# Zansa: full answer library > Zansa (from the Japanese 残差, "residual") is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. Source: https://zansa.co.za/answers ## About Zansa ### What is Zansa and what does the company actually do for businesses? Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world. It is based in South Africa and works remotely. Zansa connects the systems a business already uses, reconciles numbers that do not match, and builds dashboards and machine-learning forecasts on top of clean data. The aim is simple: a business should be able to trust its numbers and see them in one place. Typical work covers three things: - Connecting the systems you already use, such as accounting, e-commerce and spreadsheets - Reconciling figures that disagree between systems - Building dashboards and forecasts that run on the cleaned data If you are unsure whether your situation fits, a free intro call is the easiest way to find out. URL: https://zansa.co.za/answers/what-is-zansa-and-what-does-the-company-actually-do-for-businesses ### What services does Zansa offer to small and mid-sized companies? Zansa offers three core services: connecting the systems a business already uses, reconciling numbers that do not match between those systems, and building dashboards and machine-learning forecasts. These are delivered through a Data Health Check, a Foundation Build, and optional ongoing support. The work is staged so that you never commit to more than you understand. The Data Health Check is a fixed-fee audit that shows where your numbers disagree and what to fix first. The Foundation Build then connects systems and creates the data models, dashboards and, where useful, forecasts. Ongoing support is optional and covers maintenance and further development after launch. Everything built is owned by the client from day one. URL: https://zansa.co.za/answers/what-services-does-zansa-offer-to-small-and-mid-sized-companies ### Who is Zansa's data analytics service designed for? Zansa is designed for small and mid-sized companies that have data spread across several systems or spreadsheets and do not have a full in-house data team. It suits owners and managers who want numbers they can trust and a clear view of the business without building a large internal department. Common starting points include figures that differ between accounting and sales reports, monthly reporting done by hand in Excel, or a wish to forecast demand or cash. Zansa works with retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services, and will discuss other sectors on a call. Whether your size and situation fit is best checked on a free intro call. URL: https://zansa.co.za/answers/who-is-zansa-s-data-analytics-service-designed-for ### Does Zansa work with companies outside South Africa? Yes. Zansa works with small and mid-sized companies anywhere in the world. It is based in South Africa but the work is done remotely, so clients do not need to be nearby. Systems are accessed through read access granted by the client, and meetings happen over video calls. Because data lives in a cloud account owned by the client and the team collaborates online, geography is rarely a barrier. Privacy law matters more than location, so Zansa is designed around laws including POPIA and GDPR, with an operator agreement signed before any data is touched. Guidance prices are quoted in South African rand, and the final price is a fixed quote. URL: https://zansa.co.za/answers/does-zansa-work-with-companies-outside-south-africa ### Is Zansa based in South Africa and can it work remotely with overseas clients? Zansa is based in South Africa and works remotely, serving clients worldwide. Remote delivery is the normal way of working, not an exception. Access to client systems is granted online by the client, and calls, reviews and handovers all happen by video and shared documents. There is no requirement to meet in person. The client keeps control by granting read access and can revoke it at any time. Documentation is written down so that it can be followed without anyone being in the room. If you have a specific requirement, such as an on-site workshop, raise it on the intro call and it can be discussed. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/is-zansa-based-in-south-africa-and-can-it-work-remotely-with-overseas-clients ### How does Zansa handle time zone differences with clients in other countries? Zansa works remotely from South Africa, which sits in a time zone that overlaps reasonably with Europe and partly with other regions. Because the work needs only about an hour a week from the client, scheduling is usually handled by agreeing a fixed weekly slot that suits both sides. Much of the work is asynchronous: findings, questions and documentation are shared in writing so that nobody has to wait for a live call. A short weekly session with the people who know where the numbers come from covers the rest. If your team is in a distant time zone, mention it on the intro call so the schedule can be planned around it. URL: https://zansa.co.za/answers/how-does-zansa-handle-time-zone-differences-with-clients-in-other-countries ### How does remote collaboration with Zansa work day to day? Day to day, collaboration is light. Zansa asks for about an hour a week from the people who know where the numbers come from, plus decisions from a sponsor. The rest of the work happens on Zansa's side using read access to your systems, with progress and questions shared in writing. A typical rhythm looks like this: - A short weekly call to answer questions about how the business runs - Written updates on findings and what was built - Reviews of dashboards before they are finalised Because the scope, price and timeline are agreed in writing before any build starts, there are few surprises, and the client can see what is happening throughout. URL: https://zansa.co.za/answers/how-does-remote-collaboration-with-zansa-work-day-to-day ### What is Zansa's process from first contact to a finished data project? Zansa's process has four steps: a free intro call, a Data Health Check, a Foundation Build, and optional ongoing support. The intro call explains how the business runs, the Health Check audits the data, the build delivers the dashboards, and support keeps everything running afterwards. In order: - Free intro call to learn how the business runs and which numbers matter - Data Health Check, a fixed-fee audit of systems, spreadsheets and numbers that do not match, taking 1-2 weeks - Foundation Build, typically 4-8 weeks, with price and timeline agreed in writing before any build starts - Optional ongoing support You can stop after any step if it is not right for you. URL: https://zansa.co.za/answers/what-is-zansa-s-process-from-first-contact-to-a-finished-data-project ### What happens on Zansa's free intro call and is there any obligation? The free intro call is a conversation to learn how your business runs and which numbers matter most to you. It is not a sales commitment. Afterwards you can decide whether a Data Health Check makes sense, and Zansa can say honestly if your situation is not a good fit. It helps to bring a rough list of the systems you use, the reports you rely on, and the figures that cause arguments. You do not need technical language. The call also lets you ask about privacy, tooling and pricing. If the work looks worthwhile, the next step is a fixed-fee Data Health Check. If not, you leave with a clearer view of your options. URL: https://zansa.co.za/answers/what-happens-on-zansa-s-free-intro-call-and-is-there-any-obligation ### What is a Data Health Check and why would a business need one? A Data Health Check is a fixed-fee audit of your systems, spreadsheets and the numbers that do not match, completed in one to two weeks. It shows where your data is unreliable and what to fix first. It is the step that makes a later build accurately scoped and priced. Many businesses find that sales, stock and accounting figures disagree but nobody knows why. The Health Check traces those differences to their source, whether that is duplicated records, manual spreadsheet edits or mismatched definitions. It is useful even if you never proceed to a build, because the findings stand on their own. Because the fee is fixed, you know the cost before work begins. URL: https://zansa.co.za/answers/what-is-a-data-health-check-and-why-would-a-business-need-one ### What do you actually receive at the end of a Zansa Data Health Check? At the end of a Data Health Check you receive a clear picture of your systems, spreadsheets and the numbers that do not match, along with what is causing the problems and what is worth fixing first. It also gives the basis for a fixed quote if you want a Foundation Build. The audit covers the systems you use, the spreadsheets people rely on, and the places where figures disagree. For businesses with paper-based records, it also says what is worth digitising. The exact format of the report is agreed on the intro call, so ask there if you need something specific, such as a summary for a board or an owner. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/what-do-you-actually-receive-at-the-end-of-a-zansa-data-health-check ### How long does a Zansa Data Health Check take and what does it cost? A Data Health Check takes one to two weeks and is a fixed-fee engagement. As a guide, prices start from R25,000 in South African rand. The final price is a fixed quote given once Zansa understands which systems and spreadsheets are involved. The price depends mainly on how many systems and spreadsheets there are to review, and how tangled the numbers are. Because the quote is fixed, the cost does not grow if the audit turns up more than expected. The R25,000 figure is guidance, not a final quote. The intro call is the place to get a realistic estimate for your situation. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/how-long-does-a-zansa-data-health-check-take-and-what-does-it-cost ### What is a Foundation Build at Zansa? in practice A Foundation Build is the main project in which Zansa connects your systems, creates clean data models, and builds the dashboards and, where useful, forecasts. It typically takes four to eight weeks, with the price and timeline agreed in writing before any build starts. The build creates the reliable base your reporting stands on. Data from sources such as Xero, Sage, Shopify, SQL databases and spreadsheets is loaded daily into a cloud account that you own, usually Google BigQuery. Dashboards are built in Metabase by default or Power BI. Because the client owns the code, data models, dashboards and documentation from day one, nothing is locked inside Zansa. URL: https://zansa.co.za/answers/what-is-a-foundation-build-at-zansa-in-practice ### How long does a Zansa Foundation Build usually take? A Foundation Build typically takes four to eight weeks. The exact timeline depends on how many systems are connected, the state of the data, and how quickly questions can be answered. The timeline is agreed in writing, along with the fixed price, before any build starts. Faster projects tend to have fewer systems and a sponsor who can make decisions quickly. Longer ones usually involve messy source data or many data sources. The Data Health Check comes first and is what makes the timeline realistic, since the main uncertainties are discovered there. The weekly hour from people who know where the numbers come from also keeps things moving. URL: https://zansa.co.za/answers/how-long-does-a-zansa-foundation-build-usually-take ### Does Zansa work on fixed prices or hourly rates? Zansa works on fixed prices. A Data Health Check is a fixed fee, and a Foundation Build has a fixed price and timeline agreed in writing before any build starts. The final price is a fixed quote given once Zansa understands your systems. Fixed pricing means you know the cost up front and are not billed for time spent. The Health Check exists partly to make that possible, since it reveals the complexity before the build is quoted. Ongoing support is a monthly arrangement with guidance pricing from R12,000 per month. If anything about scope changes, that is discussed and agreed, rather than assumed. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/does-zansa-work-on-fixed-prices-or-hourly-rates ### What are Zansa's guidance prices for health checks, builds and support? As a guide in South African rand, Data Health Checks start from R25,000, Foundation Builds typically cost R80,000 to R250,000, and ongoing support starts from R12,000 per month. These are guidance figures only. The final price is a fixed quote once Zansa understands your systems. Where you land in the build range depends on the number of systems, the quality of the data, the number of dashboards and whether forecasting is included. Prices are in ZAR. Clients abroad can ask on the intro call how invoicing in another currency would work. A fixed quote, agreed in writing, is always given before any build starts, so the guidance range is only a starting point. URL: https://zansa.co.za/answers/what-are-zansa-s-guidance-prices-for-health-checks-builds-and-support ### How much does a Foundation Build cost and what drives the price? A Foundation Build typically costs between R80,000 and R250,000, as a guide. The final amount is a fixed quote agreed in writing before any build starts. The price depends mainly on how many systems are connected, the condition of the data, and how many dashboards or forecasts are needed. Factors that tend to move the price: - Number and type of source systems, for example Xero, Sage, Shopify, SQL databases or spreadsheets - How much reconciliation and cleaning is needed - Number of dashboards and users - Whether machine-learning forecasts are included The Data Health Check identifies most of these factors, which is why the quote after it is reliable. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/how-much-does-a-foundation-build-cost-and-what-drives-the-price ### What does Zansa's ongoing support include and how much does it cost? Ongoing support is an optional monthly arrangement after the build, with guidance pricing from R12,000 per month. It keeps the data connections, models and dashboards running and lets the business develop them further. The exact scope and final price are agreed in writing. Support suits businesses that want to keep improving their reporting but do not want to hire in-house. Typical needs include adding a new data source, changing a dashboard, or updating a forecast. Because the client owns everything built, support is never a condition of keeping your work. You can stop at any time and still have your code, models, dashboards and documentation. URL: https://zansa.co.za/answers/what-does-zansa-s-ongoing-support-include-and-how-much-does-it-cost ### Is ongoing support required after a Zansa project, or is it optional? Ongoing support is optional. After a Foundation Build you can choose to continue with monthly support, or you can run things yourself, since you own the code, data models, dashboards and documentation from day one. Nothing about the build depends on a support contract. Documentation is part of what you own, so your team or another provider can pick the work up. Some clients like having a team on hand for changes, while others prefer to handle it internally. If you are unsure which suits you, discuss it before the build ends. Support pricing is guided from R12,000 per month, with scope agreed in writing. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/is-ongoing-support-required-after-a-zansa-project-or-is-it-optional ### What does Zansa need from a client to run a project successfully? Zansa needs three things from a client: a sponsor who can make decisions, read access to the relevant systems, and about an hour a week from people who know where the numbers come from. With those in place, most of the work can be done remotely by Zansa. Each requirement has a purpose. The sponsor settles disagreements about definitions and priorities. Read access lets the team see the data without changing it. The weekly hour from knowledgeable staff clears up questions that no system can answer. Projects that lack one of these usually stall, so it is worth checking before you start. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/what-does-zansa-need-from-a-client-to-run-a-project-successfully ### Why does Zansa require a decision-making sponsor on every project? Zansa asks for a sponsor who can make decisions because data projects constantly hit choices, such as which definition of revenue is correct or which system is the source of truth. Without someone who can settle these quickly, projects stall or end up with numbers nobody agrees on. The sponsor does not need to be technical. They need the authority to decide, the time to respond, and the interest to see the work through. Typical decisions include what counts as a sale, which dashboards matter most, and who may see which figures. If no one has that authority, it is better to resolve that before a build starts, and the intro call is a good place to raise it. URL: https://zansa.co.za/answers/why-does-zansa-require-a-decision-making-sponsor-on-every-project ### What kind of system access does Zansa need and can it be revoked? Zansa needs read access to the systems relevant to the project. The client grants this access and can revoke it at any time. Read access means data can be viewed and copied into the client's own cloud account but not changed in the source systems. Access is limited to what the project needs, and activity is logged. The data is loaded into a cloud account owned by the client, usually Google BigQuery, and nothing is stored in a Zansa account. Because you hold the keys, you control the arrangement: removing access ends Zansa's ability to see the data. An operator agreement is signed before any data is touched. URL: https://zansa.co.za/answers/what-kind-of-system-access-does-zansa-need-and-can-it-be-revoked ### Why does Zansa ask for an hour a week from people who know the numbers? Zansa asks for about an hour a week from the people who know where the numbers come from because systems alone cannot explain why figures differ. Someone who works with the data daily can often resolve in minutes what would take days to guess. These people might be a bookkeeper, an operations manager or a sales administrator. They know which spreadsheet is the real one, which field staff misuse, and what a code means. The time is modest and regular. Skipping it tends to lengthen the project, since unanswered questions hold up the data models. If this is difficult for your team, raise it on the intro call. URL: https://zansa.co.za/answers/why-does-zansa-ask-for-an-hour-a-week-from-people-who-know-the-numbers ### How does Zansa scope a project and agree what will be built? Zansa scopes a project through the free intro call and then a Data Health Check, which audits systems, spreadsheets and mismatched numbers. Once the systems are understood, Zansa gives a fixed quote, and the price and timeline are agreed in writing before any build starts. Scoping is based on evidence rather than guesses. The intro call establishes which numbers matter. The Health Check reveals what is actually in the systems and how messy it is. The written agreement then states what will be delivered, the price and the timeline, so both sides know what done looks like. Anything outside that scope would be discussed and agreed separately. URL: https://zansa.co.za/answers/how-does-zansa-scope-a-project-and-agree-what-will-be-built ### What can Zansa do if my business data is on paper rather than in systems? If your data is on paper, Zansa starts with a simple capture process, such as structured templates or forms, so the information can be collected consistently. The Data Health Check then says what is worth digitising, so you do not spend money converting records that will never be used. Dashboards and forecasts need data in a structured, repeatable form. Paper records usually lack that, so the first job is deciding what to capture and how. This is often small, such as a standard form filled in the same way each day. Once capture is in place, the data can be loaded and analysed like any other source. The details depend on your business and are best discussed on an intro call. URL: https://zansa.co.za/answers/what-can-zansa-do-if-my-business-data-is-on-paper-rather-than-in-systems ### Who owns the dashboards, code and data models Zansa builds? The client owns everything Zansa builds, including code, data models, dashboards and documentation, from day one. The data itself sits in a cloud account owned by the client, and nothing is stored in a Zansa account, so there is no lock-in or hostage situation if the relationship ends. Ownership from day one means you do not wait for a final payment or a handover date to hold your work. Documentation is included so another analyst or provider can understand and extend it. This also applies if you choose not to continue with ongoing support. You keep the work and the data, and you can revoke Zansa's access whenever you choose. URL: https://zansa.co.za/answers/who-owns-the-dashboards-code-and-data-models-zansa-builds ### Which dashboard tool does Zansa use and why is Metabase the default? Zansa builds dashboards in Metabase by default because it has no per-user licence fees, so the cost does not rise as more people in your team view the numbers. If your team already uses Power BI, Zansa can build there instead, using the same data models underneath. The data models are the valuable part, and they are independent of the front-end tool. That is why the choice of Metabase or Power BI can be made on practical grounds such as cost and existing habits. If you already pay for Power BI licences and your people know it, staying with it may be sensible. Discuss your preference on the intro call. URL: https://zansa.co.za/answers/which-dashboard-tool-does-zansa-use-and-why-is-metabase-the-default ### Can Zansa build dashboards in Power BI instead of Metabase? Yes. If your team already uses Power BI, Zansa can build the dashboards there. The data models underneath are the same as with Metabase, so the quality and structure of the work do not depend on which dashboard tool you choose to display them. Metabase is the default because it has no per-user licence fees. Power BI makes sense when your organisation already has licences, existing reports or staff trained on it. Since the models sit in your cloud account, switching tools later is possible without rebuilding the data layer. Mention your preference on the intro call and Zansa will account for it in the fixed quote. URL: https://zansa.co.za/answers/can-zansa-build-dashboards-in-power-bi-instead-of-metabase ### Are there per-user licence fees for Zansa's dashboards? Not with the default setup. Zansa builds dashboards in Metabase, which has no per-user licence fees, so adding more viewers does not increase the cost of the tool. If you choose Power BI because your team already uses it, Microsoft's own licensing applies. This matters for growing teams, since per-user pricing can make wide access to dashboards expensive. Metabase avoids that. Hosting and cloud costs for the data warehouse, such as Google BigQuery, are separate and paid by you, since the account belongs to you. Zansa can explain what to expect on the intro call, but check current vendor pricing directly. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/are-there-per-user-licence-fees-for-zansa-s-dashboards ### Which systems and file types can Zansa connect to? Zansa works with Excel and CSV files, Xero, Sage, Shopify and SQL databases, with data loaded daily into a cloud account owned by the client. Other systems may also be possible, depending on what they offer, and are best discussed on an intro call. Most small and mid-sized companies use a mix: accounting in Xero or Sage, sales in Shopify, and a number of spreadsheets filled in by hand. Bringing these into one place is the core of a Foundation Build. If you use a system not on this list, describe it on the call and Zansa will say honestly whether it can be connected and how. URL: https://zansa.co.za/answers/which-systems-and-file-types-can-zansa-connect-to ### How often is data refreshed in the dashboards Zansa builds? Data from connected systems is loaded daily. That means dashboards reflect the previous day's activity rather than updating second by second. For most small and mid-sized businesses, daily is enough for decisions about sales, stock, cash and operations. If you want to check how this applies to your business, raise it on the free intro call. Daily loading keeps the setup simple and dependable. If you believe you need fresher figures for a particular use, raise it on the intro call so it can be considered when scoping. The load covers sources such as Excel and CSV files, Xero, Sage, Shopify and SQL databases. Spreadsheets only update when someone saves a new version, so their freshness depends on your team. URL: https://zansa.co.za/answers/how-often-is-data-refreshed-in-the-dashboards-zansa-builds ### How does Zansa reconcile numbers that do not match between systems? Zansa traces each figure back to its source in each system, compares definitions, and finds where the difference arises, such as timing, duplicates, manual edits or different rules. The agreed definitions are then built into the data models so the reconciled number is produced the same way every day. Common causes of mismatch include sales recorded on order date in one system and payment date in another, refunds handled differently, or spreadsheets edited by hand. This starts in the Data Health Check, which audits the systems, spreadsheets and numbers that do not match. The sponsor then settles which definition is correct, and the Foundation Build encodes that decision so the problem does not return. URL: https://zansa.co.za/answers/how-does-zansa-reconcile-numbers-that-do-not-match-between-systems ### What machine-learning forecasts can Zansa build for a small business? Zansa builds machine-learning forecasts on top of the cleaned data from a Foundation Build, for questions such as expected demand or sales. Which forecasts make sense depends on your data and business, which is assessed in the Data Health Check. Models run inside the client's own cloud account. A forecast is only as good as the data behind it, which is why reconciliation and data models come first. If your history is short or inconsistent, the Health Check will say so honestly. Because models run in your own account, your data does not leave your environment. Zansa does not use AI on client data without written approval. Details of what is feasible are best discussed on an intro call. URL: https://zansa.co.za/answers/what-machine-learning-forecasts-can-zansa-build-for-a-small-business ### Which industries does Zansa work with? in practice Zansa works with retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. Other sectors are possible and can be discussed on a call. The approach, connecting systems, reconciling numbers and building dashboards, is similar across sectors. If you want to check how this applies to your business, raise it on the free intro call. Sector matters mostly in the details: which systems are common, which numbers drive decisions, and what regulations apply to the data. Healthcare, for example, involves sensitive personal information, which Zansa treats by minimising or de-identifying personal data for reporting. If your sector is not listed, book an intro call and describe how your business runs. Zansa will tell you honestly whether it can help. URL: https://zansa.co.za/answers/which-industries-does-zansa-work-with-in-practice ### Where is my data stored when Zansa builds my reporting? Your data is stored in a cloud account owned by you, usually Google BigQuery. Nothing is stored in a Zansa account. You grant Zansa access to that environment and can revoke it at any time, so you stay in control of where the data lives. Because the account is yours, you choose or confirm its settings, billing and region where the platform allows it. If data location matters because of legal or contractual rules, raise it on the intro call so it can be considered when the environment is set up. The machine-learning models also run inside your account, so the data does not need to be sent elsewhere. URL: https://zansa.co.za/answers/where-is-my-data-stored-when-zansa-builds-my-reporting ### Can I remove Zansa's access to my data whenever I want? Yes. Access to your systems and your cloud account is granted by you and is revocable at any time. Because nothing is stored in a Zansa account and everything built is yours, removing access ends Zansa's ability to see your data without losing your work. This is a deliberate design choice. The client holds the account and the keys, so the relationship depends on trust and results rather than lock-in. If you end the engagement, you keep the code, data models, dashboards and documentation. It is good practice to review who has access periodically, including Zansa's, and to remove accounts you no longer need. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/can-i-remove-zansa-s-access-to-my-data-whenever-i-want ### What happens to my data and work if I stop working with Zansa? You keep everything. The client owns the code, data models, dashboards and documentation from day one, and the data sits in your own cloud account rather than in a Zansa account. You can revoke Zansa's access at any time, and your reporting continues to exist. Because documentation is part of what you own, your own staff or another provider can maintain the system. Costs such as cloud hosting continue to be billed to your account, as it is your account. Support is optional and not a condition of ownership. If you want an orderly exit, ask for a handover session before ending. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/what-happens-to-my-data-and-work-if-i-stop-working-with-zansa ### Is my company too small to benefit from working with Zansa? Zansa is built for small and mid-sized companies, so size alone is not a barrier. What matters more is whether you have data in several systems or spreadsheets, numbers that you cannot fully trust, and a sponsor who can make decisions. A free intro call can tell you quickly. Costs are the practical constraint. Guidance prices start at R25,000 for a Data Health Check and typically R80,000 to R250,000 for a Foundation Build, so the work should be weighed against the value of better decisions. If Zansa is not a good fit, it will say so on the call rather than sell you something you do not need. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/is-my-company-too-small-to-benefit-from-working-with-zansa ### How does Zansa compare with hiring a full-time data analyst? A full-time analyst gives you a daily presence and deep knowledge of your business, while Zansa gives you a team with a defined project, fixed price and timeline. Zansa can suit companies that need a foundation built before they can justify or manage a permanent hire. Considerations include: - A hire is an ongoing salary commitment, while a Zansa project has a fixed price - One person has one set of skills, while a team can cover data engineering, modelling and dashboards - Zansa leaves you owning documented code and models, which a future hire can use Some companies use Zansa to build the foundation and then hire. Which route suits you depends on your volume of work and budget. URL: https://zansa.co.za/answers/how-does-zansa-compare-with-hiring-a-full-time-data-analyst ### How does Zansa differ from hiring a freelance data analyst? A freelancer can be flexible and cost-effective for a defined task. Zansa is a team that offers a structured process, a fixed-fee audit, a fixed-price build agreed in writing, and documentation, with the client owning all of it. The right choice depends on the size and risk of your project. Questions worth asking any provider, including Zansa, are who owns the work, where the data is stored, whether an operator agreement is signed, and what happens if the person becomes unavailable. With Zansa, the client owns code, models, dashboards and documentation from day one, and data sits in your own cloud account. A freelancer may also offer these terms, so check them in writing. URL: https://zansa.co.za/answers/how-does-zansa-differ-from-hiring-a-freelance-data-analyst ### How is Zansa different from hiring a large consultancy for data work? Zansa is a focused team for small and mid-sized companies, with fixed prices that start from R25,000 for a Health Check and typically R80,000 to R250,000 for a build. A large consultancy may suit complex enterprise programmes, but is often a different scale of engagement and cost. Zansa's offer is deliberately narrow: connect systems, reconcile numbers, build dashboards and forecasts, then hand everything over. Prices are guided in advance and the final quote is fixed. If your needs are enterprise-wide transformation, regulated reporting at scale or a large internal team to support, a larger firm may be a better fit. Zansa will say so honestly on the intro call. URL: https://zansa.co.za/answers/how-is-zansa-different-from-hiring-a-large-consultancy-for-data-work ### What kinds of projects are not a good fit for Zansa? Zansa is not a good fit if you have no one who can make decisions, cannot give read access to your systems, or cannot spare about an hour a week from people who know the numbers. It is also not suited to work that requires AI processing of client data without written approval. Other mismatches include needs far outside small and mid-sized company analytics, or expecting results without any client involvement. If your data does not exist in any form yet, the first step is a capture process rather than dashboards. Zansa does not claim any particular certifications, so if a contract demands one, raise it early. When in doubt, ask on the free intro call. URL: https://zansa.co.za/answers/what-kinds-of-projects-are-not-a-good-fit-for-zansa ### How do I get started with Zansa? in practice To get started, email hello@zansa.co.za or book a call through the Zansa website. The first step is a free intro call, where Zansa learns how your business runs and which numbers matter. From there you can decide whether to proceed to a Data Health Check. Before the call it helps to note: - The systems and spreadsheets you use - The reports you rely on and the numbers that do not match - Who would be the sponsor and who knows where the numbers come from There is no obligation after the call. If you go ahead, an operator agreement is signed before any data is touched, and read access is granted by you. URL: https://zansa.co.za/answers/how-do-i-get-started-with-zansa-in-practice ### What is the email address for contacting Zansa? You can contact Zansa at hello@zansa.co.za, or book a call through the Zansa website. Email is a good way to ask a first question, describe your situation briefly, or arrange the free intro call. Zansa works remotely, so contact is online. When writing, it helps to mention your sector, the systems you use, and what is frustrating you about your current numbers. You do not need technical language. If you have privacy questions, such as which law applies to your data or what the operator agreement covers, include them. For legal certainty, you should also consult your own legal adviser. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/what-is-the-email-address-for-contacting-zansa ### Does Zansa work with Xero, Sage or Shopify data specifically? Yes. Xero, Sage and Shopify are among the systems Zansa connects, along with Excel and CSV files and SQL databases. Data is loaded daily into a cloud account owned by the client, where it is reconciled and modelled for dashboards. Combining these sources is a common reason for starting. Accounting figures from Xero or Sage often differ from sales figures in Shopify because of timing, refunds, fees or manual journals. Reconciling them is part of the work. Read access is granted by you and can be revoked. Details of how each system connects, and any limits, are covered during the Data Health Check. URL: https://zansa.co.za/answers/does-zansa-work-with-xero-sage-or-shopify-data-specifically ### In what languages and currencies can Zansa work with clients? Zansa communicates in English, using UK and South African spelling. Guidance prices are in South African rand. Handling several currencies inside your data, such as sales in dollars and costs in rand, is a modelling question that depends on your systems and is best discussed on an intro call. Multi-currency reporting needs clear rules about which exchange rate applies and when. These are decisions for the sponsor, and are encoded in the data models so every report treats them identically. If your data or users work in another language, mention it on the call. Zansa can then say honestly what is feasible rather than assume. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/in-what-languages-and-currencies-can-zansa-work-with-clients ### Will the fixed price change after the Foundation Build has started? The price and timeline for a Foundation Build are agreed in writing before any build starts, so they are not meant to shift during the work. If you ask for something outside the agreed scope, that would be discussed and agreed separately rather than added silently. The Data Health Check is what makes a fixed price realistic, since it uncovers the complexity in your systems before a quote is given. The written agreement sets out what will be delivered. A delay on the client side, such as missing access or unanswered questions, can affect the timeline, which is why the sponsor, access and weekly hour are asked for at the start. URL: https://zansa.co.za/answers/will-the-fixed-price-change-after-the-foundation-build-has-started ### Do I have to do a Data Health Check before a Foundation Build? Zansa's process is intro call, then Data Health Check, then Foundation Build, because the audit makes the build's fixed price and timeline reliable. Whether a particular situation could skip it is something to raise on the intro call rather than assume. The Health Check typically takes one to two weeks and costs from R25,000 as a guide. It identifies the systems, spreadsheets and mismatched numbers that determine how much work a build involves. Skipping it would mean quoting on assumptions, which is how projects overrun. The findings also belong to you, so the audit has value even if you build with someone else. URL: https://zansa.co.za/answers/do-i-have-to-do-a-data-health-check-before-a-foundation-build ### What does Zansa mean by connecting the systems a business already uses? It means pulling data from the tools you already use, such as accounting software, e-commerce platforms, databases and spreadsheets, into one cloud account you own, loaded daily. Once the data sits together and is reconciled, it can be reported on consistently without retyping figures between systems. Zansa does not ask you to replace your systems. Xero, Sage, Shopify, SQL databases, Excel and CSV files stay where they are, and read access lets the data be copied into your environment. The result is a single, documented set of data models that dashboards and forecasts draw from, owned by you. If you want to check how this applies to your business, raise it on the free intro call. URL: https://zansa.co.za/answers/what-does-zansa-mean-by-connecting-the-systems-a-business-already-uses ## Cross-sector KPIs ### How many KPIs should be on one dashboard? Keep a dashboard to roughly five to ten measures, each tied to a decision and an owner. There is no fixed correct number, but a dashboard that needs scrolling or explaining usually has too many, and people stop reading it. Guidelines: - Show the most important figure first, with a comparison such as last period or target. - Use drill-down pages for detail, rather than putting everything on one screen. - Remove measures nobody has acted on in the last quarter. - Define each measure in writing, including formula and data source. Commonly misread: more charts do not mean more insight. Zansa builds sector dashboards, usually in Metabase or Power BI, around a small set of agreed measures, with detail available behind them. URL: https://zansa.co.za/answers/how-many-kpis-should-be-on-one-dashboard ## Forecasting and data science ### What is demand forecasting and how does it work for a small business? Demand forecasting estimates how much of a product or service customers will want over a future period, using past sales, seasonality, prices, promotions and other drivers. For a small business it usually starts with a clean history of weekly or monthly sales, a simple method such as a seasonal average, and a regular check of how wrong it was. The output is a number per product, location and period, ideally with a range around it. That number feeds purchasing, production, staffing and cash planning. A practical sequence is: - Collect at least a couple of years of sales by product and date where possible. - Clean out stock-outs, returns and one-off events. - Build a simple baseline first. - Measure error on data the method has not seen. - Only add complexity if it beats the baseline. URL: https://zansa.co.za/answers/what-is-demand-forecasting-and-how-does-it-work-for-a-small-business ### What is the difference between a sales forecast and a demand forecast? A sales forecast predicts what you expect to sell, while a demand forecast predicts what customers would have bought if nothing limited them. The two diverge when you run out of stock, cap capacity or restrict supply. Sales history understates true demand whenever a product was unavailable during the period being measured. This matters because most businesses train forecasts on sales records. If an item was out of stock for three weeks, the zero sales in those weeks do not mean nobody wanted it, and a model that learns from them will under-order next time. Common fixes include flagging stock-out periods, filling them with estimates from comparable periods, or excluding them from training. Sales forecasts are often used for revenue targets, while demand forecasts are better for inventory and capacity decisions. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-sales-forecast-and-a-demand-forecast ### How do I build a basic cash-flow forecast for my company? Start with your current bank balance, then list expected cash coming in and going out by week, based on invoice due dates, payment habits, payroll, rent, tax and supplier terms. Roll it forward 13 weeks, update it weekly, and compare actual cash to the forecast so you learn where your assumptions are wrong. Cash flow is about timing, not profit. A sale invoiced today may not be paid for 60 days, so use real customer payment patterns instead of due dates alone. Useful practices: - Separate certain items (payroll, rent) from uncertain ones (new sales). - Build a base case, a pessimistic case and an optimistic case. - Track your minimum expected balance, not just the ending balance. A spreadsheet is often enough at first. Models help once you have many customers and enough payment history. URL: https://zansa.co.za/answers/how-do-i-build-a-basic-cash-flow-forecast-for-my-company ### What is a reorder point and how do I calculate it? A reorder point is the stock level at which you place a new order so that stock lasts until the delivery arrives. A common formula is average daily demand multiplied by supplier lead time in days, plus safety stock. When on-hand plus on-order stock falls to that number, you reorder. Example with no real data: if you sell about 10 units a day, lead time is 7 days and you hold 20 units of safety stock, the reorder point is 10 x 7 + 20 = 90 units. The formula assumes demand and lead time are reasonably stable. If either varies a lot, safety stock needs to be larger. Count stock already ordered but not yet received, or you will double order. Review the figures regularly, since demand seasonality changes the right reorder point through the year. URL: https://zansa.co.za/answers/what-is-a-reorder-point-and-how-do-i-calculate-it ### What is safety stock and how much of it should I hold? Safety stock is extra inventory held to cover uncertainty in demand and supplier lead time. How much you need depends on how variable demand and lead times are and how often you are willing to run out. More variability or a stricter service target means more safety stock, which ties up more cash. A common approach multiplies the standard deviation of demand during lead time by a service factor from the normal distribution. That is a reasonable starting point, but it assumes demand is roughly bell-shaped, which is often false for slow-moving or lumpy items. Practical advice: - Set different service levels for fast sellers, slow sellers and critical items. - Reduce variability where possible, such as shorter lead times or better forecasts. - Review holding cost against the cost of a lost sale. URL: https://zansa.co.za/answers/what-is-safety-stock-and-how-much-of-it-should-i-hold ### What is customer churn and how is it measured? Churn is the rate at which customers stop buying or cancel over a given period. It is usually calculated as customers lost during the period divided by customers at the start. For subscriptions the definition is clear, while for retail or one-off purchases you must define churn as, for example, no purchase in 180 days. The definition drives everything else, so agree it before building anything. Choose the time window and the event that counts as leaving. Watch out for: - Mixing customers who joined at different times, which distorts averages. - Treating a seasonal buyer who simply returns each year as churned. - Revenue churn versus customer churn, which can tell different stories if big customers behave differently from small ones. Cohort tables, grouping customers by sign-up month, are a simple and honest way to view it. URL: https://zansa.co.za/answers/what-is-customer-churn-and-how-is-it-measured ### How can a small company improve customer retention using data? Find which customers are leaving, when, and what they had in common before they left, then act early on those signals. Useful data includes purchase frequency, time since last order, support complaints, discounts used and product mix. Test any retention offer on a small group before rolling it out widely. Start with simple analysis: retention by cohort, by first product bought, by acquisition channel and by region. Often one or two patterns explain most of the loss. A churn model can then rank current customers by risk, but only helps if someone acts on the list. Decide in advance what the action is, such as a call, an offer or a service fix, and measure whether treated customers stay longer than a comparable untreated group. Otherwise you cannot tell whether the action worked. URL: https://zansa.co.za/answers/how-can-a-small-company-improve-customer-retention-using-data ### What does pricing analysis involve for a small or mid-sized company? Pricing analysis looks at how sales volumes and margins respond to price, costs, competitor prices, discounts and customer type. The aim is to find where price changes raise profit rather than just revenue. It combines historical data, margin calculations and, ideally, controlled tests to estimate how sensitive customers are. Start by understanding margin per product after discounts, returns and delivery costs, since many firms discover that some customers or items lose money. Next, estimate price sensitivity (elasticity) from past price changes, bearing in mind that promotions, seasonality and stock changes confuse the picture. Then test: change a price in some stores, regions or customer groups and compare against similar ones that stayed the same. Be careful with brand effects and with prices that customers compare easily. URL: https://zansa.co.za/answers/what-does-pricing-analysis-involve-for-a-small-or-mid-sized-company ### What is anomaly detection and when is it useful in business? Anomaly detection flags data points that differ markedly from normal patterns, such as a sudden drop in orders, an unusual refund, a spike in costs or a sensor reading out of range. It is useful for catching errors, fraud, system failures and emerging problems earlier than manual review would. Methods range from simple rules (alert if daily sales are more than three standard deviations from the same weekday average) to machine-learning models that learn normal behaviour from many variables. The hard part is not the algorithm but tuning. Too sensitive and people ignore constant false alarms; too lax and you miss real events. Always account for known patterns such as weekends, holidays and promotions first, otherwise they are flagged as anomalies. Each alert should have an owner and a defined response. URL: https://zansa.co.za/answers/what-is-anomaly-detection-and-when-is-it-useful-in-business ### What is customer segmentation and why does it matter? Customer segmentation groups customers who behave or look alike so you can treat each group differently in marketing, pricing, service or product decisions. Segments can be defined by rules such as spend and recency, or discovered by clustering algorithms. It matters because average customers hide very different needs and profitability. Good segments are actionable. If you cannot name a different action for each group, the segmentation is just a report. Common approaches: - RFM: recency, frequency and monetary value. - Behavioural clusters, for example by product mix or channel. - Value tiers, based on margin rather than revenue. Keep the number of segments small enough that people remember them, and re-check them periodically since customers move between groups. URL: https://zansa.co.za/answers/what-is-customer-segmentation-and-why-does-it-matter ### How can data help forecast staffing and rostering needs? Forecast the workload first, such as customer arrivals, orders, bookings or calls by hour and day, then convert that workload into the staff needed using productivity rates. Compare the resulting requirement with actual rosters to find over- and under-staffing. This keeps labour cost and service levels in balance. Workload forecasts usually need a finer time grain than sales forecasts, often hourly or by shift, and they depend strongly on day of week, holidays, weather, events and promotions. The conversion step matters as much as the forecast: how many orders can one person process per hour, what minimum cover is needed, and what labour rules apply. Build in a buffer for sickness and breaks, and have managers review the output. A forecast informs a roster; it should not override legal limits or fairness rules. URL: https://zansa.co.za/answers/how-can-data-help-forecast-staffing-and-rostering-needs ### What is machine learning explained in plain language for business owners? Machine learning is software that learns patterns from past examples instead of being given explicit rules. You show it historical data, such as past customers and whether they left, and it finds relationships it can then use to predict outcomes for new cases. Its predictions are only as good as the data it learned from. Think of it as automated pattern-fitting. A traditional program says 'if X then Y'. A machine-learning model works out the relationship between inputs and outcomes from many examples. Important caveats: - It finds correlation, not necessarily cause. - It struggles when the future differs from the past. - It can learn quirks of noisy data that do not repeat. That is why testing on data the model has not seen is essential. URL: https://zansa.co.za/answers/what-is-machine-learning-explained-in-plain-language-for-business-owners ### What is the difference between supervised and unsupervised learning? Supervised learning trains on examples where the right answer is known, such as past sales or whether a customer churned, and learns to predict that answer for new data. Unsupervised learning has no answer column and instead looks for structure, such as grouping similar customers or finding unusual records. Most business forecasting is supervised: you have historical outcomes and want to predict future ones. Regression predicts a number (next month's sales) and classification predicts a category (will this customer leave, yes or no). Unsupervised methods such as clustering support segmentation and some anomaly detection. Because there is no correct answer to score against, judging them is more subjective, so results should be checked for usefulness with people who know the business. URL: https://zansa.co.za/answers/what-is-the-difference-between-supervised-and-unsupervised-learning ### What does MAPE mean and how is it calculated? MAPE is mean absolute percentage error. For each period, take the absolute difference between actual and forecast, divide by the actual value, then average these percentages across periods. It expresses typical error as a percentage, which makes it easy to explain, but it breaks down when actual values are zero or very small. Example: if actual sales are 100 and the forecast is 90, the error is 10 percent. Average these across all periods and items to get MAPE. Limitations: - Undefined when actuals are zero. - Explodes for low-volume items. - Penalises over-forecasts and under-forecasts differently, since percentage errors on under-forecasts are capped at 100 percent while over-forecasts are unbounded. For many businesses WAPE or MAE is a safer choice. URL: https://zansa.co.za/answers/what-does-mape-mean-and-how-is-it-calculated ### Why is MAPE misleading for products with low or intermittent sales? MAPE divides each error by the actual value, so tiny actuals produce huge percentages and zero actuals make the calculation impossible. A product that sells one or two units can show an error of several hundred percent from a miss of a single unit, which distorts the average and says little about real business impact. For low-volume or intermittent items, absolute measures are more meaningful. Options include: - MAE: average error in units. - WAPE: total absolute error divided by total actual volume. - Scaled errors such as MASE, which compare your error with a naive forecast. Also consider aggregating, for example evaluating at weekly rather than daily level, or by product group, because errors often partly cancel when summed. Choose the metric that reflects the decision, such as units wasted or sales lost. URL: https://zansa.co.za/answers/why-is-mape-misleading-for-products-with-low-or-intermittent-sales ### What is WAPE and when should I use it instead of MAPE? WAPE is weighted absolute percentage error: the sum of absolute errors across all items and periods divided by the sum of actual values. It weights errors by volume, so big sellers matter more, and it avoids division by zero for individual items. Use it when you forecast many products with uneven sales. Compare with MAPE, which treats a 50 percent miss on a tiny item the same as a 50 percent miss on your best seller. WAPE instead answers: of all the volume we sold, what share did we miss by? It is simple to explain to managers. Still: - It can hide poor performance on small but important items. - Report it alongside bias. - Pick one consistent aggregation level, because WAPE changes with it. Some teams call this metric the volume-weighted MAPE. URL: https://zansa.co.za/answers/what-is-wape-and-when-should-i-use-it-instead-of-mape ### What is forecast bias and how do I detect it? Forecast bias is a consistent tendency to over-forecast or under-forecast. You detect it by summing signed errors (forecast minus actual) over many periods: if the total is persistently positive or negative rather than hovering around zero, the forecast is biased. Bias is often more costly than random error because it compounds in stock and cash. Accuracy metrics like MAPE or MAE use absolute errors, so they cannot tell you the direction. A forecast can look reasonably accurate while always being too high. Typical causes include sales targets influencing forecasts, optimism about promotions, ignoring stock-outs, and models trained on unrepresentative periods. Track bias as a percentage of total actuals, by product group and by forecaster. Once identified it is often straightforward to correct, either by fixing the process or adjusting the model. URL: https://zansa.co.za/answers/what-is-forecast-bias-and-how-do-i-detect-it ### What is the difference between MAE and RMSE for forecast error? MAE is the average size of errors in the original units, treating all errors equally. RMSE squares errors before averaging and then takes the square root, so it penalises large misses much more. Use MAE for a straightforward typical error and RMSE when big misses are disproportionately costly. If errors are 2, 2 and 20 units, MAE is 8 while RMSE is about 11.7, because the single large miss dominates. RMSE is also more sensitive to outliers and data errors, so a few bad records can make it look poor. MAE is more robust and easier to explain. Neither is percentage-based, so they are hard to compare across products with different volumes. Report them at a level where the units are meaningful, and consider showing both. URL: https://zansa.co.za/answers/what-is-the-difference-between-mae-and-rmse-for-forecast-error ### What is backtesting a forecast and why should I do it? Backtesting means pretending to forecast in the past: you train on data up to a certain date, forecast the following period, compare with what actually happened, then repeat over several earlier dates. It shows how a method would have performed without waiting months, and it prevents you trusting a model that only fits history. The key rule is that the model may only use information that was available at the time of each forecast. Using future data, even accidentally, makes results look better than reality. Good practice: - Test over several periods, including busy and quiet seasons. - Match the forecast horizon you actually need. - Compare against a simple baseline. - Look at bias as well as accuracy. Zansa treats backtesting against a baseline as a standard step before a forecast goes into a dashboard. URL: https://zansa.co.za/answers/what-is-backtesting-a-forecast-and-why-should-i-do-it ### How is time-series cross-validation different from normal cross-validation? Time-series cross-validation always trains on the past and tests on the future, moving the cutoff date forward in steps. Normal cross-validation shuffles rows randomly, which lets a model see future data while predicting the past. For forecasting that leakage gives overoptimistic results, so ordering must be respected. In a rolling-origin scheme you might train on months 1 to 24 and test on month 25, then train on 1 to 25 and test on 26, and so on. Alternatively you can use a fixed-size window that slides forward, which helps if old data is no longer representative. Average the errors across all folds, and look at how they vary, since a method that is good on average but occasionally terrible may be risky. Allow a gap between training and test if features use lagged information. URL: https://zansa.co.za/answers/how-is-time-series-cross-validation-different-from-normal-cross-validation ### What is seasonality in forecasting and how do I handle it? Seasonality is a repeating pattern tied to the calendar, such as higher sales every December, busier weekends or a quieter winter. Handle it by using a method that models the repeating cycle, such as seasonal naive, Holt-Winters, or a model with day, week and month features, and make sure you have at least two full cycles of history. There can be several layers at once: day of week, month, and year. Hourly foot traffic shows daily and weekly cycles, while ice cream sales show a yearly one. Tips: - Plot the data by period and look before modelling. - Check whether seasonal swings grow with the level of sales, which suggests a multiplicative pattern. - Do not confuse seasonality with one-off events. - Moving holidays such as Easter and Ramadan need explicit handling. URL: https://zansa.co.za/answers/what-is-seasonality-in-forecasting-and-how-do-i-handle-it ### What is trend in a time series and how is it different from seasonality? Trend is the long-term direction of a series, such as steady growth or decline, while seasonality is a regular repeating cycle within the year or week. A business can have rising sales (trend) that still peak each December (seasonality). Good forecasts separate the two so one is not mistaken for the other. Methods like decomposition split a series into trend, seasonal and remainder parts so you can see each one. The remainder is what is left over, similar to the residual in a model. Be cautious about extrapolating trends far ahead. Growth rarely continues in a straight line, and a trend fitted to a short period may be a temporary swing. Consider damped trends that flatten over time, and compare with an assumption of no trend. URL: https://zansa.co.za/answers/what-is-trend-in-a-time-series-and-how-is-it-different-from-seasonality ### How should forecasts handle public holidays and moving events? Add holidays as explicit inputs, with a flag for the day and often for the days before and after. Fixed-date holidays can partly be captured by seasonality, but moving ones, such as Easter, Eid, Chinese New Year or school holidays, shift each year and must be listed by date or the model will blur their effect. Effects differ by business. A public holiday may cut B2B orders but increase retail footfall, and a long weekend can create a bridge effect. Useful practice: - Keep a calendar table with holiday dates for each country or region you serve. - Include lead-up and aftermath windows. - Treat each major holiday as its own event if you have enough history. - Check residuals around holiday dates to see if the effect is captured. Global firms need local calendars, not just one national list. URL: https://zansa.co.za/answers/how-should-forecasts-handle-public-holidays-and-moving-events ### How do promotions affect forecasts and how should I account for them? Promotions create temporary spikes that can mislead a model if they are not recorded. Capture each promotion's dates, discount, channel and products as inputs, so the model separates promoted demand from baseline demand. Also account for the dip after a promotion, when customers who bought early no longer need to buy. If promotions are not flagged, the model may learn a false seasonal peak or inflate the baseline. Forecasting future periods also needs the planned promotion calendar, otherwise you cannot forecast the uplift. Consider: - Cannibalisation of other products. - Pull-forward and post-promotion dips. - Uplift varying by depth of discount and promotion type. With few past promotions, a simple uplift factor per promotion type, reviewed by sales staff, often beats a complex model. URL: https://zansa.co.za/answers/how-do-promotions-affect-forecasts-and-how-should-i-account-for-them ### How much historical data do I need to build a useful forecast? It depends on the method and seasonality, but a common rule of thumb is at least two full seasonal cycles, so about two years for yearly seasonality, to estimate seasonal patterns. Simple methods can work with less. Complex machine-learning models generally need more history, more observations or many similar products to learn from. With only a year of data you cannot tell whether a December peak is seasonal or a one-off. With a few months you can still forecast short-term using averages and recent trends, but you should expect wide uncertainty. More important than length is relevance and quality. Five years of data from a different business model may be less useful than eighteen months of clean recent data. Pooling similar products can also help new or slow items borrow strength from the group. URL: https://zansa.co.za/answers/how-much-historical-data-do-i-need-to-build-a-useful-forecast ### When is machine learning not worth it for forecasting? Machine learning is often not worth it when you have little history, stable demand, few items, or when a simple method is nearly as accurate. The extra cost of building, explaining and maintaining a model must be repaid by better decisions. If a seasonal average gets you close enough, use that. Signs that simpler methods may win: - Fewer than two years of data. - Noisy, tiny volumes where nothing predicts well. - Decisions that do not change with small accuracy gains. - No one available to monitor the model. - A need for full transparency. Machine learning tends to pay off with many products or locations, rich drivers like prices, promotions and weather, and decisions where small percentage gains are worth a lot of money. Test against a baseline before deciding. URL: https://zansa.co.za/answers/when-is-machine-learning-not-worth-it-for-forecasting ### What is a forecasting baseline and why do I need one? A baseline is a deliberately simple forecast, such as repeating last period's value or last year's same-week value, that any fancier method must beat. It gives context to accuracy numbers: a 20 percent error means little until you know a naive method achieves 22 or 40 percent. Without a baseline you cannot tell whether a complex model adds value. Many sophisticated models turn out barely better than a seasonal naive forecast. Good baselines include: - Naive: next period equals the last one. - Seasonal naive: next period equals the same period last cycle. - Moving average of recent periods. Measure the baseline using the same backtest and the same metric as the model. If the model does not clearly win, keep the baseline. Zansa compares every machine-learning forecast against such a baseline before recommending it. URL: https://zansa.co.za/answers/what-is-a-forecasting-baseline-and-why-do-i-need-one ### What is a naive forecast and a seasonal naive forecast? A naive forecast simply uses the most recent observed value as the forecast for every future period. A seasonal naive forecast uses the value from the same point in the previous cycle, such as the same week last year or the same weekday last week. Both are free to produce and make strong benchmarks. Naive works for series without clear seasonality, where the best guess for tomorrow is today. Seasonal naive suits series with strong repeating patterns, such as retail sales by day of week. They ignore trend and one-off effects, so they can lag when things change. Their value is as a test: if a complex model cannot beat seasonal naive in backtesting, it is probably not worth deploying. They are also useful fall-backs when a model fails or data is missing. URL: https://zansa.co.za/answers/what-is-a-naive-forecast-and-a-seasonal-naive-forecast ### What is exponential smoothing and why is it popular for business forecasts? Exponential smoothing forecasts using a weighted average of past observations where recent data gets the most weight and older data fades away. Extensions add trend and seasonality, such as Holt-Winters. It is popular because it is fast, robust, needs little data, handles many series and is easy to explain. The smoothing parameter controls how quickly the forecast reacts: a high value follows recent changes, a low one is stable. Software usually fits this automatically. Variants: - Simple: level only. - Holt: level and trend. - Holt-Winters: level, trend and seasonality. Its limits are that it cannot easily use outside drivers such as price or promotions, so a regression or machine-learning approach may be better when those matter. It remains a strong baseline. URL: https://zansa.co.za/answers/what-is-exponential-smoothing-and-why-is-it-popular-for-business-forecasts ### How do I explain a forecast to non-technical managers? Lead with the decision it supports, give a single expected number plus a plausible range, state the main assumptions in plain words, and show how accurate the method has been on past periods. Avoid jargon. Managers need to know what to do, how confident to be and what would change the answer. Helpful habits: - Use a chart of history, forecast and range rather than tables of numbers. - Express accuracy in business terms, such as typically within 10 percent per week. - Name the biggest drivers, like a planned promotion. - Show a worst-case and best-case scenario. - Be upfront about what the model cannot know, such as a new competitor. Invite challenge. A manager who knows of an upcoming event can improve the forecast. URL: https://zansa.co.za/answers/how-do-i-explain-a-forecast-to-non-technical-managers ### What is a prediction interval and why show a range instead of one number? A prediction interval is a range expected to contain the actual outcome with a stated probability, such as 80 percent. Showing a range is more honest than a single figure because every forecast is uncertain, and decisions like stock levels or cash buffers depend on how wide that uncertainty is. A single number invites false precision. If demand is forecast at 1,000 units with a likely range of 800 to 1,250, you would order differently than if the range were 950 to 1,050. Intervals should be checked: if an 80 percent interval contains the actual result only 50 percent of the time in backtests, it is too narrow. Many simple methods understate uncertainty. Intervals normally widen the further ahead you forecast. URL: https://zansa.co.za/answers/what-is-a-prediction-interval-and-why-show-a-range-instead-of-one-number ### What does it take to deploy a forecasting model into daily operations? Deployment means the model runs automatically on fresh data, writes its forecasts somewhere people use them, and has an owner. You need reliable data feeds, a schedule, stored outputs, error handling, and a fallback if it fails. A model in a notebook that someone runs by hand is not deployed. Core pieces: - Automated data pipeline with checks for missing or odd values. - Scheduled training and forecasting. - Outputs delivered to a dashboard, spreadsheet or planning system. - Version records of models and results. - Monitoring and alerts. - Documentation so someone else can run it. Plan for the people side too: who reviews the forecast, who can override it, and what happens when they disagree with it. URL: https://zansa.co.za/answers/what-does-it-take-to-deploy-a-forecasting-model-into-daily-operations ### How do I monitor a forecasting model after it goes live? Compare forecasts against actuals as they arrive and track accuracy and bias over time, by product group. Also monitor the input data for missing feeds, strange values and changes in patterns. Set thresholds that trigger a review or retraining, and keep a simple baseline running alongside for comparison. Models degrade because the world changes: new competitors, pricing shifts, product mix changes, or a data system update that alters a field. Useful checks: - Rolling error and bias against the baseline. - Share of forecasts overridden by people. - Data freshness and completeness. - Distribution of inputs versus training time. Write down what each alert means and who acts on it. Monitoring that nobody reads is the same as no monitoring. URL: https://zansa.co.za/answers/how-do-i-monitor-a-forecasting-model-after-it-goes-live ### What is model drift and how often should I retrain a forecast? Model drift is the decline in a model's accuracy because the data or relationships it learned have changed. Retrain on a regular schedule, such as monthly or quarterly depending on how fast your business changes, and also whenever monitoring shows accuracy or bias worsening beyond a set threshold. There are two kinds. Data drift means inputs look different, for example a new product range. Concept drift means the relationship changed, for example customers react differently to discounts after a competitor enters. More frequent retraining is not always better, as it can chase noise and make forecasts unstable. Test whether retraining improves backtest results. Many time-series models are refit with every forecast cycle, which is cheap, while heavier machine-learning models may be retrained less often. URL: https://zansa.co.za/answers/what-is-model-drift-and-how-often-should-i-retrain-a-forecast ### How can I safely use ChatGPT or other AI assistants with business data? Do not paste personal, confidential or client-identifiable data into a public AI tool unless you have checked its data terms and have approval. Use anonymised or aggregated data, business-grade accounts that do not train on your inputs, and always verify numbers and code the assistant produces before relying on them. Safe practice includes: - Remove names, emails, phone numbers, ID numbers and account details. - Share summaries or small samples rather than full databases. - Check settings on whether your content is used for training or retained. - Write a short company policy on what may be shared. - Keep a human responsible for checking outputs. AI assistants are good for explaining methods, drafting queries and brainstorming, but they can state wrong things confidently, so validate anything that affects decisions. URL: https://zansa.co.za/answers/how-can-i-safely-use-chatgpt-or-other-ai-assistants-with-business-data ### Can AI assistants like ChatGPT be trusted to calculate numbers from my spreadsheets? Not blindly. Language models can misread data, make arithmetic mistakes and sound confident while wrong. They are more reliable when they write code or formulas that actually run on the data, and you check the results against a known total. Treat their numbers as drafts until verified. Language models predict plausible text rather than performing exact calculation, though tools connected to a code environment can compute properly. Even then, the model may interpret columns incorrectly, drop rows or misapply a filter. Verification steps: - Reconcile totals with your source system. - Test the logic on a small case you can calculate by hand. - Ask it to show the formulas or code, then review them. - Keep sensitive data out unless approved. For important reporting, a repeatable pipeline is safer than a chat. URL: https://zansa.co.za/answers/can-ai-assistants-like-chatgpt-be-trusted-to-calculate-numbers-from-my-spreadshe ### What data privacy risks come with using AI tools on customer data? The main risks are exposing personal data to a third party, data being retained or used to train models, unauthorised cross-border transfers, and inability to honour deletion or access requests. You may also create decisions about people that are unfair or unexplainable. These can breach privacy laws and customer trust. Before using any AI tool with personal data, check: - Who processes the data and where. - Whether inputs are stored or used for training. - The legal basis and purpose for processing under laws such as GDPR or POPIA. - Whether a data processing agreement is in place. - Whether you can minimise or anonymise the data. Where possible, keep data inside your own controlled environment and share only what the task actually needs. Seek legal advice for specific obligations. URL: https://zansa.co.za/answers/what-data-privacy-risks-come-with-using-ai-tools-on-customer-data ### What are the most common forecasting mistakes small businesses make? Common mistakes include skipping a baseline, judging models on the data they were trained on, ignoring stock-outs in history, not recording promotions, forecasting at too fine a level, reporting only a single number, and never tracking accuracy or bias. Many also add complexity before cleaning their data. Further pitfalls: - Letting targets or hopes shape the forecast. - Using accuracy metrics that hide bias. - Extrapolating a short-lived trend. - Forgetting that the forecast should drive a specific decision. - Never revisiting the model after launch. Most fixes are about process, not algorithms: record events, measure error honestly, review regularly, and keep methods as simple as the decision allows. URL: https://zansa.co.za/answers/what-are-the-most-common-forecasting-mistakes-small-businesses-make ### What is a residual in forecasting and what does it tell me? A residual is the difference between an actual value and what the model fitted or predicted for it, so it is the part the model left unexplained. Studying residuals shows whether the model is missing a pattern, such as seasonality, an event or a trend, and whether its errors are random or systematic. If a model is capturing everything useful, residuals should look like random noise: centred on zero, without patterns over time and without links to inputs. Visible structure, such as regular spikes each December or errors growing with volume, signals something the model has missed. Residuals also help spot data errors and outliers, and they underpin prediction intervals. The Japanese word 残差 (zansa), which gives Zansa its name, means exactly this: the residual, what remains after the model. URL: https://zansa.co.za/answers/what-is-a-residual-in-forecasting-and-what-does-it-tell-me ### How do I do error analysis to improve a forecast? Break the errors down by product, location, time period, season, promotion status and customer segment to find where the forecast fails most and in which direction. Then look for causes: missing events, bad data, wrong method for that group. Fix the largest, most systematic errors first. A practical routine: - Rank items by total absolute error to focus on those that matter. - Plot actual against forecast and look at the worst periods. - Check bias by group to find systematic over- or under-forecasting. - Examine residuals for patterns by weekday, month or event. - Ask people in the business what happened in the worst weeks. Often a handful of items or data problems cause most of the error, so targeted fixes beat swapping algorithms. URL: https://zansa.co.za/answers/how-do-i-do-error-analysis-to-improve-a-forecast ### What is data leakage and how can it ruin a forecast? Data leakage is when information that would not be available at prediction time sneaks into model training or testing, making results look far better than they will be in real use. Examples are using future sales as an input, random train-test splits on time data, or features calculated using the whole dataset. Leakage is a leading reason models perform well in a demo and badly in production. A typical case: a churn model uses a field such as 'cancellation reason', which only exists after the customer has left. Protect yourself by: - Splitting by time, not randomly. - Asking of each feature: would I know this at the moment of prediction? - Computing summaries only from data before the forecast date. - Being suspicious of unusually high accuracy. URL: https://zansa.co.za/answers/what-is-data-leakage-and-how-can-it-ruin-a-forecast ### What is overfitting and how can I tell if my forecast model has it? Overfitting is when a model learns the noise and quirks of its training data rather than the real pattern, so it looks excellent on past data but performs poorly on new data. The tell-tale sign is a big gap between accuracy on training data and accuracy on held-out or backtest data. Complex models with many features and little data overfit easily. Ways to reduce it: - Evaluate only on data the model has not seen, split by time. - Prefer simpler models when results are similar. - Limit the number of inputs. - Use regularisation or tree depth limits in machine-learning methods. - Compare with a baseline. If adding complexity improves training error but not backtest error, you are fitting noise. URL: https://zansa.co.za/answers/what-is-overfitting-and-how-can-i-tell-if-my-forecast-model-has-it ### How should I split data into training and test sets for forecasting? Split by time: train on the earlier period and test on the later one, never shuffling. Keep the test period as long as your forecast horizon and ideally covering a full season. For stronger evidence, repeat with several cutoff dates rather than relying on a single split. For example, train on January 2022 to December 2024 and test on January to June 2025. Then roll the cutoff forward and repeat. Some teams use three parts: training, validation for choosing settings, and a final untouched test set. Do not tune against the final test set repeatedly, or it stops being an honest check. Keep the gap between training and test realistic: if you forecast eight weeks ahead in practice, test eight weeks ahead. URL: https://zansa.co.za/answers/how-should-i-split-data-into-training-and-test-sets-for-forecasting ### How do I forecast intermittent or lumpy demand for slow-moving items? Use methods designed for intermittent demand, such as Croston's method or its variants, which model how often demand occurs and how large it is when it does. Alternatively, aggregate to a longer period or product group, or manage slow items with min-max stock rules rather than precise forecasts. Standard methods perform badly when most periods are zero. Spiky series also make MAPE useless, so evaluate with measures such as MAE, scaled errors or service-level outcomes. Practical options: - Forecast at group level, then allocate to items. - Use probabilistic forecasts to set stock to a service level. - Treat very slow items with reorder-on-demand or supplier arrangements. The real goal is usually a stocking decision, so test the policy rather than only the forecast. URL: https://zansa.co.za/answers/how-do-i-forecast-intermittent-or-lumpy-demand-for-slow-moving-items ### How can I forecast demand for a new product with no sales history? Borrow from similar products: use the launch curves of comparable items, adjusted for price, channel and marketing, and combine with judgement from sales staff. Early actual sales should then replace assumptions quickly. Expect wide uncertainty and plan stock in smaller, repeated orders where possible. Approaches include: - Analogues: choose past launches that resemble the new product. - Attribute-based models: learn how product attributes relate to early sales across many products. - Pre-orders, waiting lists or test markets as demand signals. - Scenario ranges (low, expected, high). Update weekly in the first weeks. Early sales can be misleading if distribution is still ramping, so separate the effect of availability from true demand. URL: https://zansa.co.za/answers/how-can-i-forecast-demand-for-a-new-product-with-no-sales-history ### How should I treat outliers in sales history before forecasting? First find out why each outlier happened. If it was a one-off event that will not repeat, such as a data error, a bulk order or a stock-out, correct or flag it. If it reflects a recurring event like a promotion or holiday, keep it and add it as an input. Never delete points without a reason. Automatic outlier removal can erase real information, such as a genuine surge you should be ready for next time. A sensible process: - Detect unusual points with simple rules or charts. - Investigate with people who know the business. - Classify as error, one-off, or recurring. - Adjust or flag accordingly and document it. Robust methods such as using the median are less sensitive to extreme values and are a useful safeguard. URL: https://zansa.co.za/answers/how-should-i-treat-outliers-in-sales-history-before-forecasting ### How do I deal with missing data in a time series? First work out why data is missing. A zero from a stock-out or a closed shop is different from a failed data feed. Then fill gaps using sensible methods such as carrying forward, interpolating or using same-period averages, or mark them and use a model that tolerates missing values. Document what you did. Distinguish true zeros (no sales happened) from missing values (we do not know). Treating missing as zero can seriously understate demand. Options: - Interpolate short gaps in smooth series. - Use seasonal averages for seasonal ones. - Exclude long gaps from training. - Fix the upstream cause so it stops recurring. Add data completeness checks to your pipeline so gaps are caught when they happen, not months later. URL: https://zansa.co.za/answers/how-do-i-deal-with-missing-data-in-a-time-series ### How do stock-outs distort demand forecasting and what can I do? When an item is out of stock, recorded sales fall to zero or low levels even though customers wanted it, so a model learns demand is lower than it really was. Flag stock-out periods using inventory records and either exclude them, or estimate the lost demand from comparable in-stock periods. This creates a vicious circle: the forecast is too low, you order too little, and you run out again. Approaches: - Record daily stock levels so you know when items were unavailable. - Replace affected periods with estimates from before and after, or from similar items. - Use sales rate when in stock rather than raw totals. - Track lost-sales proxies such as web searches or waiting lists. Without stock data, look for unusual runs of zeros in otherwise steady sellers. URL: https://zansa.co.za/answers/how-do-stock-outs-distort-demand-forecasting-and-what-can-i-do ### What is forecast value added and how do I measure it? Forecast value added (FVA) measures whether each step in a forecasting process improves accuracy compared with the previous step or a naive benchmark. You compare the error of, for example, the statistical forecast, the manager-adjusted forecast and the naive forecast. If a step makes things worse, it adds no value. The idea exposes wasted effort. Manual overrides sometimes improve forecasts when they include real market knowledge, but often add bias, for example through optimism. To measure it: - Save forecasts at each stage before actuals are known. - Calculate the same error metric for each stage. - Compare across enough periods to be meaningful. Use the results to decide where human input is worth it, such as promotions and new launches, and where to leave the numbers alone. URL: https://zansa.co.za/answers/what-is-forecast-value-added-and-how-do-i-measure-it ### Why are my sales forecasts always too optimistic and how do I fix it? Persistent over-forecasting usually comes from targets influencing the numbers, ignoring stock-outs or returns, over-weighting promotions, or models trained on a growth period. Measure bias, separate forecasts from targets, and then correct either the process or the model, for example by calibrating against past errors. A forecast says what you expect; a target says what you want. Mixing them produces bias. Steps: - Track signed error by salesperson, region and product group. - Share the bias results openly and without blame. - Remove incentives that reward high forecasts, or penalise low ones. - Test a bias correction factor in backtests. - Record and review the assumptions behind large upward adjustments. Once bias is visible, it is often among the quickest accuracy gains available. URL: https://zansa.co.za/answers/why-are-my-sales-forecasts-always-too-optimistic-and-how-do-i-fix-it ### Should I use Prophet, ARIMA, or gradient boosting for forecasting? There is no universal winner, so backtest candidates against a baseline on your data. ARIMA and exponential smoothing suit single series with modest data. Prophet-style models handle seasonality and holidays with little tuning. Gradient boosting works well across many related series with rich driver variables. Rules of thumb: - Few series, short history, simple patterns: exponential smoothing or ARIMA. - Strong holiday and seasonal effects with irregular data: a decomposable model like Prophet. - Hundreds or thousands of related items with promotions, prices and other features: gradient boosting or similar learning models. Pick on measured accuracy, but also consider speed, ease of explaining, and who will maintain it. A slightly less accurate model people understand and trust is often the better business choice. URL: https://zansa.co.za/answers/should-i-use-prophet-arima-or-gradient-boosting-for-forecasting ### Is it better to forecast in Excel or use a machine-learning model? Excel is often enough for a few products, stable demand and monthly planning. Machine learning becomes worthwhile when you have many items or locations, rich drivers, frequent forecasting cycles and a decision where small accuracy gains are valuable. Choose by the size of the problem, not by what sounds modern. Spreadsheet strengths are transparency and familiarity. Weaknesses are fragility, version confusion, errors in formulas and difficulty refreshing at scale. Signs you have outgrown Excel: - Forecasts take days to update. - Several people maintain different copies. - You need per-store, per-product forecasts that refresh automatically. - Promotions and prices must be modelled together. An intermediate step is a scripted statistical forecast on a database, which stays simple but is repeatable and auditable. URL: https://zansa.co.za/answers/is-it-better-to-forecast-in-excel-or-use-a-machine-learning-model ### How do I forecast accounts receivable and customer payment timing? Look at how long each customer, or group of similar customers, historically takes to pay after invoicing, then apply those patterns to open invoices to predict when cash arrives. This is more realistic than assuming everyone pays on the due date, and it improves cash-flow forecasts. Group customers by payment behaviour, size, sector or country. Calculate the distribution of days-to-pay for each group, not just the average, because late payers matter most for cash. A simple model assigns each open invoice a probability of being paid in each coming week. Machine learning can refine this using customer history, invoice size and reminder activity, but needs enough past invoices. Track actual receipts against the forecast each week and adjust. URL: https://zansa.co.za/answers/how-do-i-forecast-accounts-receivable-and-customer-payment-timing ### What data do I need to build a customer churn model? You need a clear churn definition, a history of customers with their start dates and outcomes, and behaviour measured before the outcome date, such as purchase frequency, spend, product mix, support contacts and engagement. Customer attributes help, but behaviour over time is usually most predictive. Build a snapshot table: for each customer at a past date, record what you knew then and whether they churned in the following window, say 90 days. Use many snapshots to give the model enough examples. Beware of leakage from fields that are updated after leaving. Also expect class imbalance, where most customers do not churn, so use appropriate metrics such as precision, recall or lift, not just overall accuracy. Check that you hold enough churn events to learn from. URL: https://zansa.co.za/answers/what-data-do-i-need-to-build-a-customer-churn-model ### How do I know if a churn prediction model is any good? Test it on later data and check that the customers it ranks as high-risk really leave more often than others. Useful measures are precision, recall, lift or capture rate in the top-ranked group. Most importantly, check whether acting on the predictions reduces churn compared with a control group. Accuracy alone is misleading when few customers churn: a model that predicts nobody leaves can look highly accurate. Practical checks: - Rank customers by risk and measure churn in each decile. - Compare with a simple rule such as days since last purchase. - Check whether predictions are well calibrated, meaning a 30 percent risk really means about 30 percent. - Review which factors drive predictions for sense. A model is only valuable if the business can intervene effectively on the flagged customers. URL: https://zansa.co.za/answers/how-do-i-know-if-a-churn-prediction-model-is-any-good ### What is RFM analysis and how do I use it? RFM scores each customer on recency (how recently they bought), frequency (how often) and monetary value (how much they spend). Combining the three produces simple segments such as champions, loyal, at risk and lost. It needs only transaction data and is easy to explain, making it a strong starting point. Typical steps are to calculate the three measures for each customer, rank them into bands such as quintiles, and combine them into groups. Each group gets a tailored action, for example a thank-you for top customers or a win-back offer for lapsing ones. Limits: it ignores product type and margin, treats all spend equally, and band cut-offs are arbitrary. Adapt the time windows to your purchasing cycle, since a monthly grocery buyer and a once-a-year equipment buyer behave very differently. URL: https://zansa.co.za/answers/what-is-rfm-analysis-and-how-do-i-use-it ### What is price elasticity and how can I estimate it? Price elasticity measures how much quantity demanded changes when price changes. If a 10 percent price rise cuts volume by 5 percent, elasticity is about minus 0.5. Estimate it from past price changes while controlling for promotions, seasonality and stock, or run controlled price tests. Observational estimates are risky because prices rarely change at random. If you raise prices during peak season, volume may stay high for reasons unrelated to price sensitivity, leading to wrong conclusions. Better evidence comes from tests: vary prices across comparable stores, regions, customer groups or time periods and compare. Elasticity also differs by product, customer type and competitive context, and is not constant across all price levels. Combine estimates with margin calculations to see the effect on profit, not only on volume. URL: https://zansa.co.za/answers/what-is-price-elasticity-and-how-can-i-estimate-it ### How do I run a price test without damaging customer trust? Test small, defined changes on limited groups, products or periods, keep the comparison fair, and avoid showing different customers visibly different prices for identical items in the same place. Define success metrics such as profit per visitor before starting, and run long enough to cover normal weekly cycles. Options include testing across stores or regions, testing new prices on new products, or alternating prices across time periods. Be mindful of fairness, local consumer laws and the reputational risk when customers notice price differences. Guidelines: - Change one thing at a time. - Compare against similar untested groups. - Watch for stock and promotion effects. - Measure margin, not just units. - Agree a stop rule if results turn negative. Avoid price discrimination based on sensitive personal characteristics. URL: https://zansa.co.za/answers/how-do-i-run-a-price-test-without-damaging-customer-trust ### What are the common methods for detecting anomalies in business data? Common methods include threshold rules, comparison with the same period in the past, statistical limits based on standard deviation or median absolute deviation, forecast-based checks that flag actuals far outside the predicted range, and machine-learning methods like isolation forests. Start simple and add complexity only if needed. Forecast-based detection is especially practical: fit a model that accounts for seasonality and holidays, then flag points where the residual is unusually large. This avoids false alarms from normal patterns. Tips: - Define what counts as unusual in business terms and in money. - Tune alert sensitivity against past known incidents. - Group related alerts so teams are not flooded. - Record whether each alert was real, so the system can be tuned. Only alert on things someone can act on. URL: https://zansa.co.za/answers/what-are-the-common-methods-for-detecting-anomalies-in-business-data ### How do I reduce false alarms from an anomaly detection system? Model normal patterns first, including seasonality, weekdays and promotions, so predictable swings are not flagged. Then tune thresholds on past data, require anomalies to persist or exceed a money value before alerting, and feed back whether each alert was real. Fewer, better alerts are acted on more reliably. Alert fatigue is the main reason monitoring projects fail: people learn to ignore noisy messages. Tactics: - Alert on residuals from a good forecast, not on raw values. - Add materiality filters, such as ignoring deviations below a set amount. - Combine signals, for example a drop in orders together with a rise in errors. - Review a sample of alerts monthly and adjust. - Give each alert type an owner and a suggested action. URL: https://zansa.co.za/answers/how-do-i-reduce-false-alarms-from-an-anomaly-detection-system ### How do I forecast foot traffic or customer arrivals for staffing? Use counts of arrivals by hour or shift over at least a year, and model day of week, time of day, season, holidays, weather and events. Start with a seasonal average by weekday and hour, then refine. Convert the result to staffing using how many customers one person can serve in an hour. Sources can be till transactions, door counters, bookings or call logs. Be aware that transactions undercount when queues cause walk-outs. Helpful details: - Forecast at 15-minute to hourly grain depending on how flexible your rosters are. - Include marketing activity and nearby events. - Check forecast error at peak times specifically, as that is where service suffers. - Review with managers who know local quirks. Roster rules and employee preferences are separate constraints from the forecast itself. URL: https://zansa.co.za/answers/how-do-i-forecast-foot-traffic-or-customer-arrivals-for-staffing ### What is a feature in machine learning, in simple terms? A feature is an input variable the model uses to make a prediction, such as day of week, price, days since last purchase or whether a promotion was running. Choosing and building good features, known as feature engineering, often matters more to results than choosing a fancy algorithm. For forecasting, typical features include calendar variables, lagged sales (sales a week or year ago), rolling averages, prices, promotion flags and holidays. One rule is crucial: features must only use information known at the time of prediction. Lagged values need to respect the forecast horizon, because if you forecast four weeks ahead, last week's sales will not be available for the final week of the forecast. Keep features few, meaningful and documented. URL: https://zansa.co.za/answers/what-is-a-feature-in-machine-learning-in-simple-terms ### Can I explain why a machine-learning forecast gave a particular number? Often yes, to a useful degree. Simple models show their drivers directly, while complex ones can be examined with tools such as feature importance and Shapley-value explanations, which estimate how much each input pushed a prediction up or down. These explain the model's behaviour, though not necessarily real-world cause. Explanations help managers trust and challenge forecasts, for example by seeing that a high forecast was driven by a planned promotion and a holiday. Cautions: - Importance is not causation. - Correlated inputs can share credit in confusing ways. - Explanations are approximate summaries. If full transparency is required, such as for regulated decisions about people, prefer inherently interpretable models. Always pair explanations with accuracy evidence and a sanity check by domain experts. URL: https://zansa.co.za/answers/can-i-explain-why-a-machine-learning-forecast-gave-a-particular-number ### How does poor data quality affect forecasts and what should I fix first? Poor data quality leads to wrong forecasts however good the model: duplicates inflate sales, missing days appear as zero demand, and inconsistent product codes split a history in two. Fix definitions, duplicates, gaps and product or customer identifiers first, since these cause the largest and most systematic errors. A short checklist: - Are dates and time zones consistent? - Do product codes stay stable when items are renamed or replaced? - Are returns, cancellations and discounts recorded? - Are there gaps or duplicate records? - Do totals match the finance system? Reconciling with accounting totals is a powerful test. Also set up automated checks that run on every data refresh, so problems are found early, and assign an owner for each key data source. URL: https://zansa.co.za/answers/how-does-poor-data-quality-affect-forecasts-and-what-should-i-fix-first ### How can I tell whether my business is ready for data science? You are likely ready when you have a specific recurring decision, a reasonable history of reliable data, one person who will use and own the results, and agreed definitions of key measures. If data is scattered or untrusted, fix that first, because models inherit its problems. Questions to ask: - Which decision would improve if we had a better prediction? - What is that improvement worth in money or time? - Do we have at least a year or two of relevant data? - Who will act on the output? - Can we measure success? If the answers are unclear, start with reporting and data cleaning, or a simple baseline forecast. Readiness is about decisions and data discipline, not company size. URL: https://zansa.co.za/answers/how-can-i-tell-whether-my-business-is-ready-for-data-science ### How often should I update my business forecast each period? Update as often as new information arrives and decisions are made. Weekly is common for sales and inventory, daily for staffing and short-term operations, and monthly for budgets. Refreshing the data frequently is cheap; re-fitting or retraining the model can be done less often unless accuracy drops. A rolling forecast, which always looks the same number of periods ahead, is more useful than an annual one that goes stale by March. Change the forecast when facts change, not just because the calendar says so. But avoid constant fiddling: if the forecast swings wildly between updates, planners lose confidence and operations become unstable. Track forecast stability alongside accuracy, and agree a fixed cut-off after which plans freeze. URL: https://zansa.co.za/answers/how-often-should-i-update-my-business-forecast-each-period ### What are the warning signs that a forecasting model is no longer reliable? Warning signs include rising error or growing bias compared with the baseline, forecasts consistently overridden by staff, residuals showing new patterns, inputs outside their usual range, and changes in the business such as new products, prices or channels. Any of these should trigger a review and possibly retraining. Also watch for silent data problems: a feed that stops updating, a code change, or a field that changes meaning. Responses can range from a simple recalibration to rebuilding the model. In the meantime, fall back on the baseline forecast and tell users to treat the model's output with caution. Having a documented process for this is part of running models responsibly, rather than treating them as fit-and-forget. URL: https://zansa.co.za/answers/what-are-the-warning-signs-that-a-forecasting-model-is-no-longer-reliable ### Why does my forecast look accurate in testing but fail in real life? The usual causes are data leakage, random rather than time-based splits, tuning against the test set, a test period unrepresentative of the future, or inputs available in testing but not at forecast time. Real conditions may also have changed since the test period. Check each in turn: - Were any future values used, even indirectly? - Was the split by time? - Were you testing at the same horizon used in practice? - Did you select the best of many models on the same test data? - Would inputs such as planned promotions be known in advance? A useful safeguard is a final untouched test period, plus live tracking for a few weeks before relying on the model for important decisions. URL: https://zansa.co.za/answers/why-does-my-forecast-look-accurate-in-testing-but-fail-in-real-life ### How do I measure the business value of a better forecast? Translate accuracy into money: fewer stock-outs and lost sales, less excess stock and write-off, lower overtime or idle labour, and better cash use. Compare outcomes before and after, or between groups using the new and old methods, rather than quoting accuracy percentages alone. A rough approach is to simulate decisions: apply your ordering or staffing rules to past data using the old forecast and the new one, and compare resulting costs. Be honest about other changes that may influence results. Value depends on whether decisions actually change, so a more accurate forecast nobody uses is worth nothing. Include the running cost of the model, such as cloud fees and maintenance time, when calculating return. URL: https://zansa.co.za/answers/how-do-i-measure-the-business-value-of-a-better-forecast ### How much does machine-learning forecasting cost for a small or mid-sized company? Cost varies widely with data readiness, number of data sources and scope. As Zansa's guidance, a Data Health Check starts from R25,000, builds typically run R80,000 to R250,000, and ongoing support starts from R12,000 a month, all in South African rand. Other providers will differ. Main cost drivers are cleaning and connecting data, the number of forecasts and dashboards, and how much automation and monitoring are needed. Cloud running costs are usually modest for small data, but should be checked. To compare quotes, ask what is included, who owns the code and models, how accuracy will be measured against a baseline, and what ongoing maintenance is required. A fixed price and timeline in writing reduces surprises. URL: https://zansa.co.za/answers/how-much-does-machine-learning-forecasting-cost-for-a-small-or-mid-sized-company ### What is Zansa and what does it do? Zansa is a data analytics and data science team that works with small and mid-sized companies anywhere in the world. Based in South Africa and working remotely, it builds dashboards and machine-learning forecasts, and helps companies get their data in order so it can support decisions. Zansa works across sectors including retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. Its engagements follow a clear path: a free intro call, a fixed-fee Data Health Check, a Foundation Build with a fixed price and timeline agreed in writing, and optional ongoing support. The aim is practical analytics that a company can use and own, rather than a black box that only the provider understands. URL: https://zansa.co.za/answers/what-is-zansa-and-what-does-it-do ### Why is the company called Zansa and what does the name mean? Zansa comes from the Japanese word 残差 (zansa), meaning residual: the part of the data that a model leaves unexplained. In forecasting and statistics, studying the residual shows what a model has missed, which is where improvement and insight usually come from. A residual is the actual value minus the model's prediction. If a model has captured the real patterns, what remains looks like random noise. If it has not, the leftover shows structure, such as a missed seasonal effect or a one-off event. The name reflects an approach of looking carefully at what is not yet explained, being honest about the limits of a model, and treating error analysis as a routine part of the work rather than an afterthought. URL: https://zansa.co.za/answers/why-is-the-company-called-zansa-and-what-does-the-name-mean ### What is Zansa's process for working with a new client? Zansa starts with a free intro call, then a fixed-fee Data Health Check lasting one to two weeks, then a Foundation Build typically taking four to eight weeks with fixed price and timeline agreed in writing. Optional ongoing support follows after the build. Each stage has a purpose: - Intro call: understand the business problem and whether Zansa is a fit. - Data Health Check: assess data quality, sources and opportunities, and recommend what to build. - Foundation Build: deliver the agreed dashboards, data pipelines and forecasts. - Support: monitoring, updates and improvements if the client wants them. The stages let a company commit step by step, with a clear scope and cost before each move, instead of signing up to a large open-ended project. URL: https://zansa.co.za/answers/what-is-zansa-s-process-for-working-with-a-new-client ### What happens in a Zansa Data Health Check? The Data Health Check is a fixed-fee review lasting one to two weeks. Zansa examines your data sources, quality and definitions, identifies gaps and opportunities, and recommends what to build first. It gives you a clear picture of what your data can support before you commit to a larger project. The review typically looks at where data lives, how reliable it is, whether key measures are defined consistently, and which decisions could benefit from dashboards or forecasts. From the Zansa pricing guidance, Health Checks start from R25,000 (South African rand). The outcome is a practical plan, which you can use with Zansa or independently. Understanding data readiness first reduces the risk of building forecasts on foundations that cannot support them. URL: https://zansa.co.za/answers/what-happens-in-a-zansa-data-health-check ### What does a Zansa Foundation Build include and how long does it take? A Foundation Build typically takes four to eight weeks, with the price and timeline agreed in writing before work starts. It creates the data foundations and deliverables identified in the Health Check, such as dashboards and machine-learning forecasts, built to run in the client's own cloud account. Zansa guidance puts builds typically between R80,000 and R250,000, depending on scope. The client owns everything built. The work usually involves connecting and cleaning data sources, defining shared measures, building the dashboards, and, where justified, creating and backtesting forecasts against simple baselines. Because the scope is fixed and written down, a company knows the cost and delivery period up front rather than facing open-ended hourly billing. URL: https://zansa.co.za/answers/what-does-a-zansa-foundation-build-include-and-how-long-does-it-take ### Who owns the models and dashboards that Zansa builds? The client owns everything Zansa builds. Machine-learning models run inside the client's own cloud account, usually Google BigQuery, so the data, code and models sit under the client's control rather than on a provider's platform. This avoids lock-in if the relationship ends. Ownership matters because forecasting systems become part of daily operations. If a provider holds the models, a company can lose its forecasting capability when it changes supplier. When evaluating any analytics provider, ask: - Where does the data live? - Who holds the code and the model files? - Can another team take over? - Is documentation provided? With models hosted in your own cloud account, you keep access and can change who maintains them. URL: https://zansa.co.za/answers/who-owns-the-models-and-dashboards-that-zansa-builds ### Where do Zansa's machine-learning models run and who controls them? Zansa's machine-learning models run inside the client's own cloud account, usually Google BigQuery. Data does not need to be copied to Zansa's systems for modelling, and the client keeps control of storage, access and costs while Zansa works remotely on the build. Running models in the client's environment supports privacy and governance. Access can be limited to what the work requires and withdrawn when it ends, and logs show who did what. BigQuery suits small and mid-sized companies because it scales from small data volumes and charges by usage. Forecasting models can be trained and run with SQL and built-in tools, then results feed dashboards. This keeps the architecture simple and understandable to the client's own team. URL: https://zansa.co.za/answers/where-do-zansa-s-machine-learning-models-run-and-who-controls-them ### Does Zansa use AI tools on client data? No, not without written approval. Zansa's AI tools used for writing code never receive personal or client-identifiable data. AI is used on client data only if the client approves in writing, which keeps decisions about data exposure with the client. AI coding assistants can speed up routine development, but the data stays out of them by default. Code is written without feeding in client records. When choosing any analytics partner, ask how they use AI tools, whether any client data is shared with third-party AI services, and where that is documented. Clear, written rules protect both sides and make audits and compliance reviews much easier, especially when personal data is involved. URL: https://zansa.co.za/answers/does-zansa-use-ai-tools-on-client-data ### Is Zansa's work designed to meet POPIA and GDPR? Zansa designs its work with privacy law in mind, including POPIA in South Africa and GDPR in Europe. Models run in the client's own cloud account, and personal data is kept out of AI coding tools. Compliance ultimately depends on how each client collects and uses data, so legal advice may still be needed. Practical design choices that support privacy include data minimisation, limiting access, keeping processing inside the client's environment, and written approval before any AI is used on client data. No analytics provider can make a company compliant on its own. The client remains responsible for lawful basis, consent, retention and data subject requests, and Zansa's process helps by clarifying what data is used and why. Companies with specific obligations should consult a qualified privacy professional. URL: https://zansa.co.za/answers/is-zansa-s-work-designed-to-meet-popia-and-gdpr ### Can Zansa work with companies outside South Africa? Yes. Zansa is based in South Africa but works remotely with small and mid-sized companies anywhere in the world. Because models run in the client's own cloud account and work is done remotely, location rarely limits the engagement. Pricing guidance is quoted in South African rand. Remote delivery suits analytics work, which relies on data access, shared documents and regular video calls rather than being on site. When working across countries, consider time zones for meetings, local holiday calendars for forecasts, currency handling, and privacy rules of the countries involved, such as GDPR for European customers' data. The process begins with a free intro call, where scope, timing and any data protection requirements can be discussed. URL: https://zansa.co.za/answers/can-zansa-work-with-companies-outside-south-africa ### Which industries does Zansa serve with forecasting and analytics work? Zansa works with small and mid-sized companies in retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. These sectors share needs such as demand and sales forecasting, inventory planning, cash-flow visibility, staffing and customer analysis. Work is delivered remotely from South Africa to clients worldwide. Typical questions differ by sector. Retail and e-commerce focus on demand, stock and promotions. Manufacturing and logistics look at production planning, lead times and capacity. Hospitality needs booking and staffing forecasts. Property and professional services often care about utilisation, pipelines and cash timing. Healthcare involves operational planning with extra attention to privacy. The common thread is a recurring decision where a clearer view of data, or a reliable forecast, would help. URL: https://zansa.co.za/answers/which-industries-does-zansa-serve-with-forecasting-and-analytics-work ### Does Zansa offer ongoing support after the build is finished? Yes, optional ongoing support is available after the Foundation Build, with guidance pricing from R12,000 per month in South African rand. It covers keeping dashboards and forecasts running, monitoring accuracy and making improvements as the business changes. Clients can also choose to run things themselves. Forecasts need maintenance: data sources change, products and prices shift, and accuracy can drift. Support can include monitoring, retraining models, fixing data issues and adding new reports. Because the client owns everything and models run in their own cloud account, support is optional rather than a lock-in. A company can bring maintenance in-house or use another provider. Ask any provider what monitoring is included and how problems are reported. URL: https://zansa.co.za/answers/does-zansa-offer-ongoing-support-after-the-build-is-finished ### When should a company hire a data scientist instead of using a consultancy? Hire in-house when you have a steady pipeline of data science problems, enough data infrastructure, and the management capacity to guide and retain someone. Use an external team when needs are project-based, you need several skills briefly, or you want results before committing to a permanent role. A single hire can struggle without data engineering, domain support and clear priorities. Salary, recruitment time and turnover risk matter for small companies. A common path is to start with an external team to build foundations, then hire to run and extend them. Insist that everything is documented and owned by you, so a future hire can take over. Zansa, for example, builds in the client's own cloud account and the client owns the result, which supports either path. URL: https://zansa.co.za/answers/when-should-a-company-hire-a-data-scientist-instead-of-using-a-consultancy ### Can a small company really benefit from machine-learning forecasts, or is it only for big firms? Yes, if the problem fits. Small and mid-sized companies can benefit when they have repeated decisions, such as buying stock or scheduling staff, enough clean history, and many product-location combinations that are too numerous to forecast manually. Cloud tools have made the technology affordable, but data and discipline still matter. Where small firms often gain is in using modest, well-chosen models rather than elaborate ones. A seasonal baseline plus a model that adds promotions and prices can be enough. Test honestly: if a simple method performs nearly as well, use it. Start with a pilot on one product group or location, measure the effect on decisions, then scale. Zansa works with small and mid-sized firms on this basis, comparing machine learning with simple baselines first. URL: https://zansa.co.za/answers/can-a-small-company-really-benefit-from-machine-learning-forecasts-or-is-it-only ### How does Zansa use residuals to check a forecast is trustworthy? Zansa treats residuals, the gaps between actual results and model predictions, as a diagnostic tool. If they look like random noise, the model has captured the main patterns. If they show structure, such as errors on holidays or growing with volume, the model is missing something and needs refining before it is relied upon. This reflects the origin of the name, from the Japanese 残差 (zansa), meaning residual. Typical residual checks include plotting errors over time, by weekday and month, around promotions and across product groups, looking for consistent over- or under-prediction, and checking whether the spread of errors is stable. Findings feed into model changes, such as adding a holiday flag, and into the width of prediction ranges shown to users, so forecasts are presented honestly. URL: https://zansa.co.za/answers/how-does-zansa-use-residuals-to-check-a-forecast-is-trustworthy ### What should I ask a data science provider before hiring them to build forecasts? Ask how they will measure accuracy against a simple baseline, who owns the code and models, where data and models will run, how AI tools are used with your data, what the fixed price and timeline are, and how monitoring and handover work. Clear written answers signal a mature provider. Further useful questions: - Will you backtest over several periods and show bias as well as error? - How will you handle stock-outs, promotions and holidays? - What happens if the model stops working? - Can my own team maintain it? - How is personal data protected under laws such as POPIA or GDPR? Be cautious of promises of guaranteed accuracy before seeing your data, since achievable accuracy depends on how predictable the underlying demand is. URL: https://zansa.co.za/answers/what-should-i-ask-a-data-science-provider-before-hiring-them-to-build-forecasts ### Is it safe to upload sales spreadsheets with customer names to an AI chatbot? Generally no. Customer names and contact details are personal data, and uploading them to a public AI chatbot may breach privacy law, your contracts or customer expectations. Remove or replace identifiers first, share only the aggregates the task needs, or use an approved business tool with proper data agreements. Before uploading anything, ask: - Does the file contain names, emails, phone numbers, addresses or ID numbers? - Is the tool's retention and training policy acceptable? - Does a data processing agreement exist? - Could aggregated or pseudonymised data do the job? For many analyses, customer IDs replaced by random codes and totals by product and month are enough. Note that pseudonymised data can still count as personal data if people could be re-identified, so remove unnecessary columns too. URL: https://zansa.co.za/answers/is-it-safe-to-upload-sales-spreadsheets-with-customer-names-to-an-ai-chatbot ### What is the difference between anonymised and pseudonymised data? Anonymised data has been altered so individuals cannot be identified by any reasonably likely means, so privacy laws generally no longer treat it as personal data. Pseudonymised data replaces direct identifiers with codes, but the person could still be re-identified using extra information, so it usually remains personal data. Replacing a customer name with an ID number is pseudonymisation. If you keep a lookup table, or if the remaining fields (postcode, birth date, purchase history) could single someone out, the data is still linkable to a person. True anonymisation is hard. Techniques include aggregation, removing or generalising fields, and adding noise. Small groups are risky: a count of one in a small area may identify someone. Pseudonymisation is still valuable as a safeguard, but check the specific legal tests that apply to you. URL: https://zansa.co.za/answers/what-is-the-difference-between-anonymised-and-pseudonymised-data ### What are hallucinations in AI and how do they affect data analysis? Hallucinations are confident but false statements generated by a language model, such as invented figures, nonexistent sources or wrong explanations. In data analysis they can appear as fabricated numbers, misdescribed methods or code that runs but is wrong. Always verify against source data and run the code yourself. Models generate plausible text, not guaranteed truth. They can be especially unreliable with exact arithmetic, obscure statistics and recent events. Reduce risk by: - Asking for formulas or code, then checking them. - Reconciling outputs with known totals. - Requesting reasoning step by step and reviewing it. - Never citing statistics from a chatbot without finding the original source. Treat AI as a fast but fallible junior assistant, with a human accountable for the final analysis. URL: https://zansa.co.za/answers/what-are-hallucinations-in-ai-and-how-do-they-affect-data-analysis ### Should my business have an AI usage policy for staff handling data? Yes. A short written policy reduces the risk of staff pasting confidential or personal data into public AI tools. It should state which tools are approved, what data may never be shared, how outputs must be checked, and who to ask when unsure. Review it regularly as tools change. Elements of a practical policy: - Approved tools and account types. - Data categories that are prohibited, such as personal data, credentials and client contracts. - Rules on anonymising or aggregating before use. - A requirement to verify numbers, code and legal or financial claims. - Disclosure rules for AI-assisted work to clients where relevant. Pair it with brief training. Policies fail when they are so strict that staff quietly work around them, so provide approved alternatives that actually meet their needs. URL: https://zansa.co.za/answers/should-my-business-have-an-ai-usage-policy-for-staff-handling-data ### Do AI vendors train their models on the data I send them? It depends on the vendor, the product tier and your settings. Some consumer services may use inputs to improve models unless you opt out, while many business and API offerings state they do not train on customer data by default. Read the current terms and data processing agreement before sending anything sensitive. Check: - Whether inputs and outputs are retained, and for how long. - Whether humans may review conversations. - Where data is processed and stored. - Whether there is a signed data processing agreement. - How to delete data. Terms change, so do not rely on memory or assumptions. If the terms are unclear, treat the tool as public and share nothing confidential or personal. URL: https://zansa.co.za/answers/do-ai-vendors-train-their-models-on-the-data-i-send-them ### How can LLMs help with forecasting and analytics without risking my data? Use them for tasks that need no real data: explaining methods, drafting SQL or Python against a described schema with made-up sample rows, reviewing logic, writing documentation and brainstorming features. Run the generated code yourself inside your own environment, where the real data stays, and check the results. This keeps the data in your controlled systems while still gaining speed. Describe column names and types rather than pasting records, and use synthetic examples to test. LLMs are also useful for turning technical results into plain-language summaries, provided the figures you give them contain nothing sensitive and you check the wording against the facts. Zansa follows this approach: AI coding tools never receive personal or client-identifiable data, and AI is not used on client data without written approval. URL: https://zansa.co.za/answers/how-can-llms-help-with-forecasting-and-analytics-without-risking-my-data ### How do I forecast when something big has changed, like a pandemic or new competitor? When the past no longer represents the future, treat the disrupted period carefully: flag or down-weight it, use shorter recent windows, rely more on scenarios and judgement, and update frequently. Do not blindly extrapolate or ignore the break. Track errors closely and widen your forecast ranges until patterns stabilise. Options include: - Marking the disruption with an indicator variable so the model does not treat it as normal. - Excluding the affected period from seasonal estimates. - Using leading indicators, such as enquiries, bookings or web traffic. - Building scenarios for how quickly recovery or decline happens. Be honest that statistical models cannot foresee unprecedented events. Keep a simple baseline running for comparison, and pair statistical output with input from people close to customers. URL: https://zansa.co.za/answers/how-do-i-forecast-when-something-big-has-changed-like-a-pandemic-or-new-competit ### How do I forecast revenue for a service or professional-services firm? Build it from drivers: capacity (billable staff and hours), utilisation, rates, and the sales pipeline weighted by historical conversion and timing, plus recurring contracts. This is usually more reliable than extrapolating past revenue, because work is constrained by people and driven by deals, not just seasonality. Split revenue into recurring or contracted, probable from pipeline, and speculative. Apply stage-based win rates from your own history instead of guesses, and include expected delays between winning and billing. Check capacity: a forecast that exceeds what staff can deliver is unrealistic. Track utilisation, write-offs and payment delays, since they affect cash as well as revenue. Review forecast versus actual monthly and refine win rates and timing. URL: https://zansa.co.za/answers/how-do-i-forecast-revenue-for-a-service-or-professional-services-firm ### How can hospitality businesses forecast bookings and occupancy? Combine booking pace, which is bookings on hand compared with the same point in previous years, with seasonality, events, holidays, pricing and cancellations patterns. Forecast final occupancy from how reservations typically accumulate before arrival, and update daily. This supports pricing, staffing and purchasing decisions. Booking pace is a powerful short-term signal: if you are well ahead of last year at 30 days out, demand is likely stronger. Additional considerations: - Cancellation and no-show rates by channel. - Length-of-stay patterns. - Local events and school calendars. - Group bookings treated separately from individual ones. Restaurants can forecast covers by day and service and link them to ingredient ordering and staffing. Keep forecasts at the granularity of the decision, such as per night or per service. URL: https://zansa.co.za/answers/how-can-hospitality-businesses-forecast-bookings-and-occupancy ### What data is needed to forecast production and raw-material needs for a manufacturer? You need demand history or orders, current stock, work in progress, bills of materials, supplier lead times and minimum order quantities, production capacity and scrap rates. Forecast end-product demand, then convert it to material requirements through the bill of materials, adjusting for stock on hand and orders in transit. Lead time variability and supplier reliability often matter more than the demand forecast itself, so track actual delivery performance. Useful practices: - Forecast at product-family level when item-level data is noisy. - Separate make-to-order from make-to-stock items. - Set safety stock by material criticality and lead time. - Capture planned downtime and maintenance. Compare forecast material usage with actual consumption to find errors in bills of materials or scrap assumptions. URL: https://zansa.co.za/answers/what-data-is-needed-to-forecast-production-and-raw-material-needs-for-a-manufact ### How can a logistics company use forecasting to plan capacity and routes? Forecast shipment volumes, parcels or orders by day, region and service type, then convert them into vehicles, drivers, warehouse labour and storage needs. Account for peak seasons, promotions by clients, public holidays and delivery time windows. Good short-term forecasts reduce overtime, idle capacity and missed deliveries. Volumes from large clients may be known in advance through their promotional calendars, so sharing information improves accuracy. Consider: - Forecast at the level of the dispatch decision, such as depot and day. - Error at peak days, which drives service failures. - Separate forecasts for weight, volume and number of stops where they constrain different resources. - Fall-back capacity such as temporary labour or subcontractors. Routing itself is an optimisation problem that uses the forecast as an input. URL: https://zansa.co.za/answers/how-can-a-logistics-company-use-forecasting-to-plan-capacity-and-routes ### Can healthcare practices use forecasting while protecting patient privacy? Yes. Many useful forecasts, such as appointment demand, no-shows, stock use and staffing needs, can be built on aggregated or de-identified data. Keep patient-identifiable information out of analytics wherever possible, restrict access, process data in a controlled environment and follow the health data rules that apply to you. Health data is generally treated as especially sensitive under privacy laws such as POPIA and GDPR, so data minimisation is key. Ask whether the forecast needs individual patient records at all, or whether daily counts by service are enough. Never put patient data into public AI tools. Document who has access and why, and involve your privacy or compliance officer early. Models that influence individual patient care carry extra requirements and need clinical oversight. URL: https://zansa.co.za/answers/can-healthcare-practices-use-forecasting-while-protecting-patient-privacy ### How can e-commerce stores forecast demand across many products and channels? Forecast at the level you buy and fulfil, using sales history by product and channel, promotions, price, traffic, seasonality and stock availability. For large catalogues use methods that learn across products and group similar items, and treat new, slow and bestselling items with different approaches. E-commerce sales are strongly affected by marketing activity, so record campaigns, email sends and ad spend. Also record out-of-stock periods and page views, which indicate unmet demand. Additional points: - Returns affect net demand and should be modelled. - Event days such as Black Friday behave differently from normal days. - Review accuracy by product tier, using WAPE and bias. A machine-learning model across the catalogue can beat per-item methods if you have enough data, but test against seasonal baselines. URL: https://zansa.co.za/answers/how-can-e-commerce-stores-forecast-demand-across-many-products-and-channels ### How do I get started with data science if my data is messy and scattered? Start by choosing one valuable decision, list the data it needs, and consolidate that data into one place with clear definitions. Fix the most important quality issues, build a simple dashboard to establish trust in the numbers, then try a baseline forecast. Only then consider machine learning. A step-by-step path: - Pick one use case, such as weekly stock ordering. - Identify sources and who owns them. - Export or connect data into a central store. - Reconcile totals to finance records. - Document definitions. - Report, then forecast. Progress in small, finished steps beats a large plan that stalls. A structured review, like Zansa's fixed-fee Data Health Check, is one way to find what is usable before committing to a larger build. URL: https://zansa.co.za/answers/how-do-i-get-started-with-data-science-if-my-data-is-messy-and-scattered ## Fundamentals ### What is data analytics and what does it mean for a small business? Data analytics is the process of collecting, cleaning and examining business data to answer questions and support decisions. For a small business it usually means turning sales, finance, stock, marketing and customer records into clear numbers that show what is happening, why, and what to do next. Analytics ranges from simple reporting, such as monthly revenue by product, to deeper questions such as why margins fell in one region. It does not require a big team or expensive software. The core ingredients are: - Data that is reasonably accurate and consistently defined - A clear question the business wants answered - A way to present the answer so people actually use it The value comes from better decisions, not from the charts themselves. URL: https://zansa.co.za/answers/what-is-data-analytics-and-what-does-it-mean-for-a-small-business ### What is business intelligence (BI) and how is it different from analytics? Business intelligence is the set of tools and practices that gather company data and present it in reports and dashboards so people can monitor performance. Analytics is broader and more investigative. BI usually answers what happened and what is happening now, while analytics also explores why and what might happen next. In practice the terms overlap and many vendors use them interchangeably. A useful rule of thumb: - BI: recurring, standardised reporting such as a weekly sales dashboard - Analytics: ad hoc investigation, such as why a customer segment stopped buying Most small and mid-sized companies need solid BI first. Deeper analytics becomes valuable once the basic numbers are trusted and agreed across the business. URL: https://zansa.co.za/answers/what-is-business-intelligence-bi-and-how-is-it-different-from-analytics ### What is the difference between data analytics and data science? Data analytics mainly describes and explains what has happened using reports, queries and statistics. Data science goes further by building predictive or automated models, often using machine learning and programming. Analytics supports everyday decisions, while data science is typically used for forecasting, classification or recommendations once the foundations are in place. The boundary is blurry, and job titles vary between organisations. Broadly: - Analysts focus on business questions, metrics and communication - Data scientists focus on modelling, experimentation and prediction Most smaller companies get more value from strong analytics first. Predictive models are only as good as the underlying data, so skipping the foundations tends to produce models nobody trusts or uses. URL: https://zansa.co.za/answers/what-is-the-difference-between-data-analytics-and-data-science ### What is a dashboard in business terms and what should it show? A dashboard is a single screen that displays the key numbers a team needs to monitor performance, usually as charts, tables and headline figures that update from underlying data. A good one shows a handful of agreed metrics for a specific audience and makes it obvious what needs attention. Effective dashboards are built around decisions, not around available data. Before building, ask who will use it, how often, and what action a change in each number should trigger. Common features of a useful dashboard: - A small number of metrics with clear definitions - Comparison to a target, prior period or benchmark - Filters for the dimensions people really use, such as date or region - Drill-down to the detail behind a number URL: https://zansa.co.za/answers/what-is-a-dashboard-in-business-terms-and-what-should-it-show ### What is a KPI and how is it different from a regular metric? A KPI, or key performance indicator, is a metric tied directly to a business goal and used to judge progress. Any number you can measure is a metric, but only the few that reflect success against an objective deserve to be called KPIs. Page views are a metric; qualified leads per month might be a KPI. A good KPI is clearly defined, measurable from reliable data, owned by a person, and linked to a target or threshold. If nobody would change behaviour when the number moves, it is probably just a metric. Typical examples differ by business: gross margin and stock turnover for retail, on-time delivery for logistics, utilisation for professional services. Keep the list short so attention is not diluted. URL: https://zansa.co.za/answers/what-is-a-kpi-and-how-is-it-different-from-a-regular-metric ### What is a data warehouse and why do companies use one? A data warehouse is a central database designed for reporting and analysis, where data from different systems such as accounting, CRM and e-commerce is stored together in a consistent structure. Companies use one so everyone reports from the same cleaned, combined data rather than from scattered exports. Operational systems are built to run transactions, not to answer questions across the whole business. A warehouse copies data from those systems, standardises it and keeps history, so you can compare sales with ad spend or stock with demand. Modern cloud warehouses charge mainly for storage and queries used, so a small company can often run one at modest cost. The harder work is modelling the data correctly, not the technology itself. URL: https://zansa.co.za/answers/what-is-a-data-warehouse-and-why-do-companies-use-one ### What is the difference between ETL and ELT? ETL means extract, transform, load: data is cleaned and reshaped before it is placed in the warehouse. ELT means extract, load, transform: raw data is loaded first and transformed inside the warehouse. ELT is now more common with cloud warehouses because storage and processing are cheap and flexible. Both move data from source systems to a place where it can be analysed. Why ELT has become popular: - Raw data is preserved, so you can rebuild logic later without re-extracting - Transformations can be written in SQL and version controlled - Cloud warehouses handle heavy processing well ETL still makes sense when you must filter or mask sensitive fields before data reaches the warehouse, for example for privacy reasons. URL: https://zansa.co.za/answers/what-is-the-difference-between-etl-and-elt ### What is business forecasting and how does it differ from planning? Forecasting estimates what is likely to happen, such as future sales or cash flow, based on historical patterns and known drivers. Planning sets what the business intends to achieve and how. A forecast is a prediction; a plan or budget is a commitment, and comparing the two shows where reality is drifting. Forecasts can be simple, such as a moving average or seasonal trend, or more advanced, using statistical or machine learning models. Complexity does not guarantee accuracy. Good practice includes: - Stating the forecast horizon and the margin of uncertainty - Tracking forecast error over time - Updating as new data arrives A simple, well-understood forecast that people trust usually beats a sophisticated one nobody can explain. URL: https://zansa.co.za/answers/what-is-business-forecasting-and-how-does-it-differ-from-planning ### What is machine learning in plain English for a business owner? Machine learning is a way of getting software to find patterns in historical data and use them to make predictions or decisions, instead of following hand-written rules. Examples include predicting which customers may leave, estimating demand for a product, or flagging unusual transactions. A model is trained on past examples where the outcome is known, then applied to new cases. It is not magic and it is not always right; it gives probabilities that need to be checked against reality. Machine learning needs enough clean, relevant history to learn from. For many small businesses, good reporting and simple statistics deliver most of the value, and machine learning is added later for specific, well-defined problems. URL: https://zansa.co.za/answers/what-is-machine-learning-in-plain-english-for-a-business-owner ### What are descriptive, diagnostic, predictive and prescriptive analytics? They are four levels of analytics. Descriptive shows what happened, diagnostic explains why, predictive estimates what will likely happen, and prescriptive recommends what action to take. Most businesses should master descriptive and diagnostic work before investing in predictive or prescriptive approaches. Think of them as a ladder of difficulty and value: - Descriptive: sales fell 8 percent last month - Diagnostic: the fall came from one region after a stock-out - Predictive: that region will likely miss target next month without restocking - Prescriptive: reorder these items by this date Each level depends on the one beneath it. Predictions built on inconsistent historical numbers rarely survive contact with reality. URL: https://zansa.co.za/answers/what-are-descriptive-diagnostic-predictive-and-prescriptive-analytics ### What is the difference between a data lake and a data warehouse? A data lake stores large volumes of raw data in many formats, structured or not, with little upfront organisation. A data warehouse stores cleaned, structured data modelled for reporting. Warehouses suit business reporting; lakes suit raw storage and data science, and most small companies only need a warehouse. A lake offers flexibility but can become a swamp of undocumented files if poorly managed. A warehouse enforces structure, which makes analysis faster and more consistent. Modern cloud platforms blur the line, and many setups keep a raw layer and a modelled layer in the same warehouse. For a small or mid-sized company, that single-platform approach is usually simpler and cheaper than running a separate lake. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-data-lake-and-a-data-warehouse ### What is a data model and why does it matter for reporting? A data model is the agreed structure of your business data: which tables exist, how they connect, and how things like customer, order or revenue are defined. It matters because consistent definitions let every report calculate numbers the same way, so results match across the business. Without a model, each report author interprets the raw data differently, which is a common cause of conflicting numbers. A typical model separates: - Fact tables, which record events such as sales or shipments - Dimension tables, which describe things such as products, customers and dates A well-designed model is also documented, so a new analyst can understand it without reverse engineering. URL: https://zansa.co.za/answers/what-is-a-data-model-and-why-does-it-matter-for-reporting ### What is a data pipeline and what can go wrong with one? A data pipeline is an automated process that moves data from source systems into a place where it can be analysed, often cleaning and reshaping it on the way. Common problems include broken connections, changed source formats, duplicated records and silent failures that leave dashboards showing stale numbers. Pipelines run on schedules or triggers and need monitoring, like any other system. Reliable pipelines usually have: - Alerts when a job fails or data stops arriving - Automated checks, such as row counts and null tests - Logging so problems can be traced - Documentation of what each step does The most dangerous failure is the quiet one, where everything appears to work but the numbers are wrong. URL: https://zansa.co.za/answers/what-is-a-data-pipeline-and-what-can-go-wrong-with-one ### What is the difference between a data analyst, data scientist and data engineer? A data analyst answers business questions with reports and dashboards. A data engineer builds and maintains the pipelines and warehouses that deliver reliable data. A data scientist builds predictive models. Small companies often need analyst and engineering skills first, and may rely on one person or an outside team for both. Roles overlap, especially in smaller organisations. - Analyst: SQL, BI tools, business understanding, communication - Engineer: pipelines, databases, cloud infrastructure, data quality - Scientist: statistics, machine learning, experimentation Hiring a data scientist before the engineering and analytics foundations exist is a common and expensive mistake, because they end up spending most of their time cleaning data instead of modelling. Knowing which role you actually need avoids paying for skills you cannot yet use. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-data-analyst-data-scientist-and-data-engineer ### When does a small or mid-sized business actually need data analytics? A business needs analytics when decisions depend on numbers that are slow to produce, disputed, or spread across several systems. Typical triggers are growth beyond what one person can track in spreadsheets, multiple sales channels, leadership asking questions nobody can answer quickly, or recurring surprises in margin or cash flow. Company size matters less than complexity. A ten-person e-commerce shop with several marketplaces may need it sooner than a fifty-person firm with one simple service line. Good signs the time has come: - Monthly reporting takes days of manual work - Departments quote different figures for the same measure - You cannot see profit by product, customer or channel - Decisions are made on instinct because the data is hard to reach URL: https://zansa.co.za/answers/when-does-a-small-or-mid-sized-business-actually-need-data-analytics ### What are the signs that my company has outgrown spreadsheets? Warning signs include reports that take days to compile, several versions of the same file, formulas only one person understands, numbers that differ between departments, files that crash or run slowly, and decisions delayed because nobody trusts the figures. When these appear, spreadsheets are creating risk and wasted time. Spreadsheets remain excellent for modelling and quick analysis. The trouble starts when they become the system of record. Further signs: - Manual copy and paste from other systems each week - No history of who changed what - Fear of editing a critical file - Data entry errors that are found late Moving to a warehouse and dashboards does not mean abandoning spreadsheets; it means they stop carrying the whole load. URL: https://zansa.co.za/answers/what-are-the-signs-that-my-company-has-outgrown-spreadsheets ### Spreadsheets versus BI tools: which should a growing business use? Use spreadsheets for small datasets, one-off analysis and modelling, and move to a BI tool when data comes from multiple systems, needs regular refreshing, or must be shared with many people. BI tools add automation, consistent definitions, access control and interactive dashboards that spreadsheets handle poorly at scale. The comparison is not either/or. Many companies keep spreadsheets for planning and what-if analysis while BI tools handle recurring reporting. BI tools generally win on: - Automatic refresh from source systems - Handling millions of rows - Permissions by user or team - A single shared definition of each metric Spreadsheets win on flexibility and familiarity. The best setup uses each for what it does well. URL: https://zansa.co.za/answers/spreadsheets-versus-bi-tools-which-should-a-growing-business-use ### Does a small business really need a data warehouse? Not always. If you have one or two systems and simple reporting, a BI tool connected directly to the source may be enough. A warehouse becomes worthwhile when you combine several systems, need history, face slow or inconsistent reports, or want a reliable foundation for forecasting and future growth. Cloud warehouses have made this far more affordable than a decade ago, since you generally pay for storage and the queries you run rather than for hardware. Consider a warehouse when: - Data lives in three or more systems - Source systems do not keep history you need - Reports are slow or put load on operational software - You want one agreed set of metric definitions If none of these apply, start simpler and revisit later. URL: https://zansa.co.za/answers/does-a-small-business-really-need-a-data-warehouse ### Does my business need a data scientist or just a data analyst? Most small and mid-sized businesses need an analyst, and the engineering to feed them, long before a data scientist. A data scientist adds value when you have a specific predictive problem, enough clean historical data, and a way to act on the predictions. Without these, the role tends to be underused. Ask whether your questions are about what happened and why, or about predicting the future at scale. The first is analyst territory. A data scientist makes sense for problems such as demand forecasting across many products, churn prediction or pricing optimisation, provided that: - The data is trustworthy and has enough history - Someone owns acting on the output - The benefit plausibly exceeds the cost Many firms use an outside specialist for such projects rather than a permanent hire. URL: https://zansa.co.za/answers/does-my-business-need-a-data-scientist-or-just-a-data-analyst ### When is machine learning worth it for a small or mid-sized company? Machine learning is worth it when you have a repeated, well-defined prediction problem, several months or years of relevant clean data, and a process that can act on the output. Demand forecasting, churn prediction and anomaly detection are common fits. If basic reporting is still unreliable, fix that first. Questions to test readiness: - Is the outcome measurable and recorded consistently? - Is there enough history for the model to learn from? - Would a simpler rule or trend line do almost as well? - Who will use the prediction, and what will they do differently? Simple baselines should always be tried first. If a model barely beats them, the added complexity and maintenance are rarely justified. URL: https://zansa.co.za/answers/when-is-machine-learning-worth-it-for-a-small-or-mid-sized-company ### Should we build analytics in-house, buy software, or outsource it? It depends on scale, skills and budget. Buy off-the-shelf tools for standard needs, build in-house when analytics is core to your product or you have steady demand for a team, and outsource when you need expertise quickly or lack enough work for a full-time hire. Many companies blend all three. Consider each route honestly: - Buy: fast and cheap to start, but limited to what the vendor supports and may not fit your data - Build: maximum control and fit, but recruiting, management and key-person risk - Outsource: speed and breadth of skills, but you must manage handover and ownership A common pattern is outsourcing the foundation build, then using an internal analyst or part-time support to run it day to day. URL: https://zansa.co.za/answers/should-we-build-analytics-in-house-buy-software-or-outsource-it ### Should I hire a data analyst, use an agency, or use a freelancer? Hire an analyst when you have steady, ongoing work and can manage them. Use a freelancer for a small, well-defined task. Use an agency or consultancy when you need several skills at once, such as engineering, modelling and dashboards, and want continuity if one person leaves. Trade-offs in general terms: - Employee: deep company knowledge, but one person rarely covers engineering, analysis and modelling, and leaving creates a gap - Freelancer: flexible and often cost-effective, but availability and documentation vary - Agency: broader skills and continuity, usually at a higher day rate Whichever you choose, insist that you own the code, data and documentation so you are never locked in. URL: https://zansa.co.za/answers/should-i-hire-a-data-analyst-use-an-agency-or-use-a-freelancer ### How does Zansa compare with hiring a data analyst or freelancer? Zansa is a data analytics and data science team for small and mid-sized companies, so one engagement brings engineering, modelling and dashboard skills together rather than a single person. The client owns everything built, which reduces key-person risk, and work is delivered at an agreed fixed fee. A single hire or freelancer can be an excellent choice for steady, narrow work. A team is more useful when you need a full foundation built quickly: data connections, a warehouse, data models and dashboards. Zansa is based in South Africa and works remotely with clients worldwide. Its process is a free intro call, a fixed-fee Data Health Check, a Foundation Build with fixed price and timeline agreed in writing, and optional ongoing support. Whether that suits you depends on your volume of work and in-house capacity. URL: https://zansa.co.za/answers/how-does-zansa-compare-with-hiring-a-data-analyst-or-freelancer ### How much does it cost to hire an in-house data analyst? Cost varies widely by country and seniority, but you should budget for salary, employer taxes and benefits, software, training and management time, which together often add a substantial amount on top of base pay. A single analyst also rarely covers data engineering, so factor in extra support. Beyond salary, consider: - Recruitment time, which can take months for good candidates - Software licences and cloud costs - Onboarding before the person is productive - Replacement cost if they leave, taking knowledge with them For many smaller companies, a fixed-scope outsourced build plus part-time support costs less than a full-time hire in the first year, while a hire can win once work is continuous. Compare total cost, not salary alone. URL: https://zansa.co.za/answers/how-much-does-it-cost-to-hire-an-in-house-data-analyst ### How much does a business dashboard project typically cost? Costs depend on the number of data sources, data quality, number of dashboards and who builds it. As a general guide, a simple single-source dashboard may cost a few thousand US dollars, while a multi-source warehouse with modelled data and several dashboards can run into the tens of thousands. The biggest cost driver is usually not the dashboard itself but the data work behind it: connecting systems, cleaning, reconciling definitions and building a reliable model. Other factors: - Number and complexity of source systems - How messy the existing data is - Software licences, which differ widely between tools - Training and documentation Get a written scope and fixed price where possible, and ask what ongoing costs follow launch. URL: https://zansa.co.za/answers/how-much-does-a-business-dashboard-project-typically-cost ### What does Zansa charge for data analytics work? Zansa gives guidance pricing in South African rand: Data Health Checks start from R25,000, Foundation Builds are typically R80,000 to R250,000, and ongoing support starts from R12,000 per month. Final prices are fixed and agreed in writing after scoping, because they depend on systems and complexity. The stages are: - Data Health Check: a fixed-fee audit taking one to two weeks - Foundation Build: typically four to eight weeks, with fixed price and timeline in writing - Optional support: monthly, after the build These figures are guidance rather than a quote. The number of data sources, data quality and dashboard scope move the price. A free intro call is the first step, so you can get a realistic range before committing to anything. URL: https://zansa.co.za/answers/what-does-zansa-charge-for-data-analytics-work ### How do you measure the ROI of data analytics? Compare the measurable benefits of analytics, such as time saved, revenue gained, costs avoided or losses prevented, against the total cost of building and running it. Set a baseline before starting, attribute benefits to specific decisions or processes, and be conservative about what you credit. Returns often show up as: - Hours of manual reporting removed - Reduced stock-outs or overstock - Better pricing or marketing spend allocation - Earlier detection of problems Some benefits, like confidence in decisions, are real but hard to quantify. It helps to define two or three target outcomes at the start, such as reporting time cut or margin improved, and review them after a set period. Avoid claims of precise ROI that cannot be traced. URL: https://zansa.co.za/answers/how-do-you-measure-the-roi-of-data-analytics ### What is data maturity and why should a business assess it? Data maturity describes how well an organisation collects, manages and uses data to make decisions. Assessing it shows where you are, what to fix first and what to avoid, so you invest in the right next step instead of buying advanced tools your foundations cannot support. A maturity assessment looks at several areas: - Data availability and quality - Tools and infrastructure - Skills and ownership - Governance and security - How decisions are actually made It need not be elaborate. A simple honest review of where numbers come from, who trusts them and what is manual is often enough to set priorities and build a sensible roadmap. URL: https://zansa.co.za/answers/what-is-data-maturity-and-why-should-a-business-assess-it ### What are the typical stages of data maturity for a growing company? Models vary, but a common progression is: ad hoc spreadsheets, then basic standardised reporting, then a central warehouse with trusted metrics, then self-service analytics, and finally predictive and automated decision-making. Most small and mid-sized companies sit between the first and third stages. A rough description of each: - Ad hoc: data in individual files, answers depend on who you ask - Reporting: regular reports, still manual, definitions vary - Integrated: data centralised, shared definitions, automated refresh - Self-service: staff explore data themselves within governed limits - Predictive: models inform planning and operations You do not need to reach the top. Aim for the stage that matches your decisions and resources. URL: https://zansa.co.za/answers/what-are-the-typical-stages-of-data-maturity-for-a-growing-company ### What does single source of truth mean in analytics? A single source of truth means one agreed, trusted place and definition for each key business number, so everyone reports from the same data. When finance, sales and operations all pull revenue from the same governed source, meetings stop being arguments about whose spreadsheet is right. It is less about one physical system and more about agreement and discipline: - Data from sources is brought together in one place, often a warehouse - Each metric has one documented definition - Dashboards read from the shared layer, not from private copies It rarely happens overnight. A pragmatic path is to start with the five or ten metrics that cause the most disagreement and fix those first. URL: https://zansa.co.za/answers/what-does-single-source-of-truth-mean-in-analytics ### What are the basics of data governance for a small company? Data governance means clear rules for who owns data, who can access it, how it is defined and how quality and privacy are protected. For a small company, start light: name an owner for each key dataset, document metric definitions, control access and record where personal data lives. Governance sounds heavy but can be proportionate. A practical starting set: - A named owner for each key dataset or metric - A shared glossary of definitions - Role-based access, with least privilege - A list of where personal data is stored and why - A simple process for reporting and fixing data errors Write it down briefly and review it yearly. Policies nobody reads do not count as governance. URL: https://zansa.co.za/answers/what-are-the-basics-of-data-governance-for-a-small-company ### What is data quality and how can a business improve it? Data quality is how accurate, complete, consistent, timely and unique your data is. Improve it by fixing problems at the source, validating entries, removing duplicates, standardising formats and adding automated checks that flag anomalies before they reach reports. Perfect data is rarely required, only data fit for the decision. Cleaning data repeatedly downstream is costly. Better to prevent errors: - Use required fields and dropdowns in entry systems - Agree naming and coding standards - Monitor for missing values, duplicates and sudden volume changes - Assign someone to resolve issues Perfect data is not the goal. Aim for data that is good enough for the decisions it supports, and be transparent about known limitations. URL: https://zansa.co.za/answers/what-is-data-quality-and-how-can-a-business-improve-it ### What is a data dictionary and do small businesses need one? A data dictionary is a document listing each field or metric in your data with its meaning, source, format and calculation. Small businesses benefit because it prevents misunderstandings, speeds up onboarding and keeps definitions such as active customer or net revenue consistent over time. It can be as simple as a shared sheet with columns for name, definition, source system, owner and notes. Useful entries to start with: - Revenue, margin and cost measures - Customer and order status definitions - Date conventions, such as order date versus ship date - Known exceptions or data quirks Keep it near the work, update it when definitions change, and make it readable by non-technical staff. URL: https://zansa.co.za/answers/what-is-a-data-dictionary-and-do-small-businesses-need-one ### How should a small business start with data analytics? Start with the decisions that matter most, list the questions you cannot currently answer, and identify which systems hold the relevant data. Then audit data quality, agree definitions for a few key metrics, and build one useful dashboard end to end before expanding. A sensible sequence: - Pick two or three business questions with clear value - Inventory your data sources and who owns them - Check the data for gaps and mismatches - Define the metrics precisely - Build a small first version and get feedback - Expand only once people actually use it Resist the urge to connect everything at once. Early wins build trust and show what is worth doing next. URL: https://zansa.co.za/answers/how-should-a-small-business-start-with-data-analytics ### How does Zansa help a company that is just starting with data? Zansa starts with a free intro call, then a fixed-fee Data Health Check lasting one to two weeks. It audits your systems, spreadsheets and numbers that do not match, so you get a clear picture and priorities before committing to a larger build. The Health Check suits companies unsure where to begin or who do not trust their current reports. It identifies what data exists, where it conflicts and what is worth fixing first. If it makes sense to continue, the next step is a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Dashboards are usually built in Metabase or Power BI, and the data sits in a cloud account owned by the client. You are not obliged to proceed after the Health Check. URL: https://zansa.co.za/answers/how-does-zansa-help-a-company-that-is-just-starting-with-data ### Which KPIs should a small business track first? Start with a few KPIs that reflect revenue, profitability, cash and customers: for example revenue, gross margin, cash position, customer acquisition and retention. Then add one or two operational measures specific to your model, such as stock turnover, on-time delivery or billable utilisation. Choose KPIs that are: - Linked to a goal you actually care about - Measurable from data you trust - Owned by a person who can influence them - Reviewed on a regular rhythm Five to ten well-chosen KPIs usually beat fifty. Ask of each one: if this number moved sharply, what would we do? If the answer is nothing, drop it. URL: https://zansa.co.za/answers/which-kpis-should-a-small-business-track-first ### What are the most common data analytics mistakes small businesses make? Common mistakes include starting with tools instead of questions, building too many dashboards, ignoring data quality, leaving metric definitions vague, hiring a data scientist too early, giving no one ownership, and relying on a supplier who keeps the code and data. Each leads to wasted spend and low trust. Other frequent errors: - Trying to connect every system at once - Skipping user feedback during design - Forgetting ongoing maintenance and monitoring - Not documenting what was built - Treating analytics as a one-off project Most are avoidable by starting small, agreeing definitions early, checking data first and planning who will look after the system after launch. A short written plan and a named owner for each item usually prevents most of these problems. URL: https://zansa.co.za/answers/what-are-the-most-common-data-analytics-mistakes-small-businesses-make ### How do I choose a data analytics consultancy? Look for clear scoping, a fixed or transparent pricing model, relevant experience in your kind of business, written deliverables, and a promise that you will own the code, data and documentation. Ask how they handle data quality, privacy and handover, and speak to people who have worked with them. A sensible approach: - Describe the problem and see whether they ask good questions before proposing tools - Request a small paid first stage, such as an audit, before a large commitment - Check how they explain trade-offs in plain language - Confirm who actually does the work - Ask for examples of documentation Beware of anyone who recommends a solution before understanding your data. URL: https://zansa.co.za/answers/how-do-i-choose-a-data-analytics-consultancy ### What are the red flags when hiring a data consultancy? Red flags include promising results before seeing your data, vague scopes and open-ended billing, refusing to hand over code or documentation, tying you to proprietary tools, hiding who does the work, blaming your data without evidence, and pushing machine learning when you lack reliable reporting. Further warning signs: - Guaranteed return on investment figures - Heavy jargon and no plain explanation - Storing your data in the supplier's own account - No plan for handover or ongoing support - Pressure to sign quickly - No discussion of privacy or security A good consultancy is comfortable being asked hard questions and puts the answers in writing. URL: https://zansa.co.za/answers/what-are-the-red-flags-when-hiring-a-data-consultancy ### What questions should I ask a data consultancy before signing? Ask who will do the work, how scope and price are fixed, who owns the code, data and dashboards, where data will be stored, how privacy is handled, what happens when the engagement ends, and how they test accuracy. Request answers in writing within the proposal. Useful questions include: - What exactly will I have at the end, and when? - What do you need from my team? - Which tools will you use and what are their ongoing costs? - How do you check the numbers are right? - Can another team maintain this without you? - How do you treat personal data and use of AI tools? Clear, specific answers are a good sign. Vague ones suggest the same vagueness in delivery. URL: https://zansa.co.za/answers/what-questions-should-i-ask-a-data-consultancy-before-signing ### How do I measure whether a dashboard is actually working? Measure usage, trust and impact. Check how often and by whom it is opened, whether people still keep side spreadsheets, whether numbers match across teams, and whether decisions or processes have changed because of it. A dashboard nobody consults is not working, however polished. Practical indicators: - View counts and active users over time - Time saved on recurring reporting - Fewer questions or disputes about the figures - Specific decisions that cited the dashboard - Feedback from users at regular reviews Most BI tools provide usage statistics. Combine them with a short conversation every few months, then retire or rework what is unused. URL: https://zansa.co.za/answers/how-do-i-measure-whether-a-dashboard-is-actually-working ### Why do so many business dashboards end up unused? Dashboards go unused when they are built without a clear user and decision in mind, show numbers people do not trust, are slow or cluttered, lack ownership, or are never updated as the business changes. Technical quality alone does not create adoption. Common causes: - Built from available data rather than real questions - Metrics defined differently from how teams think about them - No training or habit, such as using them in weekly meetings - Errors that were never fixed, eroding trust - No owner to maintain them Involve the actual users from the start, review the dashboard in a recurring meeting, and fix reported errors quickly. URL: https://zansa.co.za/answers/why-do-so-many-business-dashboards-end-up-unused ### How many KPIs or charts should one dashboard have? Fewer than most people expect. A focused dashboard typically shows around five to ten key figures or visuals that fit on one screen, aimed at one audience and purpose. If users must scroll through dozens of charts, split it into separate views or add drill-down pages. There is no fixed rule, but clutter dilutes attention. Helpful principles: - Put the most important figures at the top - Show comparison to target or prior period - Use simple chart types people already understand - Make filters obvious - Move detail to secondary pages Test it by asking a user to find an answer. If it takes more than a few seconds, simplify. URL: https://zansa.co.za/answers/how-many-kpis-or-charts-should-one-dashboard-have ### How do Metabase, Power BI, Tableau and Looker Studio compare for small companies? Power BI suits Microsoft-centred teams and is relatively low cost per user. Tableau is powerful for visual analysis but costlier. Looker Studio is free and simple for Google-based data. Metabase is open source with an easy interface and can avoid per-user fees when self-hosted. The right choice depends on your stack and users. Considerations beyond features: - Licence model: per user, per capacity or free - Fit with your existing tools and data sources - Skills your team already has - Governance and permission needs - Hosting and maintenance effort Pricing and features change often, so check current terms with each vendor. Tool choice matters less than clean, well-modelled data underneath. Run a short trial with your own data before committing, since demos rarely reflect real complexity. URL: https://zansa.co.za/answers/how-do-metabase-power-bi-tableau-and-looker-studio-compare-for-small-companies ### Why does per-user BI licensing matter as a company grows? Per-user licensing means costs rise with every person who needs access, which can discourage sharing dashboards widely. For a growing company, tools without per-user fees or with flat pricing can make it affordable to give everyone data access instead of rationing seats. Think about how many people truly need to see numbers. Often it is more than the analysts: managers, sales staff, warehouse leads. When comparing tools, model the cost at your likely headcount in two or three years, including viewer versus creator licences. Also consider hosting and maintenance costs for any free or open-source option. Zansa builds mainly in Metabase, which has no per-user licence fees, or in Power BI where the client already uses Microsoft. URL: https://zansa.co.za/answers/why-does-per-user-bi-licensing-matter-as-a-company-grows ### What is self-service BI and is it right for a small company? Self-service BI lets non-technical staff explore data and build their own reports without waiting for an analyst. It suits small companies if the underlying data is clean, modelled and well documented. Without that, self-service tends to produce conflicting numbers and confusion. Governed definitions are the prerequisite. Self-service works best when: - A governed data model exists with agreed definitions - Users get brief training - Sensitive data is protected by permissions - An analyst reviews widely used reports A common middle path is curated dashboards for most users plus self-service exploration for a few trained power users. Without these safeguards, self-service often increases the number of conflicting reports. Start with a small pilot group and widen access as trust grows. URL: https://zansa.co.za/answers/what-is-self-service-bi-and-is-it-right-for-a-small-company ### Who should own the data and dashboards built by a consultancy? You should. The code, data models, dashboards, documentation and the cloud account holding your data should belong to your business, so you can change supplier, bring work in-house or continue without permission. Confirm this in the contract before work begins. Ownership problems arise when a supplier hosts data in its own account, uses proprietary tooling, or keeps the logic hidden. Check that the agreement covers: - Ownership of all code and models - Access credentials in your name - Documentation sufficient for another team - What happens at termination This is a basic protection, not an unreasonable demand, and reputable suppliers will agree readily. URL: https://zansa.co.za/answers/who-should-own-the-data-and-dashboards-built-by-a-consultancy ### How does Zansa handle ownership of data, code and dashboards? With Zansa, the client owns everything built: code, data models, dashboards and documentation. Data lives in a cloud account owned by the client, usually Google BigQuery, so there is no lock-in and the client can continue with anyone, including in-house staff. Because the account belongs to the client from day one, Zansa works inside it rather than copying data elsewhere. This also simplifies security and privacy oversight, since the client controls access. Documentation is part of the Foundation Build so that another analyst or team can understand and maintain the work. Optional support is available afterwards, but it is not required to keep using what was delivered. URL: https://zansa.co.za/answers/how-does-zansa-handle-ownership-of-data-code-and-dashboards ### What is vendor lock-in in analytics and how can I avoid it? Vendor lock-in is when switching supplier or tool becomes very costly because your data, logic or dashboards are tied to their proprietary system. Avoid it by owning your cloud account, using open or widely supported tools, keeping code in your repositories and requiring documentation. Practical steps: - Hold the data in an account under your name - Prefer standard languages such as SQL - Keep transformation logic in version-controlled code - Avoid black-box tools without export options - Ask for an exit plan in the contract Some convenience tools trade flexibility for speed. That can be fine, provided you know the cost of leaving before you commit. URL: https://zansa.co.za/answers/what-is-vendor-lock-in-in-analytics-and-how-can-i-avoid-it ### What privacy laws affect business analytics, such as GDPR and POPIA? Laws such as the EU and UK GDPR and South Africa's POPIA regulate how personal information is collected, used, stored and shared. Analytics that touches customer or employee data must have a lawful basis, limit data to what is needed, secure it and respect individuals' rights. Check with a qualified adviser for your situation. Common principles across such laws: - Collect only what is necessary for a stated purpose - Keep data secure and limit access - Retain it no longer than needed - Be transparent with people - Control cross-border transfers Many analytics needs can be met with aggregated or pseudonymised data, which reduces exposure. This is general information, not legal advice. Rules differ by country and by industry, so confirm the specific obligations that apply where your customers and data are located. URL: https://zansa.co.za/answers/what-privacy-laws-affect-business-analytics-such-as-gdpr-and-popia ### How does Zansa design analytics work around POPIA and GDPR? Zansa designs its work for privacy law including POPIA and GDPR. Data stays in the client's own cloud account, access is controlled by the client, and analytics are designed to avoid unnecessary personal data. Clients should still take their own legal advice on compliance. Practical measures include keeping personal information out of places it does not need to be, limiting who can see identifiable fields, and documenting where such data sits. Because the data environment belongs to the client, they retain control and visibility. Privacy-by-design is easier at the start of a build than retrofitted later, which is one reason the Data Health Check looks at what personal data exists in your systems and spreadsheets. URL: https://zansa.co.za/answers/how-does-zansa-design-analytics-work-around-popia-and-gdpr ### Is it safe to use AI tools on business data? It can be, if you control what data they receive. Avoid pasting personal or client-identifiable data into public AI tools, check each provider's data-use terms, prefer enterprise agreements where needed, and keep sensitive processing inside your own cloud environment with proper access controls. Key risks are leakage of confidential information, using data for model training without consent, and breaching privacy law. Sensible rules: - Anonymise or aggregate before sharing - Use approved tools only - Record what data goes where - Review AI-generated code and answers before relying on them AI can speed up analytics work considerably, but human review remains essential for accuracy. URL: https://zansa.co.za/answers/is-it-safe-to-use-ai-tools-on-business-data ### How does Zansa use AI tools without exposing client data? Zansa uses AI tools to help write code, but those tools never receive personal or client-identifiable data. Any machine learning models run inside the client's own cloud account, so data does not leave the environment the client controls. This keeps client confidentiality intact. This separates two things: AI as a coding assistant, and models that learn from your data. The assistant works from structure and requirements, not from real customer records. Models trained on your data stay inside your account. This matters for privacy law such as POPIA and GDPR and for client confidentiality. Clients should still ask any supplier, Zansa included, to describe this in writing in the contract. URL: https://zansa.co.za/answers/how-does-zansa-use-ai-tools-without-exposing-client-data ### What are data silos and how do they harm a business? Data silos are pockets of information held in separate systems or teams that do not connect. They cause duplicated work, conflicting figures, incomplete views of customers and slow decisions, because nobody can see the whole picture without manual effort. Joining them up is usually the first priority. Silos arise naturally as different departments adopt their own tools. Signs include sales, finance and operations quoting different numbers, and staff rekeying data between systems. Ways to reduce them: - Centralise data in a warehouse - Agree shared identifiers, such as customer or product codes - Define metrics once - Assign cross-team ownership The goal is not to remove departmental systems but to connect their data. URL: https://zansa.co.za/answers/what-are-data-silos-and-how-do-they-harm-a-business ### Why do the numbers in different reports never match? Numbers differ because of different definitions, time periods, filters, data sources, refresh times, duplicates, manual adjustments or errors. For example, revenue may be counted at order date in one report and invoice date in another. The fix is shared definitions and one governed source. Investigating a mismatch usually means asking: - Which system and extract does each report use? - What date field and filters apply? - Are refunds, tax and cancellations treated the same way? - When was each report refreshed? - Has anyone edited numbers manually? Once causes are found, document the agreed definition and rebuild reports from the same data layer so the problem does not recur. URL: https://zansa.co.za/answers/why-do-the-numbers-in-different-reports-never-match ### What is a data audit or data health check and what does it involve? A data audit reviews what data you have, where it lives, how reliable it is and how well it supports your decisions. It typically covers source systems, spreadsheets, definitions, quality issues, security and privacy, and ends with prioritised recommendations. It is the logical first step before any larger build. A good audit produces something actionable, not just a list of problems: - An inventory of systems and spreadsheets - Findings on mismatched or missing numbers - Risks around privacy and access - Recommended priorities and rough effort It is useful before any big investment because it shows the real state of the data. Zansa offers this as a fixed-fee Data Health Check lasting one to two weeks. URL: https://zansa.co.za/answers/what-is-a-data-audit-or-data-health-check-and-what-does-it-involve ### Is fixed-price or time-and-materials better for a data project? Fixed price suits well-defined work, giving budget certainty and an incentive for the supplier to deliver efficiently. Time and materials suits uncertain or exploratory work. A common compromise is a fixed-fee discovery or audit first, then a fixed-price build based on what it finds. Risks to watch: - Fixed price with a vague scope leads to disputes over changes - Time and materials with no cap can run away Whichever you choose, insist on a written scope, deliverables, timeline and a change process. Zansa follows the staged approach: a fixed-fee Health Check, then a Foundation Build with fixed price and timeline agreed in writing, so costs are known before the main work starts. URL: https://zansa.co.za/answers/is-fixed-price-or-time-and-materials-better-for-a-data-project ### How long does it take to build a business dashboard or data foundation? A simple dashboard from one clean source can take days to a couple of weeks. A proper data foundation with several sources, a warehouse, modelled data and dashboards typically takes weeks to a few months, depending on data quality and decisions needed from the business. Time is driven by: - Number and accessibility of source systems - Messiness of the data - Speed of the business answering definition questions - Number of dashboards and users Delays more often come from waiting on access and decisions than from technical work. Zansa Foundation Builds typically run four to eight weeks, with the timeline agreed in writing. Agree a realistic scope and give the supplier prompt access to systems, since waiting on access is a frequent source of delay. URL: https://zansa.co.za/answers/how-long-does-it-take-to-build-a-business-dashboard-or-data-foundation ### Do I need real-time data or is daily refresh enough? Daily refresh is enough for most management reporting. Real-time data is worth the extra cost and complexity only when decisions must be made within minutes, such as fraud checks, live operations or inventory alerts. Match the refresh rate to how quickly someone can actually act. Ask what would change if the data were fresher. If a manager reviews numbers each morning, hourly updates add cost without value. Typical options: - Daily: most financial and sales reporting - Hourly: operational monitoring - Near real time: live logistics or monitoring Higher frequency means more infrastructure, more failure points and sometimes higher cloud query bills. A useful test is to ask who would act on the data within the hour and what exactly they would do. URL: https://zansa.co.za/answers/do-i-need-real-time-data-or-is-daily-refresh-enough ### How can I combine data from accounting, CRM and e-commerce systems? Extract data from each system using built-in connectors or APIs, load it into a central warehouse, then model it with shared keys such as customer ID, product code and date so the sources can be joined. Reporting tools then read from the combined model. Challenges to expect: - Different identifiers for the same customer or product - Different time zones, currencies and tax treatments - APIs with rate limits or changing formats - Duplicate or incomplete records Plan a mapping between systems early, and decide which system is authoritative for each piece of information. Start with the two or three sources that answer your most valuable question. URL: https://zansa.co.za/answers/how-can-i-combine-data-from-accounting-crm-and-e-commerce-systems ### Why do many companies choose a cloud data warehouse such as BigQuery? Cloud warehouses like Google BigQuery need no hardware, scale automatically and typically charge for storage and queries used, so small companies can start cheaply. They connect to many tools and are managed by the provider, which reduces administration compared with running your own servers. Alternatives include Snowflake, Amazon Redshift and Azure-based options, and each has strengths. Choice often follows your existing cloud and skills. Things to manage: - Query costs, which can grow if unmanaged - Access permissions - Region and data residency Zansa usually uses Google BigQuery, set up in an account owned by the client. Whatever you choose, confirm the data sits in an account you control so you can switch supplier or tool later. URL: https://zansa.co.za/answers/why-do-many-companies-choose-a-cloud-data-warehouse-such-as-bigquery ### Why is documentation important in business analytics and data projects? Documentation records what data exists, how it is transformed and what each metric means, so the system can be understood, trusted and maintained by someone other than its creator. Without it, the business depends on one person's memory and every change is risky. Useful documentation covers: - Source systems and how data is extracted - Data model and table descriptions - Metric definitions and business rules - Refresh schedules and known issues - How to add or change reports It need not be long. Short, accurate and kept up to date beats exhaustive and stale. Treat documentation as a deliverable with an owner, rather than something written if time remains at the end. URL: https://zansa.co.za/answers/why-is-documentation-important-in-business-analytics-and-data-projects ### What is data literacy and how can a team improve it? Data literacy is the ability to read, question and use data sensibly. Teams improve it through short practical training on their own dashboards, clear metric definitions, encouraging questions about where numbers come from, and using data routinely in meetings rather than occasionally. It does not mean everyone must code. Core skills are understanding averages versus totals, spotting misleading charts, knowing the difference between correlation and causation, and recognising data limitations. Tactics that help: - Short walkthroughs when a dashboard launches - A glossary of terms - Leaders who ask for evidence - A named person to answer questions Even brief, regular sessions tied to real company numbers tend to work better than generic classroom training. URL: https://zansa.co.za/answers/what-is-data-literacy-and-how-can-a-team-improve-it ### How do you build a data-driven culture in a small business? Lead by example: ask for evidence in decisions, use the same dashboards in regular meetings, agree definitions, celebrate decisions improved by data, and make data accessible. Culture changes through repeated habits, not through tools or announcements. Culture follows consistent habits. Practical steps: - Start each management meeting with the shared numbers - Make it safe to report bad numbers - Give teams ownership of their metrics - Fix data errors visibly and quickly - Combine data with experience instead of replacing it Data should inform judgement, not override it. Over-reliance on single metrics can cause its own distortions. Be patient: habits typically take months to shift, and repeated use by senior people matters more than any policy statement. URL: https://zansa.co.za/answers/how-do-you-build-a-data-driven-culture-in-a-small-business ### How accurate can business forecasts really be for a small company? Accuracy depends on how stable and predictable the business is, how much history exists and how far ahead you forecast. Short-term forecasts for steady, seasonal products can be quite reliable; long-range forecasts or new products are much less certain. Always express forecasts with a range. Ways to keep forecasting honest: - Compare against a simple baseline, such as last year's figure - Track error over time - Update as conditions change - Account for known events like promotions External shocks cannot be forecast from history, so treat forecasts as decision aids, not guarantees. Reviewing past forecasts against actual results every month is the best way to learn how far to trust them. URL: https://zansa.co.za/answers/how-accurate-can-business-forecasts-really-be-for-a-small-company ### How does demand forecasting help retail and e-commerce businesses? Demand forecasting estimates how much of each product customers will buy, helping retailers set stock levels, plan purchasing and avoid both stock-outs and overstock. Even simple seasonal forecasts can improve ordering decisions and reduce cash tied up in slow stock. Inputs often include sales history, seasonality, promotions, price changes and lead times from suppliers. More advanced methods add external factors. Practical points: - Forecast at a level where data is stable, such as category before individual item - Treat new products separately - Review forecast errors regularly Start with a simple method and measure it before adding complexity. Share forecasts with buyers and operations teams, since their knowledge of supplier delays and promotions often improves the numbers. URL: https://zansa.co.za/answers/how-does-demand-forecasting-help-retail-and-e-commerce-businesses ### What is churn prediction and when is it useful? Churn prediction uses past customer behaviour to estimate who is likely to leave, so you can act before they do. It is useful for subscription or repeat-purchase businesses with enough customer history and the ability to run retention actions. Without a retention action, predictions add little. Before building a model, check that you can define churn clearly, for example no purchase in 90 days, and that you track relevant behaviour. Often a simple analysis of what leaving customers have in common gives most of the insight. A model adds value when you have many customers and need to prioritise limited retention effort. Without a response plan, the prediction has no benefit. URL: https://zansa.co.za/answers/what-is-churn-prediction-and-when-is-it-useful ### What is customer segmentation and how is it used in analytics? Customer segmentation groups customers by shared characteristics or behaviour, such as spend, frequency, product mix or industry, so you can tailor marketing, pricing and service. It shows which groups drive profit and which need attention. Start with simple rules based on value and recency. Segments can be rule-based, like top 20 percent by revenue, or discovered with clustering techniques. To be useful, segments should be: - Understandable to staff - Stable enough to act on - Linked to different actions Start with simple rules based on value and recency. More sophisticated methods help when you have rich behavioural data. Review segments periodically, because customers move between groups as behaviour changes. Name segments in plain business language so staff remember and use them. URL: https://zansa.co.za/answers/what-is-customer-segmentation-and-how-is-it-used-in-analytics ### What is A/B testing and can small companies use it? A/B testing compares two versions of something, such as a web page or email, by showing each to a random group and measuring which performs better. Small companies can use it if they have enough traffic or customers to detect a real difference rather than random noise. Key rules: - Test one change at a time - Decide the success metric beforehand - Run until you have enough observations - Do not stop early on a lucky streak With low volumes, tests may need to run a long time or only detect large effects. In that case, bigger changes or qualitative feedback may be more practical. Record every test and its result so the business builds knowledge rather than repeating earlier experiments. URL: https://zansa.co.za/answers/what-is-a-b-testing-and-can-small-companies-use-it ### What is cohort analysis and what business questions does it answer? Cohort analysis groups customers by a shared starting point, such as the month they first bought, and tracks how each group behaves over time. It answers questions about retention, repeat purchase and whether newer customers are better or worse than older ones. Overall averages can hide changes. A business may appear stable while recent cohorts retain less than earlier ones. Typical uses: - Retention curves by signup month - Revenue per cohort over time - Effect of an onboarding or pricing change It needs customer-level history with consistent identifiers, which is another reason data foundations matter. Keep cohorts large enough to be meaningful, and compare them at the same age, such as the first three months after joining, rather than at the same calendar date. URL: https://zansa.co.za/answers/what-is-cohort-analysis-and-what-business-questions-does-it-answer ### What does data cleaning involve and why does it take so long? Data cleaning means fixing errors, removing duplicates, handling missing values, standardising formats and reconciling inconsistent records so data is reliable. It takes long because real-world data is messy, sources disagree and fixing issues often needs business decisions, not just code. Typical tasks: - Standardising names, dates, currencies and units - Merging duplicate customers - Deciding how to treat blanks and outliers - Reconciling totals against source systems It is often the largest part of analytics work. Fixing issues at the source and automating checks reduces the effort over time. Budget time for it explicitly in any project plan, because underestimating cleaning work is a leading cause of overruns. URL: https://zansa.co.za/answers/what-does-data-cleaning-involve-and-why-does-it-take-so-long ### What are the basic data security practices a small business should follow? Use strong authentication, give people only the access they need, keep data in reputable cloud services, encrypt sensitive information, back up regularly, remove access when staff leave and keep an inventory of where sensitive data lives. These steps cover most common risks. Additional measures: - Avoid emailing spreadsheets that contain personal data - Log and review who accesses sensitive data - Keep software updated - Train staff on phishing - Have a plan for incidents Security is ongoing, not a one-off setup, and proportionate effort matters more than perfection. Write a one-page security checklist, assign an owner, and review it twice a year to keep it current. URL: https://zansa.co.za/answers/what-are-the-basic-data-security-practices-a-small-business-should-follow ### How is generative AI changing business analytics for small companies? Generative AI speeds up writing code and queries, summarising findings and drafting documentation, and lets people ask questions of data in plain language. It does not remove the need for clean data, clear definitions and human checking, since it can produce confident but wrong answers. Useful applications: - Drafting SQL and transformation code for review - Explaining dashboards in plain English - Classifying text such as customer comments Cautions: - Privacy of data sent to external services - Hallucinated or incorrect results - Inconsistent definitions leading to inconsistent answers Good data foundations make AI tools far more reliable. Treat it as an assistant that increases the productivity of skilled people, not a replacement for them. URL: https://zansa.co.za/answers/how-is-generative-ai-changing-business-analytics-for-small-companies ### Can I just ask an AI chatbot questions about my business data? Sometimes, but reliability depends on how well your data is modelled and documented. Chat interfaces over messy or ambiguous data can return plausible but wrong numbers. They work best on a governed data model with clear definitions, and answers should be verifiable. Before relying on one, check: - Does it query a trusted model, not raw tables? - Does it show the query or logic used? - Is sensitive data protected, and where is it processed? - Are results consistent when you ask in different words? Treat it as a convenience for exploration, with official dashboards remaining the reference for key figures. URL: https://zansa.co.za/answers/can-i-just-ask-an-ai-chatbot-questions-about-my-business-data ### How do I prepare my data for AI and machine learning projects? Make sure data is accurate, consistently defined, centralised and has enough history. Document it, clarify the outcome you want to predict, check for bias and gaps, and confirm you are allowed to use it for this purpose. Foundations matter more than the algorithm. Preparation steps: - Define the target outcome precisely - Combine relevant sources into one dataset - Clean and standardise records - Check history covers different conditions, such as seasons - Keep a clear record of how the data was produced Many AI projects fail because of poor data, not poor algorithms. Involving the people who understand how the data was recorded often reveals quirks that no algorithm would detect. URL: https://zansa.co.za/answers/how-do-i-prepare-my-data-for-ai-and-machine-learning-projects ### How can a business trust that its dashboard numbers are correct? Reconcile dashboard totals against source systems, add automated tests such as row counts and null checks, document definitions, review changes before release and have users flag anomalies. Trust is built by verification, transparency and quick fixes when errors appear. Transparency is what builds confidence. Practical controls: - Monthly reconciliation of revenue and cost to the ledger - Alerts when a data load fails or volumes change sharply - Version control of data logic - A visible last-updated time on dashboards An honest note of known limitations helps credibility more than hiding them. Make known limitations visible on the dashboard itself, since users trust a source that admits its gaps more than one that appears flawless. URL: https://zansa.co.za/answers/how-can-a-business-trust-that-its-dashboard-numbers-are-correct ### What is version control and why should analytics code use it? Version control, usually Git, records every change to code with who made it and why, and lets you roll back mistakes. Analytics code, such as SQL transformations, benefits because metric logic changes over time and you need to trace and review those changes. Benefits for analytics: - Reviewing changes before they affect reports - Restoring a previous working version - Collaborating without overwriting each other - Keeping a history of definition changes It also supports the client owning the code, since the repository can be handed over complete with history. Even a team of one benefits, since the history protects against accidental breakages and unexplained changes in reported figures. URL: https://zansa.co.za/answers/what-is-version-control-and-why-should-analytics-code-use-it ### What are the limits of Excel for business analytics? Excel struggles with very large datasets, automatic refresh from many systems, multi-user collaboration, audit trails and version control. Formulas can contain hidden errors, and files get copied and diverge. It remains excellent for modelling and ad hoc analysis, but weak as a shared reporting system. Specific pain points: - Row and performance limits on large files - Manual copy and paste from sources - Errors in complex formulas are hard to spot - Limited access control - Many conflicting copies Power Query and Power Pivot extend Excel considerably, and can be a reasonable stepping stone before a full BI setup. If your spreadsheets are central to operations, treat them like any system: restrict editing, document formulas and keep backups. URL: https://zansa.co.za/answers/what-are-the-limits-of-excel-for-business-analytics ### How can automated reporting save time for a small team? Automated reporting replaces manual compilation of recurring reports with scheduled data refreshes and dashboards or emailed summaries. It saves hours of copying and formatting, reduces errors and frees staff to analyse rather than assemble numbers. The time saved is usually easy to measure. Good candidates for automation are weekly sales reports, month-end packs and stock summaries that follow the same format. Steps: - Identify recurring reports and how long they take - Connect them to a reliable data source - Build once and schedule - Monitor for failures The time saved can be estimated directly, which makes this one of the easier returns to measure. Report owners should still review output periodically, because automation can repeat an error just as efficiently as it repeats a correct result. URL: https://zansa.co.za/answers/how-can-automated-reporting-save-time-for-a-small-team ### What is a data strategy and does a small business need one? A data strategy is a short plan describing how data will support business goals: which decisions matter, what data and tools are needed, who is responsible and what to do in what order. Small businesses need a light version, often one or two pages. A practical strategy answers: - What are our key business goals? - Which decisions need better data? - What is the current state? - What will we do in the next 6 to 12 months? - Who owns it and how will we measure success? Avoid grand plans disconnected from budget. A modest roadmap you actually follow is better. Revisit it at least yearly, because priorities, systems and budgets change and an out-of-date plan quickly loses credibility. URL: https://zansa.co.za/answers/what-is-a-data-strategy-and-does-a-small-business-need-one ### How do I build a sensible data roadmap on a limited budget? Rank possible projects by business value and effort, start with foundational work that unlocks several uses, and deliver in small stages that each produce something useful. Fix data quality and definitions before advanced analytics, and review the plan every few months. A staged example: - Stage 1: audit and agree key metrics - Stage 2: centralise main data sources and build core dashboards - Stage 3: extend to more sources and automate reports - Stage 4: add forecasting or machine learning where justified Each stage should have a clear benefit so you can stop or adjust if priorities change. Be willing to drop items that no longer pay off; a roadmap is a set of options, not a promise to do everything listed. URL: https://zansa.co.za/answers/how-do-i-build-a-sensible-data-roadmap-on-a-limited-budget ### How do I get staff and management to actually use new dashboards? Involve users in design, build around decisions they already make, train them briefly, build the dashboards into regular meetings and have a leader use them visibly. Fix errors promptly and retire dashboards nobody uses. Adoption is earned through relevance and reliability. Adoption tactics: - Ask what they check now and why - Replace an existing report instead of adding another - Show quick wins early - Provide a contact for questions - Review usage after a month Resistance often comes from distrust of the numbers or fear of scrutiny. Address both openly. Ask users after a month what they still do outside the dashboard. The answers show exactly what to improve next. URL: https://zansa.co.za/answers/how-do-i-get-staff-and-management-to-actually-use-new-dashboards ### Who should be responsible for data in a small company? Assign a senior owner accountable for data overall, plus named owners for key datasets and metrics, usually the managers who understand them, such as finance for revenue. Technical staff or a supplier maintain systems, but business owners decide definitions and priorities. Roles to consider: - Executive sponsor who backs the effort - Metric owners in each department - A technical custodian, internal or external - Someone who handles access and privacy requests One person may hold several roles in a small firm. What matters is that responsibilities are named, not left to everyone and therefore no one. Write the roles down and share them, so questions about a number always have an obvious person to ask. URL: https://zansa.co.za/answers/who-should-be-responsible-for-data-in-a-small-company ### What are the risks of outsourcing data analytics? Risks include dependence on the supplier, unclear ownership of code and data, weaker understanding of your business, privacy exposure and poor handover. You can manage them by owning the cloud account, requiring documentation, fixing scope in writing and keeping an internal contact who understands the work. Mitigations: - Hold data and code in your own accounts - Define deliverables and acceptance criteria - Require documented handover - Limit supplier access to what is needed - Include privacy and confidentiality clauses Outsourcing works well when you remain the owner and decision-maker. A good supplier will welcome these safeguards, because they make the relationship clearer and easier to end or extend. URL: https://zansa.co.za/answers/what-are-the-risks-of-outsourcing-data-analytics ### What should a good handover at the end of a data project include? A good handover includes all code and models in your repository, documentation of data sources and metric definitions, access credentials in your name, a walkthrough for your team, a list of known issues and a plan for who maintains and monitors it afterwards. Also request: - A diagram of how data flows - Refresh schedules and failure alerts - Instructions for common changes - Costs of ongoing services Test the handover by asking someone who did not build it to make a small change. If they cannot, the documentation needs work. Agree the handover requirements at the start of the engagement rather than the end, when leverage and goodwill are both lower. URL: https://zansa.co.za/answers/what-should-a-good-handover-at-the-end-of-a-data-project-include ### What does Zansa deliver in a Foundation Build? A Foundation Build, typically four to eight weeks, delivers connected data sources, a cloud data warehouse in the client's account, modelled data with agreed metric definitions, dashboards, usually in Metabase or Power BI, and documentation. Price and timeline are fixed and agreed in writing. The scope depends on what the Data Health Check found. The client owns the code, data models, dashboards and documentation. Zansa designs for privacy law including POPIA and GDPR, and machine learning models, where used, run inside the client's own account. After the build, optional support is available from R12,000 per month, but the client is free to run things themselves or with another provider. URL: https://zansa.co.za/answers/what-does-zansa-deliver-in-a-foundation-build ### How can retail and e-commerce businesses use data analytics? Retailers use analytics to see margin by product and channel, forecast demand, manage stock, understand customer retention, measure marketing return and spot pricing opportunities. Combining sales, inventory, advertising and customer data in one place is usually the key first step. Common questions: - Which products and channels are truly profitable after costs and returns? - What stock is slow moving? - What is customer lifetime value by acquisition source? - How do promotions affect margin? Multiple sales channels often produce conflicting numbers, so consistent definitions for revenue, returns and discounts are essential. Start with the one or two reports your buyers or owners look at most often, and make those fully trustworthy first. URL: https://zansa.co.za/answers/how-can-retail-and-e-commerce-businesses-use-data-analytics ### How can manufacturing companies use data analytics to improve operations? Manufacturers use analytics to track production output, yield, downtime, scrap, costs and on-time delivery, forecast demand, plan materials and spot quality issues early. Integrating production, inventory and sales data gives a clearer picture of true product profitability. Start with one line or one question. Typical measures include overall equipment effectiveness, cost per unit and supplier performance. Challenges: - Data locked in machines or legacy systems - Manual shop-floor records - Differing definitions of downtime or scrap Starting with one line or one question, such as the main source of lost time, is more practical than attempting plant-wide analytics at once. Where machine data is hard to reach, even a manual daily log entered consistently can provide a useful starting dataset. URL: https://zansa.co.za/answers/how-can-manufacturing-companies-use-data-analytics-to-improve-operations ### How can logistics and distribution companies use analytics? Logistics firms use analytics to monitor delivery performance, route and fleet costs, warehouse productivity, capacity and customer service levels. Combining order, tracking, cost and customer data helps identify unprofitable routes or clients and improve on-time delivery. Start with delivery performance. Useful measures include on-time in full, cost per delivery, vehicle utilisation and dwell time. Challenges often include data from multiple carriers and systems, inconsistent location or customer codes and the need for timely data. Operational dashboards with exception alerts tend to deliver quick value, while forecasting and route optimisation can follow once data is reliable. Customer-facing partners often value shared performance reports, which can strengthen contract discussions and service reviews. URL: https://zansa.co.za/answers/how-can-logistics-and-distribution-companies-use-analytics ### What should healthcare providers consider when using data analytics? Healthcare data is highly sensitive, so privacy, consent, security and legal compliance come first. Within those limits, analytics can improve scheduling, utilisation, billing accuracy and operational efficiency, often using aggregated or de-identified data rather than patient-level detail. Aggregated data is often sufficient. Considerations: - Applicable health data laws and professional rules - Strict access control and audit logs - De-identification where possible - Clear separation of operational reporting from clinical decisions Zansa serves healthcare among its sectors and designs for privacy law including POPIA and GDPR, with data kept in the client's own cloud account. Legal advice should still be taken for specific requirements. URL: https://zansa.co.za/answers/what-should-healthcare-providers-consider-when-using-data-analytics ### How can property and hospitality businesses use data analytics? Property businesses use analytics for occupancy, rent collection, maintenance costs and portfolio performance. Hospitality businesses use it for occupancy, average rate, revenue per available room, covers, food cost and guest feedback. Both benefit from combining operational and financial data. Both sectors depend on seasonal patterns. Common challenges are data split across property management, booking, point-of-sale and accounting systems, and seasonal patterns that need careful forecasting. Zansa works with property and hospitality clients, among other sectors, typically starting with a Data Health Check to see which systems and numbers need reconciling before dashboards are built. Guest and tenant feedback can be combined with financial data to show which service issues actually cost money. URL: https://zansa.co.za/answers/how-can-property-and-hospitality-businesses-use-data-analytics ### How can professional services firms use data analytics? Professional services firms use analytics to track utilisation, billable hours, realisation, project profitability, pipeline and client concentration. Linking time-tracking, billing and CRM data shows which clients and services are truly profitable and where capacity is wasted. Time-entry discipline is the main challenge. Key questions: - Are projects delivered within budgeted hours? - Which clients generate the best margin? - What does the pipeline suggest about future capacity? Data quality depends on consistent time entry, so process and culture matter as much as tools. Zansa serves professional services among its sectors. Pipeline data from your CRM, when kept current, helps predict staffing needs weeks or months ahead of demand. URL: https://zansa.co.za/answers/how-can-professional-services-firms-use-data-analytics ### How should I handle personal data in analytics reports? Use aggregated or pseudonymised data wherever possible, restrict identifiable fields to those who need them, document your lawful basis and retention, and avoid sharing personal details in exports or emails. Most reporting questions can be answered without exposing individuals. Most questions need no individual-level detail. Practical techniques: - Replace names with IDs in analysis datasets - Hide small groups that could identify individuals - Mask sensitive fields in dashboards - Delete data once no longer needed Privacy laws such as GDPR and POPIA require appropriate safeguards. This is general guidance, so check your specific obligations with a qualified adviser. Also apply the same care to files saved on laptops or shared drives, which are a common and overlooked place for personal data. URL: https://zansa.co.za/answers/how-should-i-handle-personal-data-in-analytics-reports ### Can a small business do data analytics without a big budget? Yes. Start with free or low-cost tools, focus on a few high-value questions, clean your most important data and build simple dashboards. Cloud services charge by usage, so costs can begin small. Spend where it removes the biggest pain, not on tools you cannot yet use. Low-cost approaches: - Use existing reports from your accounting or sales systems - Spreadsheets with careful structure - Free or open-source BI tools - A short audit to find the highest-value opportunities Budget should also cover time, since staff effort is often the larger cost. Staging the work lets you stop when the marginal value falls. Whatever you spend, tie it to one concrete outcome so you can tell afterwards whether it was worthwhile. URL: https://zansa.co.za/answers/can-a-small-business-do-data-analytics-without-a-big-budget ### How much does cloud data storage and querying cost for a small company? Costs depend on data volume and usage, but for many small and mid-sized companies, cloud warehouse bills can be modest, often far below the cost of staff time. The risk is unmanaged queries, so set budgets, alerts and efficient designs from the start. Cloud warehouses usually charge for stored data and for data processed by queries. Small datasets are often very cheap to store and query. Keep costs predictable by: - Setting spending alerts and quotas - Limiting who can run expensive queries - Refreshing data only as often as needed - Reviewing the bill monthly Check current provider pricing, as it changes. URL: https://zansa.co.za/answers/how-much-does-cloud-data-storage-and-querying-cost-for-a-small-company ### How do I decide which data sources to connect first? Connect the sources that answer your most valuable business question first, usually sales or revenue, then costs, then customer or marketing data. Prefer sources that are accessible, reasonably clean and have clear owners, and expand once the first dashboard is trusted and used. A simple scoring approach: - Value: how much would the answer improve decisions? - Accessibility: is there an API or export? - Quality: is the data reliable? - Effort: how long will it take? Start with high value and moderate effort. Connecting everything at once delays benefits and multiplies reconciliation problems. Revisit the order after each phase, since new questions and newly accessible systems can change what is most valuable next. URL: https://zansa.co.za/answers/how-do-i-decide-which-data-sources-to-connect-first ### What is a good first project to prove the value of analytics? Pick a small project with a clear owner, visible pain and measurable benefit, such as automating a weekly sales or stock report, or building a profitability view by product or customer. Deliver in a few weeks so the value is evident before larger commitments. Good first projects share traits: - A manager who needs the answer and will use it - Data from only a few sources - A before-and-after comparison, for example hours saved - Low risk if imperfect Use the project to establish metric definitions and build trust that later work can build on. Share the result widely and honestly, including what did not work, so later projects start with credibility. URL: https://zansa.co.za/answers/what-is-a-good-first-project-to-prove-the-value-of-analytics ### How do I know if my company is ready for predictive analytics? You are ready when your descriptive reporting is trusted, you have consistent historical data covering enough time, the outcome you want to predict is clearly defined and someone will act on the predictions. If reports still disagree, fix that first. Readiness checklist: - Reliable data from the relevant systems - Sufficient history, ideally covering seasonal cycles - A clear business decision linked to the prediction - Skills to build and maintain the model - A plan to measure whether it works Starting with simple trend or seasonal forecasts is a low-risk way to test appetite. If you are unsure, a short data audit can tell you whether history, quality and ownership are good enough yet. URL: https://zansa.co.za/answers/how-do-i-know-if-my-company-is-ready-for-predictive-analytics ### What does Zansa mean and where does the name come from? Zansa comes from the Japanese word 残差 (zansa), meaning residual. In statistics, the residual is the gap between what a model predicts and what actually happens, which is where the useful insight often sits. Zansa is a data analytics and data science team based in South Africa. The name reflects an approach of looking closely at what does not add up: numbers that do not match, results that differ from expectations, and gaps between reports and reality. Zansa works remotely with small and mid-sized companies anywhere in the world, serving sectors such as retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. The name is also a reminder that the unexplained differences in your numbers are worth investigating, not ignoring. URL: https://zansa.co.za/answers/what-does-zansa-mean-and-where-does-the-name-come-from ### What is the process for working with Zansa from first contact to support? Zansa begins with a free intro call, then a fixed-fee Data Health Check of one to two weeks, then a Foundation Build typically lasting four to eight weeks at a fixed price and timeline agreed in writing, followed by optional ongoing support. Each stage can be a decision point. Guidance pricing in South African rand: Health Checks from R25,000, builds typically R80,000 to R250,000 and support from R12,000 per month. Dashboards are usually built in Metabase, with no per-user licence fees, or Power BI. Data lives in a cloud account owned by the client, usually Google BigQuery, and the client owns the code, data models, dashboards and documentation. Zansa works remotely from South Africa with clients worldwide. URL: https://zansa.co.za/answers/what-is-the-process-for-working-with-zansa-from-first-contact-to-support ## Healthcare ### How do I measure clinic or practice utilisation? Practice utilisation is booked and attended appointment time divided by available clinical time, multiplied by 100. Available time is the hours a practitioner or room is open for patients. It shows how much of your capacity produces patient care and revenue. Formula in words: appointment minutes attended divided by minutes available. It matters because unused slots cost salary, rent and equipment time without income. Commonly misread: - Counting booked time, not attended time, which ignores no-shows. - Dividing by total opening hours rather than bookable hours, hiding blocked admin or leave time. - Treating 100 percent as ideal when some slack is needed for emergencies and overruns. Measure by practitioner, day of week and room, and separate booked utilisation from attended utilisation to see where the gap comes from. URL: https://zansa.co.za/answers/how-do-i-measure-clinic-or-practice-utilisation ### How do I calculate no-show rate for appointments? No-show rate is the number of appointments where the patient did not attend and did not cancel in advance, divided by the number of appointments scheduled, multiplied by 100. It shows how much booked capacity is lost to missed visits. It matters because a missed slot is usually unrecoverable revenue and wastes practitioner time. Commonly misread: - Mixing no-shows with late cancellations, which need different fixes. Record both. - Excluding walk-ins or double-booked slots inconsistently. - Ignoring patterns by appointment type, day, lead time or practitioner. Break the rate down by how far in advance an appointment was booked and by reminder method. Reminders, deposits and easy rescheduling can be tested, but compare against your own baseline before and after rather than external figures. URL: https://zansa.co.za/answers/how-do-i-calculate-no-show-rate-for-appointments ### What does claims paid first time mean for a medical practice? Claims paid first time, or first-pass payment rate, is the number of claims paid in full on first submission divided by claims submitted, multiplied by 100. It shows how clean your billing is before any rework. It matters because every claim not paid first time requires correction, resubmission and waiting, which adds cost and delays cash. Commonly misread: - Counting a claim as paid when it was partly paid or short-paid. - Using claims still outstanding as if they were failed. - Mixing claims across different payers with different rules. Analyse by payer, practitioner, billing code and reason for non-payment. Matching remittance advice to submitted claims is needed, as the payment record shows what the payer actually did with each claim. URL: https://zansa.co.za/answers/what-does-claims-paid-first-time-mean-for-a-medical-practice ### How is rejected claims rate calculated and what should I do with it? Rejected claims rate is the number of claims rejected by the payer divided by the number of claims submitted, multiplied by 100. Use reason codes to find the cause, then correct the process at source, for example in capture, coding or eligibility checks. It matters because rejections delay payment and, if not resubmitted in time, may be lost. Commonly misread: - Treating rejection as the same as denial. Rejected usually means the claim failed before assessment and can be fixed and resubmitted. Check how your payer uses the terms. - Counting a claim each time it is resubmitted. - Looking only at the total rate and not the top reasons. Rank reasons by number and by value, fix the biggest at the point of data entry, and track the resubmission success rate. URL: https://zansa.co.za/answers/how-is-rejected-claims-rate-calculated-and-what-should-i-do-with-it ### How do I calculate debtor days for a healthcare practice? Debtor days is trade receivables divided by credit revenue for the period, multiplied by the number of days in the period. It estimates the average number of days between billing and getting paid. Example: 150,000 owed, 600,000 billed over 90 days gives 150,000 / 600,000 x 90, about 22.5 days. It matters because slow collection ties up working capital. Commonly misread: - Using total revenue that includes cash payments taken at the desk, which lowers the figure. - Mixing medical-aid or insurer debt and patient debt. Their delays and risks differ, so report them separately. - Relying on the average, which hides a small number of very old balances. Read it alongside an aged debtors report, and track it monthly against your own history. URL: https://zansa.co.za/answers/how-do-i-calculate-debtor-days-for-a-healthcare-practice ### What is de-identified reporting in healthcare analytics and why use it? De-identified reporting means producing management reports from patient data with identifying details, such as names, ID numbers and contact details, removed or replaced, so staff can analyse utilisation, claims and revenue without seeing who the patients are. It matters because most management questions, such as no-shows by day or rejected claims by code, do not require knowing the patient. Good practice includes: - Replacing identifiers with irreversible codes or keeping a separate, restricted key. - Watching for small groups where someone could still be identified from combinations of details. - Logging who accesses any identifiable data. Zansa designs healthcare reporting with a de-identified layer and logged access, with POPIA and GDPR in mind. This is not legal advice, so confirm requirements with a qualified adviser. URL: https://zansa.co.za/answers/what-is-de-identified-reporting-in-healthcare-analytics-and-why-use-it ### How does privacy law like POPIA or GDPR affect practice analytics? Health information is treated as special personal information under POPIA and special category data under GDPR, so analytics on it needs a lawful basis, limited access, security and purpose limits. Using de-identified data for management reporting reduces exposure. Practical points for practices: - Collect and use only what the analysis needs. - Restrict who can see identifiable records and log access. - Keep data in a controlled cloud account with appropriate security. - Be careful with third-party tools and data transfers across borders. Privacy requirements depend on your jurisdiction and circumstances, so seek advice from a qualified legal or data protection professional. Zansa designs its work for privacy law, including POPIA and GDPR, but this does not replace legal advice for your practice. URL: https://zansa.co.za/answers/how-does-privacy-law-like-popia-or-gdpr-affect-practice-analytics ### How do I calculate revenue per consultation and revenue per practitioner? Revenue per consultation is total billed (or collected) revenue divided by the number of consultations in the period. Revenue per practitioner divides revenue by the number of practitioners, or better, by their clinical hours, so part-time staff compare fairly. It matters because it shows the income each unit of clinical time produces. Commonly misread: - Mixing billed and collected revenue. Billed amounts that are rejected or unpaid overstate real income. - Comparing practitioners with different patient mix or procedure types. - Ignoring consultation length; a long, complex visit and a short review are not equivalent. Use revenue per available clinical hour as the fairest comparison, and always state whether the figure is billed or collected. URL: https://zansa.co.za/answers/how-do-i-calculate-revenue-per-consultation-and-revenue-per-practitioner ### What is an aged debtors report and how should a clinic use it? An aged debtors report lists amounts owed to you grouped by how long they have been outstanding, typically current, 30, 60, 90 and over 90 days. It shows not only how much is owed but how old, and therefore how risky, each balance is. It matters because the older an invoice, the harder it usually is to collect. How to use it: - Work the oldest and largest balances first. - Split by payer type, such as insurer, medical scheme and patient. - Check for unallocated payments that make balances look older than they are. Commonly misread: - A low debtor days figure can hide a few very old balances. - Treating all buckets as equally collectable. Review weekly, assign follow-up to a named person, and record write-offs with a reason. URL: https://zansa.co.za/answers/what-is-an-aged-debtors-report-and-how-should-a-clinic-use-it ### How do I measure patient wait time and appointment lead time? Patient wait time is the time from scheduled appointment start to the start of consultation. Appointment lead time is the time between the booking request and the appointment date. Both are calculated as averages and as the proportion exceeding a chosen threshold. It matters because long waits affect patient experience and attendance, and long lead times can drive cancellations and no-shows. Commonly misread: - Averages hide the longest waits. Report the 90th percentile as well. - Recording wait times by hand at the end of the day, which makes them unreliable. - Not splitting by practitioner, day or appointment type. Capture check-in and consultation start times in your practice system, and review them together with utilisation, since overbooking raises utilisation but lengthens waits. URL: https://zansa.co.za/answers/how-do-i-measure-patient-wait-time-and-appointment-lead-time ### How do I measure days to payment for medical claims? Days to payment is the average number of days between claim submission and receipt of payment from the payer. Calculate it per claim using submission and remittance dates, then average by payer, month or claim type. It matters because slow payers affect cash flow and show where follow-up effort should go. Commonly misread: - Using invoice date instead of submission date, which includes your own internal delay. - Leaving unpaid claims out of the average, which understates the delay. - Averaging across payers with very different terms. Report the percentage of claims paid within 30, 60 and 90 days, not only the average. This needs remittances matched to claims, because bank deposits often combine many claims in one payment. URL: https://zansa.co.za/answers/how-do-i-measure-days-to-payment-for-medical-claims ### What does Zansa's healthcare dashboard for clinics and practices actually include? Zansa's healthcare dashboard reconciles billing and remittances and reports utilisation and rejected claims. It is built on a de-identified reporting layer with logged access, designed with privacy law, including POPIA and GDPR, in mind. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process starts with a free intro call and a fixed-fee Data Health Check of one to two weeks, followed by a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Support is optional. Dashboards are usually in Metabase or Power BI, with data in a cloud account owned by the client, usually Google BigQuery. The client owns everything built. Privacy requirements vary by situation, so legal advice remains a separate matter. URL: https://zansa.co.za/answers/what-does-zansa-s-healthcare-dashboard-for-clinics-and-practices-actually-includ ### How can a practice track collection rate on patient-paid accounts? Collection rate is the amount actually collected divided by the amount billed, for a defined group of invoices, multiplied by 100. For patient-paid accounts, track it separately from insurer or medical-scheme payments, as the causes of non-payment differ. Example: 80,000 collected from 100,000 billed gives 80 percent for that group of invoices. It matters because billed revenue is not income until collected. Commonly misread: - Measuring collections against invoices from the same month only, when many payments arrive later. Use invoice cohorts followed over time. - Including write-offs and adjustments inconsistently. - Ignoring the cost of chasing small balances. Track the rate by 30, 60 and 90 days after billing, and by payment method. Collecting payment at the time of the visit often changes results, which you can test and measure against your own history. URL: https://zansa.co.za/answers/how-can-a-practice-track-collection-rate-on-patient-paid-accounts ## Hospitality ### What is hotel occupancy rate and how is it calculated? Hotel occupancy rate is rooms sold divided by rooms available, multiplied by 100. If 70 of 100 available rooms were sold on a night, occupancy is 70 percent. It shows how much of your room inventory is being used. It matters as a basic measure of demand, but it says nothing about price. Commonly misread: - Rooms out of order may be removed from available rooms or counted, depending on policy. Choose one rule and keep it. - Complimentary and house-use rooms may or may not count as sold. - High occupancy achieved through deep discounting can earn less than lower occupancy at higher rates. Always read occupancy together with average daily rate and RevPAR, which combine volume and price. URL: https://zansa.co.za/answers/what-is-hotel-occupancy-rate-and-how-is-it-calculated ### What is ADR (average daily rate) in hospitality and how is it calculated? ADR, average daily rate, is room revenue divided by the number of rooms sold, for a period. It shows the average price paid per occupied room, and excludes unsold rooms from the calculation. Formula in words: total room revenue divided by rooms sold. Example: 21,000 currency units of room revenue from 210 rooms sold gives an ADR of 100. It matters because it shows pricing strength independent of how full the property is. Commonly misread: - Including taxes, breakfast or other charges in room revenue in some periods but not others. - Using it alone: a high ADR with low occupancy may earn less than a lower ADR with higher occupancy. - Comparing across room types or seasons without adjusting for mix. Use it with occupancy and RevPAR. URL: https://zansa.co.za/answers/what-is-adr-average-daily-rate-in-hospitality-and-how-is-it-calculated ### What is RevPAR and how do I calculate it? RevPAR, revenue per available room, is room revenue divided by rooms available, or equivalently occupancy multiplied by ADR. It combines how full you are and how much you charge into a single measure of room revenue performance. Example: 70 percent occupancy and an ADR of 100 gives a RevPAR of 70. It matters because occupancy and ADR can move in opposite directions, and RevPAR shows the net effect. Commonly misread: - It is revenue, not profit. A high RevPAR bought with high commission and labour costs can earn less. - It ignores revenue outside rooms, such as food and beverage. - Using different definitions of available rooms in different periods. Pair it with a profit-based measure such as GOPPAR to see what remains after costs. URL: https://zansa.co.za/answers/what-is-revpar-and-how-do-i-calculate-it ### How do I calculate labour cost percentage in a hotel or restaurant? Labour cost percentage is total labour cost divided by total revenue for the same period, multiplied by 100. Include wages, overtime, employer contributions, benefits and, where used, agency or casual staff, so the figure reflects the full cost of labour. It matters because labour is typically one of the largest controllable costs in hospitality. Commonly misread: - Leaving out management salaries or agency costs, which understates the real figure. - Matching costs to the wrong revenue period, such as payroll cycles that do not match calendar months. - Treating a falling percentage as success when it comes from understaffing that hurts service. Track it daily or weekly against actual covers or occupancy, as well as monthly. Labour hours per occupied room or per cover gives additional, more operational insight. URL: https://zansa.co.za/answers/how-do-i-calculate-labour-cost-percentage-in-a-hotel-or-restaurant ### What is GOPPAR and how does it differ from RevPAR? GOPPAR, gross operating profit per available room, is gross operating profit divided by rooms available in the period. Unlike RevPAR, which stops at room revenue, GOPPAR reflects revenue from all departments minus the costs of running the property. Formula in words: total revenue minus departmental costs and undistributed operating expenses (before fixed charges such as financing and depreciation, depending on the accounting standard), divided by available rooms. It matters because it shows whether revenue growth is actually leaving profit. Commonly misread: - Gross operating profit definitions vary, so check which standard or internal policy applies. - Comparing it between properties with different services such as spas or restaurants. Always document the cost items included, so that results are comparable across properties and over time. URL: https://zansa.co.za/answers/what-is-goppar-and-how-does-it-differ-from-revpar ### How do I measure the true cost of each booking channel? Channel cost is the commissions and fees paid on bookings from that channel, divided by the revenue those bookings produce, giving an acquisition cost percentage. For true cost, add other costs such as payment fees, marketing and staff time, then compare with net revenue. Formula in words: (commission + fees + channel-related costs) divided by booking revenue from the channel. It matters because the headline room rate on each channel can look the same while the amount you keep differs. Commonly misread: - Comparing gross revenue by channel instead of net revenue after commission. - Overlooking cancellations and no-shows, which differ by channel. - Ignoring repeat guests that a channel introduced who later book direct. Report net ADR by channel, plus cancellation rates and length of stay, to see which channels are profitable. URL: https://zansa.co.za/answers/how-do-i-measure-the-true-cost-of-each-booking-channel ### How do I forecast covers for a restaurant? Forecast covers by starting with historical covers for the same day of week and similar periods, then adjusting for known factors such as reservations, hotel occupancy, events, weather and holidays. Compare the forecast with actuals every day to improve it. A cover is one guest served. A simple approach is the average of recent comparable days, adjusted by a ratio of current reservations to the reservations typical for that day at the same lead time. It matters because staffing and food purchasing depend on it. Commonly misread: - Using a single past day as the forecast. - Mixing walk-in-driven and booked days. - Ignoring cancellations and no-shows. Measure forecast accuracy as the average gap between forecast and actual, and review it regularly. Keep any model simple enough that managers trust it. URL: https://zansa.co.za/answers/how-do-i-forecast-covers-for-a-restaurant ### What is prime cost in a restaurant and why is it tracked? Prime cost is cost of goods sold (food and beverage) plus total labour cost. Prime cost percentage is that sum divided by revenue, multiplied by 100. It captures the two largest controllable costs in a restaurant in one figure. It matters because food cost and labour trade off: cutting one can raise the other, and the combined figure shows the net effect. Commonly misread: - Excluding some labour, such as management or payroll costs, in some periods. - Using purchases instead of actual usage, which ignores stock movements. Opening stock plus purchases minus closing stock gives true usage. - Reading monthly figures only; weekly tracking catches problems sooner. What counts as healthy depends on concept and market, so build your own baseline from history instead of copying a number. URL: https://zansa.co.za/answers/what-is-prime-cost-in-a-restaurant-and-why-is-it-tracked ### How do I calculate food cost percentage correctly? Food cost percentage is the cost of food used divided by food sales, multiplied by 100. Cost of food used equals opening stock plus purchases minus closing stock. Using purchases alone gives a misleading figure whenever stock levels change. It matters because food cost is a major driver of restaurant margin. Commonly misread: - Comparing actual with theoretical food cost without understanding the gap. Theoretical cost comes from recipe costs and items sold; the gap reflects waste, over-portioning, theft or errors. - Ignoring staff meals and comps. - Using outdated recipe costs when supplier prices have changed. Cost each recipe, update prices often, count stock on a regular schedule, and investigate gaps between actual and theoretical cost by category. URL: https://zansa.co.za/answers/how-do-i-calculate-food-cost-percentage-correctly ### What is RevPASH and when is it useful in a restaurant? RevPASH, revenue per available seat hour, is revenue in a period divided by the number of seats multiplied by the hours open. It shows how well seating capacity is turned into revenue, and highlights quiet and busy times. Example: 5,000 currency units in a four-hour service with 50 seats gives 5,000 / (50 x 4) = 25 per seat hour. It matters because seats are a fixed capacity that cannot be stored. Commonly misread: - Calculating over the whole day hides peaks and gaps. Use hourly or service-period views. - Raising it by crowding tables may harm service and repeat visits. - Ignoring costs. A high-revenue hour with high staffing and ingredient cost may not be the most profitable. Use it to adjust pricing, menus, bookings and staffing for different periods. URL: https://zansa.co.za/answers/what-is-revpash-and-when-is-it-useful-in-a-restaurant ### What is direct booking share and why does it matter? Direct booking share is the proportion of bookings (or revenue) that comes through your own channels, such as your website, phone or email, divided by total bookings. It matters because direct bookings usually carry lower commission and give you the guest relationship. Formula in words: direct bookings divided by all bookings, over the same period. Calculate it by count and by revenue. Commonly misread: - Treating a higher direct share as good without checking net revenue; a discount offered to win direct bookings can cancel out commission savings. - Mislabelling repeat guests who found you on a third-party site. - Source data from the booking engine and property system not matching. Compare net ADR, cancellation rate and length of stay between direct and third-party bookings to see the full picture. URL: https://zansa.co.za/answers/what-is-direct-booking-share-and-why-does-it-matter ### What is a hospitality morning dashboard and what should it show? A morning dashboard is a one-screen daily summary for managers of last night's results and today's outlook. For a hotel it typically shows occupancy, ADR and RevPAR, labour cost against revenue, and arrivals and departures for the day ahead. It matters because daily decisions on staffing, pricing and purchasing depend on fresh numbers, not month-end reports. Good practice: - Keep it to a handful of measures. - Show actuals against forecast and against the same day last year or last week. - Highlight exceptions rather than listing everything. Zansa builds a morning dashboard with occupancy, RevPAR and labour, using PMS, POS and payroll data joined per property. URL: https://zansa.co.za/answers/what-is-a-hospitality-morning-dashboard-and-what-should-it-show ### What does Zansa's hospitality analytics service include for hotels and restaurants? Zansa joins PMS, POS and payroll data per property and builds a morning dashboard with occupancy, RevPAR and labour, plus a forecast of covers and rooms. It is aimed at hotels, lodges and restaurants that want decisions based on daily numbers. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process is a free intro call, a fixed-fee Data Health Check of one to two weeks, then a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Optional support follows. Dashboards are usually in Metabase or Power BI. Data lives in a cloud account owned by the client, usually Google BigQuery, and the client owns everything built. Guidance pricing: Health Checks from R25,000; builds typically R80,000 to R250,000; support from R12,000 a month. URL: https://zansa.co.za/answers/what-does-zansa-s-hospitality-analytics-service-include-for-hotels-and-restauran ### How can hotels and restaurants forecast rooms and covers with limited data? Start simply: use the same weekday's history, add current bookings compared with what was typical at the same lead time, and adjust for events and seasons. Measure the gap between forecast and actual each day, then refine. Complex models are only worth it if they beat this baseline. A forecast is useful when it drives a decision, such as how many staff to roster or how much to prepare. Steps: - Collect at least a year of daily history, if available. - Separate days with events or closures. - Compare forecasts with actuals and keep a record. Commonly misread: - Treating a forecast as a certainty rather than a range. - Judging accuracy on one good or bad day. Zansa builds forecasts of covers and rooms for hospitality clients, using data joined from PMS, POS and payroll systems. URL: https://zansa.co.za/answers/how-can-hotels-and-restaurants-forecast-rooms-and-covers-with-limited-data ## Integration and BI tools ### Why do my Xero and Shopify sales numbers never match each other? Xero and Shopify usually disagree because they record different things at different times. Shopify counts orders when they are placed, while Xero records invoices, payouts, fees, refunds and tax when they are posted. Differences in timing, payment gateway fees, discounts, shipping and tax treatment explain most gaps, so the numbers need a defined bridge between them. Common causes of the gap: - Shopify reports gross orders, while your payment provider pays out net of fees on a delay. - Refunds and chargebacks land in different periods. - Tax-inclusive and tax-exclusive prices are mixed. - Gift cards and store credit are treated as sales in one system and liabilities in another. The fix is a reconciliation that walks from orders to payouts to the bank and ledger, with each difference labelled. Once the bridge exists, you can automate it and only investigate the leftover items. URL: https://zansa.co.za/answers/why-do-my-xero-and-shopify-sales-numbers-never-match-each-other ### How do I reconcile sales figures between my till system and my accounting software? Start by agreeing one definition of a sale, then compare the same period, the same currency and the same tax basis in both systems. Reconcile till end-of-day totals to the bank deposits and to the journals posted in accounting, and list every difference by type, such as cash variances, voids, discounts and card fees. A practical routine: - Export daily totals from the till by tender type (cash, card, vouchers). - Export the matching journal or deposit lines from accounting. - Match on date and tender type, then list unmatched items. Typical reasons for differences are late card settlements, cash banked a day later, voided sales, tips, and tax rounding. Keep a short log of recurring differences so they stop being surprises. If the till can export transaction-level data, load it into a database rather than relying on end-of-day summaries. URL: https://zansa.co.za/answers/how-do-i-reconcile-sales-figures-between-my-till-system-and-my-accounting-softwa ### What is ETL and how is it different from ELT? ETL means extract, transform, then load: data is cleaned and reshaped before it reaches the warehouse. ELT means extract, load, then transform: raw data lands in the warehouse first and is transformed there using SQL. Modern cloud warehouses make ELT the common choice because storage is cheap and transformations are easier to version and rerun. With ETL, the transformation happens in a separate tool or script, so the warehouse only holds finished data. If the logic is wrong, you may need to re-extract from the source. With ELT, the raw copy is kept, so you can change your logic and rebuild without calling the source system again. Tools such as dbt handle the transform step in SQL. For small and mid-sized companies, ELT is usually simpler to maintain, but ETL can still make sense when you must remove sensitive fields before data leaves a source system. URL: https://zansa.co.za/answers/what-is-etl-and-how-is-it-different-from-elt ### What is dbt and does a small business need it? dbt is an open-source tool that lets you write data transformations as SQL files, test them, document them and run them in order inside a warehouse. A small business does not strictly need it, but it becomes valuable once you have several data sources, shared metrics and more than one person touching the logic. dbt brings software habits to analytics: version control, tests, documentation and dependency graphs. Instead of a pile of hidden spreadsheet formulas or ad hoc queries, each model is a readable SQL file. It is probably overkill if you have one source and a handful of reports. It starts paying off when numbers must be consistent across dashboards, when you need automated checks, or when you want someone new to understand how a figure is calculated without asking the person who built it. URL: https://zansa.co.za/answers/what-is-dbt-and-does-a-small-business-need-it ### What is a star schema and why does it matter for dashboards? A star schema organises data into one central fact table of events, such as sales lines, surrounded by dimension tables that describe them, such as customers, products and dates. It matters because it makes dashboards faster, simpler to build and more consistent, since every report uses the same definitions and joins. A fact table holds measurable events with numeric values and keys. Dimension tables hold descriptive attributes you filter and group by. Benefits for BI: - Business users can drag fields without writing joins. - Metrics are defined once, so two dashboards do not give different totals. - Performance is usually better than querying many raw operational tables. You do not need a perfect textbook design. Even a simple version, with a clean sales fact and a few dimensions, is a big improvement over reporting straight from raw exports. URL: https://zansa.co.za/answers/what-is-a-star-schema-and-why-does-it-matter-for-dashboards ### Should a small company use BigQuery, Snowflake or Postgres as its data warehouse? For most small and mid-sized companies, any of the three works, and the choice depends on data volume, existing cloud, skills and billing model. BigQuery and Snowflake are managed cloud warehouses that scale easily. Postgres is a free, mature database that is often enough for modest data volumes and simple needs. Rough guide: - BigQuery: serverless, charges mainly by storage and data scanned or compute reserved, and fits well if you already use Google tools. - Snowflake: separates storage and compute, bills by usage, and runs on several clouds. - Postgres: familiar, cheap, easy to self-host or use managed, but you handle more tuning as data grows. Pricing and limits change, so check current terms. If your data is under a few gigabytes and queries are simple, Postgres is often fine. If you expect growth or many sources, a cloud warehouse reduces operational work. URL: https://zansa.co.za/answers/should-a-small-company-use-bigquery-snowflake-or-postgres-as-its-data-warehouse ### How does Snowflake pricing work in general terms? Snowflake generally charges separately for storage and for compute. Compute is consumed when virtual warehouses run queries, and you pay for the time they are active, while storage is charged by the amount of data kept. Exact rates and plans vary and change, so check current terms before committing. Key ideas: - Warehouses can be sized up or down and can suspend automatically when idle, which keeps costs down. - Heavy dashboards that refresh often on a large warehouse can become costly if not managed. - Separate warehouses for loading and for reporting prevent one workload slowing the other. For a small company, the main risk is leaving compute running unnecessarily or letting many users refresh heavy queries. Set auto-suspend, resource monitors and sensible refresh schedules. URL: https://zansa.co.za/answers/how-does-snowflake-pricing-work-in-general-terms ### Is Postgres good enough as a reporting database for a small business? Yes, often. Postgres handles reporting for small and mid-sized datasets very well, and it is free, widely supported and works with every major BI tool. It becomes limiting when tables reach hundreds of millions of rows or many users run heavy queries at once against your operational system. Good practice if you use it for reporting: - Do not report directly from your live production database. Use a read replica or a separate reporting database. - Add indexes for common filters and joins. - Use views or materialised views for repeated calculations. If your volumes grow, you can move to a cloud warehouse later because the SQL you wrote is largely portable. Starting simple on Postgres is a reasonable choice when budget is tight. URL: https://zansa.co.za/answers/is-postgres-good-enough-as-a-reporting-database-for-a-small-business ### Metabase vs Power BI: which is better for a small company? It depends on your existing tools and user count. Metabase is open-source, simple and easy for non-technical people to explore, and its open-source edition has no per-user licence fee. Power BI is more powerful for complex modelling and integrates tightly with Microsoft 365, but is licensed per user in most setups. Choose Metabase when you want quick, clean dashboards, easy self-service questions and predictable costs across many viewers. Choose Power BI when your team already lives in Microsoft tools, needs advanced DAX calculations, or wants deep Excel integration. Things to compare: - Licence model and how many viewers you have. - Who will build and maintain reports. - Hosting preference, since Metabase can be self-hosted. - Features like row-level security and embedding, which differ by edition. The data model underneath matters more than the tool. URL: https://zansa.co.za/answers/metabase-vs-power-bi-which-is-better-for-a-small-company ### Is Looker Studio good enough for business reporting? Looker Studio is good for lightweight, shareable reports, especially on Google data such as Analytics, Ads and Sheets, and it has no per-user licence cost for the core product. It is weaker for complex data modelling, large datasets and governed metrics, so it works best on top of a clean, pre-modelled source. Strengths: - Quick to start and easy to share by link. - Native connectors for Google products. - Works well with BigQuery. Limitations: - Calculations and joins inside the tool can get messy and slow. - Version control and testing are limited. - Some third-party connectors are paid or unreliable. A good pattern is to do the heavy lifting in a warehouse or SQL view, then use Looker Studio only for presentation. Otherwise logic gets buried in individual reports. URL: https://zansa.co.za/answers/is-looker-studio-good-enough-for-business-reporting ### When is Tableau worth it compared with cheaper BI tools? Tableau is worth considering when you need highly custom, visually sophisticated analysis, have analysts who already know it, or work with large, varied datasets and want strong visual exploration. For standard operational dashboards, cheaper or open-source tools often deliver most of the value at lower licence cost. Tableau is typically licensed per user with different roles for creators, explorers and viewers, so costs scale with headcount. Check current terms for details. Ask yourself: - Do we need visual flexibility that simpler tools cannot offer? - Do we have people who can build and maintain Tableau workbooks? - How many people only need to view dashboards? If the answers lean towards basic KPIs for many viewers, a lower-cost option is usually more sensible. If you have a few expert analysts doing deep exploration, Tableau can justify itself. URL: https://zansa.co.za/answers/when-is-tableau-worth-it-compared-with-cheaper-bi-tools ### Can Excel be used as a business intelligence tool? Yes, up to a point. Excel with Power Query and PivotTables can combine files, clean data and produce useful reports, and many businesses run on it for years. It struggles when many people edit copies, data grows large, or the logic becomes impossible to audit or refresh reliably. Excel works well for ad hoc analysis, small datasets and finance-led reporting. Warning signs that you have outgrown it: - Multiple versions of the same file circulating. - Manual copy-and-paste steps every week. - Formulas only one person understands. - Files that crash or take minutes to open. A common next step is keeping Excel as a familiar front end but feeding it from a central database or data model, so everyone starts from the same numbers. URL: https://zansa.co.za/answers/can-excel-be-used-as-a-business-intelligence-tool ### How can I tell when my spreadsheets have outgrown Excel? You have outgrown Excel when reports take hours of manual work, two people produce different totals from the same data, files are slow or break, or nobody trusts the numbers. These are signs that the process, not just the file, needs a proper data source and automated refresh. Check for these signals: - The monthly report depends on one person's laptop. - Data is copied between files by hand. - You cannot easily trace where a number came from. - You need data from several systems in one view. - Sensitive data is emailed around. Moving on does not mean abandoning Excel. It means loading source data into a database or warehouse, applying documented transformations, and letting Excel or a BI tool read from that shared source. URL: https://zansa.co.za/answers/how-can-i-tell-when-my-spreadsheets-have-outgrown-excel ### What is a data pipeline in plain English? A data pipeline is an automated process that moves data from where it is created, such as an accounting system or online shop, to a place where it can be analysed, and prepares it along the way. It runs on a schedule so reports stay current without manual exports. A basic pipeline has three parts: - Extract: pull data from a source via an API, database connection or file. - Load: store it in a central location such as a warehouse. - Transform: clean, join and reshape it into reporting tables. Good pipelines also include logging, error alerts and tests, so you know when something fails. Without those, a pipeline can break silently and dashboards show stale or wrong data. URL: https://zansa.co.za/answers/what-is-a-data-pipeline-in-plain-english ### How do APIs let me pull data out of business systems? An API is a structured way for software to request data from another system. A pipeline sends a request to, for example, Xero or Shopify, authenticates, and receives the data in a format like JSON, which it then stores in a database. This replaces manual exports with repeatable, automatic extraction. Things to know: - Most APIs require credentials, such as OAuth tokens or API keys, which must be stored securely. - Many limit how many requests you can make in a given time, so large pulls are paged or throttled. - Fields and endpoints can change, so pipelines need monitoring. - Some data, such as certain historical reports, may not be available via the API. Always read the vendor's documentation for what the API exposes, since coverage differs between systems and plans. URL: https://zansa.co.za/answers/how-do-apis-let-me-pull-data-out-of-business-systems ### How do I connect Xero to a BI tool or database? You can connect Xero through its API using a pipeline or connector that authenticates with OAuth, pulls invoices, payments, contacts and journals, and loads them into a database or warehouse. Your BI tool then reads from that database. Some BI tools also offer direct connectors, though these can be limited. Typical steps: - Create an app or authorise a connector in Xero with read-only scopes. - Extract entities such as invoices, bank transactions, accounts and tracking categories. - Load them into a warehouse with timestamps. - Model them into clean reporting tables. Remember Xero's API has rate limits and the data model reflects accounting logic, so you need to handle credit notes, multi-currency and tracking categories carefully. Keeping a history of loaded data also helps you spot changes made to past periods. URL: https://zansa.co.za/answers/how-do-i-connect-xero-to-a-bi-tool-or-database ### How do I get Sage data into a reporting database? The route depends on which Sage product you use. Cloud versions typically offer an API or export options, while desktop products often need database access, ODBC connections or scheduled report exports. Whichever route, load the data into a central database and model it there rather than reporting straight from Sage. Sage covers several different products with different technical options, so confirm yours first. General approaches: - API extraction for cloud editions. - Scheduled CSV or Excel exports where no API is available. - Direct read-only database or ODBC access for some on-premises versions, where the vendor allows it. Always check licensing and support terms before connecting directly to a database. Use read-only credentials and run extracts outside peak hours. Keep raw extracts so you can re-run transformations if logic changes. URL: https://zansa.co.za/answers/how-do-i-get-sage-data-into-a-reporting-database ### Can I connect QuickBooks to Power BI or Metabase? Yes, usually by extracting QuickBooks data through its API into a database or warehouse and then connecting your BI tool to that. Some BI tools offer direct connectors, but a staging database gives you better performance, history and control over how accounting data is modelled. Considerations: - QuickBooks Online has an API with authentication and rate limits; desktop versions work differently and may need export-based approaches. - Standard reports in QuickBooks may not match raw transaction data unless you replicate its logic. - Classes, locations and custom fields should be pulled explicitly if you report on them. Plan for reconciling your loaded totals to QuickBooks' own profit and loss and balance sheet, so you can trust the data before building dashboards. URL: https://zansa.co.za/answers/can-i-connect-quickbooks-to-power-bi-or-metabase ### How do I pull Shopify data into a data warehouse? Use Shopify's API or a connector to extract orders, line items, customers, products, refunds and transactions on a schedule, then load them into a warehouse such as BigQuery. Model them into clean sales tables, and reconcile them against payouts and your accounting system. Tips: - Pull orders with line-level detail, not only order totals. - Capture refunds, discounts, shipping and taxes as separate fields. - Decide how to treat test orders, cancelled orders and draft orders. - Keep timestamps and time zones consistent, as Shopify store time zones can cause date mismatches. Customer data is personal information, so store it with appropriate access controls and respect privacy law such as POPIA or GDPR. URL: https://zansa.co.za/answers/how-do-i-pull-shopify-data-into-a-data-warehouse ### How do I connect WooCommerce to reporting tools? WooCommerce stores data in a MySQL or MariaDB database and also exposes a REST API. You can either extract orders and products through the API on a schedule or replicate the database to a reporting copy, then model the data and connect your BI tool to that. Considerations: - Orders, line items, coupons and refunds sit in several related tables or API objects. - Plugins often add custom fields that you must map explicitly. - Querying the live shop database directly can slow the website, so use a replica or off-peak extracts. - HPOS (high-performance order storage) changes where order data lives in newer versions, so check your setup. As with any shop data, reconcile against payment gateway reports and accounting totals before trusting dashboards. URL: https://zansa.co.za/answers/how-do-i-connect-woocommerce-to-reporting-tools ### How can I get data out of an old till or point-of-sale system? Check what the system offers: a built-in API, scheduled report exports to CSV or Excel, or a database you can read from. If it offers none, ask the vendor, since many can provide exports. Automate whatever route exists and load the files into a central database daily. Options in order of preference: - API or cloud export. - Scheduled file exports dropped in a shared folder or storage bucket. - Read-only database access, if permitted by the vendor. - Manual export as a last resort, with a strict routine. Transaction-level data (each sale line) is much more useful than daily summaries because it lets you analyse products, times and staff. Watch for quirks such as voids, returns and end-of-day cut-offs that differ from calendar days. URL: https://zansa.co.za/answers/how-can-i-get-data-out-of-an-old-till-or-point-of-sale-system ### How do I integrate an ERP with a BI tool? Extract data from the ERP into a separate reporting database or warehouse, model it into business-friendly tables, and connect the BI tool to that. Avoid reporting directly on the ERP's live tables, which are complex, can affect performance and may not match business terminology. ERP databases are built for transactions, not analysis. Tables are numerous, codes are cryptic and business rules are buried in application logic. Steps: - Identify the entities you need, such as customers, items, orders and stock movements. - Extract them via API, ODBC or vendor-supported export, using read-only access. - Translate codes into meaningful names in a modelling layer. - Test totals against ERP reports. Involve someone who knows how the ERP is configured, since custom fields and processes differ between companies. URL: https://zansa.co.za/answers/how-do-i-integrate-an-erp-with-a-bi-tool ### How do I combine CRM data with finance data in one report? Find a shared key, such as customer ID, company name or email domain, that links CRM records to accounting contacts, then build a customer master table that maps both. Join revenue from finance to pipeline and activity from the CRM through that mapping. The hard part is almost always matching customers, since names are typed differently in each system. Approach: - Create a master customer list with one ID per real customer. - Store mapping rows from CRM IDs and accounting IDs to the master ID. - Handle parent and child accounts explicitly. - Review unmatched records manually and save the decisions so they are not repeated. Once linked, you can report on customer profitability, sales cycle versus payment behaviour, and pipeline versus actual revenue. URL: https://zansa.co.za/answers/how-do-i-combine-crm-data-with-finance-data-in-one-report ### What is master data and why does it matter? Master data is the core reference information your business depends on, such as customers, products, suppliers, locations and chart of accounts. It matters because when each system holds a different version of it, reports disagree and joins between systems fail. Examples of master data problems: - The same customer under three spellings across CRM, billing and shop. - Product codes that differ between ERP and e-commerce. - Branch names that change over time. Good practice is to decide which system owns each type of master data, keep a mapping table where systems cannot be aligned, and assign someone responsible for changes. You do not need expensive software to start. A well-maintained mapping table in a database can solve most small-company issues. URL: https://zansa.co.za/answers/what-is-master-data-and-why-does-it-matter ### How do I remove duplicate customers from my data? Standardise the fields first, such as trimming spaces, lowercasing emails and normalising phone formats, then match on strong identifiers like email or tax number. For fuzzy matches on names and addresses, flag likely duplicates for human review before merging, and keep a record of every merge decision. Steps that work in practice: - Clean: trim, case-fold, remove punctuation, standardise abbreviations. - Match: exact on unique IDs, then fuzzy on name plus address or postcode. - Review: automated merging of uncertain matches creates new errors. - Merge: keep a surviving record and a mapping from old IDs to the new one. Do not delete the originals in source systems without understanding the downstream effects on invoices and history. Often it is safer to deduplicate in the reporting layer and leave source records untouched. URL: https://zansa.co.za/answers/how-do-i-remove-duplicate-customers-from-my-data ### How do I clean messy Excel data without doing it manually every month? Use Power Query in Excel or a script or SQL transformation to record your cleaning steps once and reapply them to each new file. Steps such as trimming, splitting columns, fixing types and removing duplicates then run automatically, rather than being repeated by hand. Power Query records each transformation and replays it on refresh. For larger or shared workflows, load files into a database and clean them with SQL or Python. Best practices: - Keep the raw file untouched and clean a copy. - Name each step so others can follow it. - Add checks, such as row counts and totals, before and after cleaning. - Make the process fail loudly if a column name changes. This turns a fragile monthly chore into a repeatable, documented process. URL: https://zansa.co.za/answers/how-do-i-clean-messy-excel-data-without-doing-it-manually-every-month ### What are data tests and which ones matter most? Data tests are automated checks that run on your data to catch problems before they reach dashboards. The most useful are uniqueness of keys, non-null required fields, valid value ranges, referential integrity between tables, and reconciliation checks that compare totals with a source. Core tests to start with: - Unique and not null on primary keys. - Accepted values, such as order status from a known list. - Relationships, so every sale has a valid product and customer. - Row counts and freshness, to detect missing loads. - Totals match the source system within an agreed tolerance. Tools like dbt make these easy to define alongside your models. Start with a few high-value tests on the tables that feed your key dashboards, then add more when issues occur. URL: https://zansa.co.za/answers/what-are-data-tests-and-which-ones-matter-most ### How do I monitor data pipelines so I know when they break? Set up automated alerts for failed jobs, late data and unexpected volumes, sent to a channel someone actually watches. Combine job status monitoring with data tests that check freshness, row counts and key totals, so silent failures are caught as well as obvious crashes. Useful monitoring layers: - Job-level: did each scheduled run succeed, and how long did it take? - Freshness: when was each table last updated? - Volume: did today's row count fall sharply versus normal? - Business checks: do revenue totals still match the source? Also show a last-updated timestamp on dashboards. Document who is responsible for responding and what the first diagnostic steps are. Most pipeline failures are caused by expired credentials, changed source fields or API limits. URL: https://zansa.co.za/answers/how-do-i-monitor-data-pipelines-so-i-know-when-they-break ### What should data documentation include for a small company? Good documentation covers where each dataset comes from, what each table and column means, how key metrics are calculated, how often data refreshes, who owns it and any known quirks. Keep it short, current and stored where the team can find it. A minimum useful set: - A list of source systems and how they connect. - A data dictionary with plain-English column definitions. - Metric definitions, for example how revenue, margin or active customer is calculated. - A diagram of how tables relate. - Runbook notes on what to do when a refresh fails. Documentation that lives next to the code, as in dbt, is easier to keep updated than a separate document. Outdated documentation is worse than none, so assign an owner. URL: https://zansa.co.za/answers/what-should-data-documentation-include-for-a-small-company ### What is a data dictionary and how do I create one? A data dictionary is a reference listing every field in your data with its meaning, type, source, allowed values and any business rules. Create one by listing your key tables, describing each column in plain language, and getting the people who use the data to confirm the definitions. Start small with the tables behind your main dashboards. For each column record: - Name and data type. - Plain-English description. - Source system and field. - Allowed values or units, such as currency or time zone. - Notes on edge cases. Tools such as dbt can generate documentation from descriptions you write alongside the models. Review it whenever definitions change, since the most common source of confusion is two teams using the same word differently. URL: https://zansa.co.za/answers/what-is-a-data-dictionary-and-how-do-i-create-one ### Why does the same KPI show different values in different dashboards? It usually happens because each dashboard calculates the metric separately, with different filters, date logic, currency handling or source tables. The fix is to define each metric once in a shared data model and have every dashboard use that definition. Typical causes: - One report excludes refunds, another includes them. - Date fields differ, such as order date versus invoice date. - Time zones shift transactions across days. - Different joins duplicate or drop rows. Solve it with a governed layer: metrics defined in SQL models or a semantic layer, documented, and tested. Then dashboards select from those definitions instead of rebuilding logic. When someone asks why numbers differ, you can trace both back to a single documented rule. URL: https://zansa.co.za/answers/why-does-the-same-kpi-show-different-values-in-different-dashboards ### What is a semantic layer in business intelligence? A semantic layer is a shared set of business definitions, such as revenue, customer and margin, that sits between raw data and dashboards. It ensures that everyone calculating a metric uses the same logic, and lets non-technical users build reports with consistent terms. It can be implemented in different ways: - Modelled tables in SQL that already contain the correct logic. - Metric definitions in a BI tool's modelling feature. - A dedicated metrics layer tool. For small companies, well-designed reporting tables in the warehouse often serve as the semantic layer, with documented definitions. The goal is the same: avoid each analyst or dashboard reinventing the formula, which is how contradictory numbers appear. URL: https://zansa.co.za/answers/what-is-a-semantic-layer-in-business-intelligence ### How do per-user BI licence costs add up as my team grows? With per-user licensing, cost grows roughly in line with headcount, so rolling dashboards out to many staff can become the largest part of a BI budget. Viewer-only roles are often cheaper than builder roles, and some tools charge nothing per user, so the model matters as much as the headline price. Things to model before choosing a tool: - How many people build reports versus only view them. - Whether external users, such as clients or franchisees, also need access. - Whether sharing or embedding requires a higher tier or capacity-based licence. - Whether prices are per user, per capacity, or per server. Pricing changes often, so request current quotes. A tool that looks cheap for five users may be expensive for fifty, and the reverse can be true for capacity-based plans. URL: https://zansa.co.za/answers/how-do-per-user-bi-licence-costs-add-up-as-my-team-grows ### Should I self-host my BI tool or use a cloud version? Choose cloud-hosted if you want minimal maintenance and have no one to manage servers. Choose self-hosted if you need control over data location, want to avoid per-user fees, or have the skills to run it securely. Many small companies are better served by managed hosting. Self-hosting gives control and flexibility, but you are responsible for uptime, patches, backups, SSL and access security. Cloud versions reduce workload and typically include upgrades and support, but cost more over time and may restrict some configuration. Questions to ask: - Who will be on call when it goes down? - Are there data residency or compliance requirements? - How many users will access it? Whichever you choose, ensure the data underneath is in an account you own so you can switch tools later. URL: https://zansa.co.za/answers/should-i-self-host-my-bi-tool-or-use-a-cloud-version ### Can I embed dashboards in my own website or application? Yes, most BI tools support embedding, usually through iframes or an SDK, but the options, security model and licensing vary widely. Public embedding is simple but exposes data to anyone with the link, while secure embedding passes signed tokens that restrict what each viewer sees. Questions to check: - Is embedding included in the plan, or does it need a premium tier or capacity licence? - Can you filter data per customer, using row-level security or signed parameters? - Can you style it to match your brand? - How are viewers charged? Never embed a public link for sensitive data. If you are showing clients their own data, test thoroughly that one client cannot see another's by changing parameters or URLs. URL: https://zansa.co.za/answers/can-i-embed-dashboards-in-my-own-website-or-application ### What is row-level security in BI and when do I need it? Row-level security restricts which rows of data each user can see, so a branch manager sees only their branch, or a client only their records. You need it whenever the same dashboard serves people who should not see all the data. It is typically enforced through user attributes mapped to filters, applied in the BI tool or, more robustly, in the database itself. Points to consider: - Enforce security as close to the data as possible; hiding a filter in the front end is not security. - Test with real user accounts, not just admin accounts. - Document the rule so it survives staff turnover. - Features and editions differ between tools, so check what your plan supports. This is also relevant for privacy compliance, since personal data should be available only to those who need it. URL: https://zansa.co.za/answers/what-is-row-level-security-in-bi-and-when-do-i-need-it ### What are the core principles of good dashboard design? Good dashboards answer a specific question for a specific audience, show the most important numbers first, use simple charts, and avoid clutter. Every element should help someone make a decision, and definitions and refresh time should be clear so people trust what they see. Practical principles: - Start with the decisions the viewer needs to make. - Lead with a few key figures, then supporting detail. - Use comparisons such as versus last period or target to give context. - Choose the simplest chart that works; avoid 3D effects and decoration. - Use colour sparingly and consistently. - Limit the number of visuals per page. - Label units, filters and last-updated time. Test with real users. If they have to ask what a number means, the dashboard needs work. URL: https://zansa.co.za/answers/what-are-the-core-principles-of-good-dashboard-design ### Which chart type should I use for different business questions? Use line charts for trends over time, bar charts for comparing categories, tables for exact values, and simple numbers with comparisons for key metrics. Avoid pie charts with many slices and avoid dual axes unless carefully explained, because they make comparison harder. Quick guide: - Trend: line chart. - Ranking or comparison: horizontal bar chart. - Part of a whole with few categories: stacked bar or a simple pie. - Relationship between two measures: scatter plot. - Distribution: histogram. - Exact figures, such as invoices: table. - Single KPI: big number with change versus prior period. Start bar charts at zero, order categories meaningfully, and keep scales consistent between charts that viewers will compare. URL: https://zansa.co.za/answers/which-chart-type-should-i-use-for-different-business-questions ### What is the difference between a data warehouse, data lake and database? A database stores data for running an application, a data warehouse stores cleaned, structured data for analysis, and a data lake stores raw files of any type cheaply. Small companies usually need only a warehouse, sometimes with a raw landing area alongside it. Database: optimised for fast reads and writes of individual records, such as an order system. Data warehouse: optimised for analytical queries across many rows, with data organised for reporting. Data lake: file storage for raw, semi-structured or unstructured data, such as logs, documents or JSON dumps. Modern cloud warehouses can also hold raw loaded data in separate schemas, so one platform often serves both purposes. Choose based on your needs rather than buzzwords. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-data-warehouse-data-lake-and-database ### How do I handle currency and exchange rates in combined reports? Store each transaction in its original currency and also convert it to a reporting currency using a clearly defined rate, such as the rate on the transaction date or a monthly average. Document which rate you use, because it affects totals and reconciliation to accounts. Decisions to make and write down: - Which currency is the reporting currency? - Which rate source and timing: daily, month-end or average? - Do you report in constant currency to remove exchange effects? Keep an exchange rate table in the warehouse with dates, and join to it in the model. Your accounting system may revalue balances with its own rates, so expect small differences versus management reports and explain them rather than hiding them. URL: https://zansa.co.za/answers/how-do-i-handle-currency-and-exchange-rates-in-combined-reports ### Why do dates and time zones cause mismatched reports? Systems record timestamps in different time zones and use different date fields, such as order date, payment date or invoice date. A sale near midnight can fall on different days in different systems, causing daily and monthly totals to disagree. To avoid this: - Store timestamps in UTC and convert to a business time zone for reporting. - Choose which date represents a sale for each report and document it. - Be careful with month-end cut-offs, where small timing differences move revenue between periods. - Check financial year and week-start conventions. When reconciling two systems, compare totals using the same date field and time zone first. Many unexplained differences disappear once dates are aligned. URL: https://zansa.co.za/answers/why-do-dates-and-time-zones-cause-mismatched-reports ### How should I reconcile data between two systems step by step? Pick one period and one measure, extract both datasets at the lowest common level, such as transactions, and match them on a shared key. Then categorise every unmatched or differing item by cause, fix what you can, and document the rest as known differences. A repeatable process: - Agree definitions and scope: dates, currency, tax, status filters. - Compare totals first, then drill down by day, then by transaction. - Match on IDs where possible; use amount plus date as a fallback. - Classify differences: timing, missing records, duplicates, rounding, fees. - Fix upstream causes, not just the report. - Automate the comparison so it runs each refresh and flags new differences. A small tolerance for rounding is reasonable, but anything above it should be explained. URL: https://zansa.co.za/answers/how-should-i-reconcile-data-between-two-systems-step-by-step ### What is a data model and why should small businesses care? A data model is the structured design of your data: which tables exist, what they contain and how they relate. Small businesses should care because a clear model makes reports faster to build, consistent across users and easier to maintain than patchwork spreadsheets. Without a model, each report author joins raw exports their own way, so results differ and changes break things. With a model: - Business concepts like customer, product and order have one definition. - Relationships between tables are explicit. - New data sources are added in a consistent pattern. - Dashboards can be rebuilt on a different tool without redoing the logic. It does not need to be elaborate. A simple, documented model is far better than none. URL: https://zansa.co.za/answers/what-is-a-data-model-and-why-should-small-businesses-care ### What is a fact table and what is a dimension table? A fact table records events or measurements, such as each sale line with quantity and amount. A dimension table describes the things involved, such as the product, customer, store or date. Facts are what you add up, dimensions are how you slice them. Example: a sales fact table has one row per order line, with keys pointing to product, customer and date, and numeric columns such as quantity and revenue. The product dimension holds name, category and supplier. Good habits: - Decide the grain of a fact table, meaning what one row represents, and keep it consistent. - Give dimensions stable keys. - Include a proper date dimension for fiscal periods and holidays. Getting the grain right avoids double-counting, one of the most common causes of wrong totals. URL: https://zansa.co.za/answers/what-is-a-fact-table-and-what-is-a-dimension-table ### What does 'grain' mean in a data model and why does it cause double counting? Grain is what a single row in a table represents, such as one order, one order line or one payment. Double counting happens when you join tables of different grain, like orders to lines, and then sum an order-level value that has been repeated on every line. Example: an order has a shipping fee of 10 and three lines. Join orders to lines and sum shipping, and you get 30 instead of 10. How to avoid it: - Write down the grain of every table. - Aggregate to a common grain before joining. - Keep order-level amounts in the order table and line amounts in the line table. - Test that totals after joining equal totals before joining. A simple test comparing row counts and sums before and after a join catches many of these errors. URL: https://zansa.co.za/answers/what-does-grain-mean-in-a-data-model-and-why-does-it-cause-double-counting ### How do I deal with handwritten or paper records in my data? Decide whether the records are worth digitising, then capture them with a structured form or scan and extract the data, and always validate it. Optical character recognition can help with printed documents, but handwriting typically needs human checking, so plan for review rather than trusting it blindly. Practical approach: - Replace paper with a simple digital form or spreadsheet with validation going forward. - For backlogs, prioritise records that matter for reporting or compliance. - Use scanning plus OCR for typed documents, and double entry or sampling checks for handwritten ones. - Record the source document reference for each row so errors can be traced. Paper records containing personal information must still be handled with care under laws such as POPIA or GDPR, including secure storage and disposal. URL: https://zansa.co.za/answers/how-do-i-deal-with-handwritten-or-paper-records-in-my-data ### Can OCR reliably extract data from invoices and scanned documents? OCR works reasonably well on clean, printed documents with consistent layouts, but accuracy falls with poor scans, unusual layouts, stamps or handwriting. Treat OCR output as a draft, validate key fields like totals and dates, and route low-confidence results for human review. Ways to improve reliability: - Use good-quality scans or original PDFs where possible. - Validate extracted amounts, for example that line items add up to the total. - Cross-check supplier and invoice numbers against existing records to catch duplicates. - Keep the original file linked to the extracted row. Modern document-extraction services and AI tools can handle varied layouts, but they still make mistakes. Never send sensitive personal data to a tool without checking where it is processed and stored. URL: https://zansa.co.za/answers/can-ocr-reliably-extract-data-from-invoices-and-scanned-documents ### What is the best way to import CSV files into a database reliably? Load each CSV into a raw staging table exactly as received, then validate and transform it into clean tables. Check column names, data types, encoding, delimiters and row counts on every load, and reject or flag files that do not match the expected structure. Common CSV traps: - Different character encodings that corrupt accented names. - Commas inside fields without proper quoting. - Leading zeros lost when codes are read as numbers. - Dates in day-month versus month-day order. - Headers that change between exports. Keep a record of which file was loaded and when, so you can trace and avoid loading the same file twice. Automating delivery to a shared folder or storage bucket makes the process dependable. URL: https://zansa.co.za/answers/what-is-the-best-way-to-import-csv-files-into-a-database-reliably ### How can I automate weekly Excel reports so nobody copies and pastes? Store the source data in a central place, build the calculations once in a query or data model, and connect the report to it with scheduled refresh. Distribute the output as a live dashboard or an automatically generated file instead of hand-built workbooks. Options from simple to robust: - Power Query connected to a shared folder, refreshing when opened. - A database with a BI dashboard that refreshes on a schedule. - Scheduled email of dashboard snapshots or exports, which many BI tools support. The key step is removing manual data handling. Once data lands automatically and logic lives in one place, the report becomes a view rather than a task. Add checks so an incomplete load does not quietly produce wrong figures. URL: https://zansa.co.za/answers/how-can-i-automate-weekly-excel-reports-so-nobody-copies-and-pastes ### What is Power Query and how does it help with data preparation? Power Query is a data preparation tool built into Excel and Power BI. It lets you connect to sources, clean and reshape data through recorded steps, and refresh the whole process with one click, making repetitive preparation consistent and repeatable. Common uses: - Combine many monthly files from a folder. - Unpivot cross-tab reports into proper tables. - Split, merge and standardise columns. - Merge tables from different sources. Its steps are written in a language called M, but most can be created through the interface. It is great for individual analysts and moderate data sizes. For shared, large-scale or heavily tested transformations, SQL in a warehouse is usually easier to govern. URL: https://zansa.co.za/answers/what-is-power-query-and-how-does-it-help-with-data-preparation ### What are the downsides of using Google Sheets as a data source for dashboards? Google Sheets is convenient but fragile as a source: anyone can change cells, formats drift, formulas break, performance drops with large data, and there is little validation or history. Dashboards built on it can silently show wrong numbers when someone edits a column. Risks: - No enforced data types or required fields. - Row and cell limits and slow refresh on big sheets. - Renaming a column or tab can break connections. - Hard to audit who changed what and why. If you must use Sheets, lock structure, use data validation, keep a raw input tab separate from calculations, and consider copying the data into a database on a schedule. For anything business-critical, a proper database is more dependable. URL: https://zansa.co.za/answers/what-are-the-downsides-of-using-google-sheets-as-a-data-source-for-dashboards ### How do I choose a BI tool for my business? Start with who will use it, which data sources you have, how many viewers there are, your budget model, and who will maintain it. Shortlist two or three tools, test them on your real data, and weigh licence structure, ease of use, embedding and security needs. Selection checklist: - Users: how many builders and how many viewers? - Skills: SQL, Excel or neither? - Connectors: does it work with your database or warehouse? - Licensing: per-user, capacity or open-source? - Governance: permissions, row-level security, version control. - Hosting: cloud or self-hosted. - Exit: can you move your data model if you switch? A clean data layer makes tool choice less critical, because dashboards can be rebuilt without redoing the underlying logic. URL: https://zansa.co.za/answers/how-do-i-choose-a-bi-tool-for-my-business ### What are the limitations of Power BI for small teams? Power BI is capable, but small teams may find the licensing structure, sharing requirements, DAX learning curve and refresh configuration confusing. Sharing reports with many people typically requires appropriate licences or capacity, and poorly built models can become slow and hard to maintain. Points to watch: - Licensing differs by how reports are shared, so check the current model before rollout. - DAX is powerful but takes time to learn. - Data gateways are needed to refresh on-premises sources. - Reports built directly on messy Excel files become fragile. Power BI works best when it sits on a clean, well-modelled source and the team has someone who understands the data model. It is particularly strong if you already use Microsoft 365 and want deep Excel compatibility. URL: https://zansa.co.za/answers/what-are-the-limitations-of-power-bi-for-small-teams ### How should I handle personal data in dashboards and pipelines? Collect only the personal data you need, restrict access by role, store it in an account you control, and mask or remove identifiers where reports do not require them. Follow the laws that apply to you, such as POPIA in South Africa or GDPR in Europe. Practical measures: - Pseudonymise customer identifiers in analytical tables where possible. - Apply row-level or column-level access rules. - Log who accesses sensitive data. - Set retention rules and delete data you no longer need. - Check where your vendors process and store data. If you use AI tools for coding help, do not paste personal or client-identifiable data into them. This is general guidance and not legal advice, so consult a qualified adviser for your situation. URL: https://zansa.co.za/answers/how-should-i-handle-personal-data-in-dashboards-and-pipelines ### What is the difference between batch and real-time data integration? Batch integration moves data on a schedule, such as nightly or hourly, while real-time or streaming integration moves changes continuously within seconds. Batch is simpler and cheaper and fits most reporting, while real-time suits cases like fraud detection or live operations. Batch pros: easier to build, test and recover, lower cost, good for finance and sales reporting. Real-time pros: immediate visibility, but more infrastructure, more failure modes and typically higher cost. Many businesses think they need real-time when the decisions are made daily or weekly. Ask what action would change if the data were fresher. A middle path is frequent micro-batches for a few operational tables, with daily loads for everything else. URL: https://zansa.co.za/answers/what-is-the-difference-between-batch-and-real-time-data-integration ### How do I keep a history of changes when source systems overwrite data? Take regular snapshots of the tables that change, such as stock levels, prices or customer status, and store them with a date in your warehouse. This gives you a history that the source system itself does not keep, and enables trend analysis. Approaches: - Daily snapshots: copy the full state of a table each day with a snapshot date. - Slowly changing dimensions: keep old and new versions of a record with valid-from and valid-to dates. - Event logs, if the source provides them. Start snapshots as early as possible, since history cannot be recreated afterwards. Storage is cheap compared with the value of knowing what stock, price or status was at a given time. URL: https://zansa.co.za/answers/how-do-i-keep-a-history-of-changes-when-source-systems-overwrite-data ### What are slowly changing dimensions in simple terms? Slowly changing dimensions are descriptive attributes, such as a customer's region or a product's category, that change over time. The question is whether reports should show the current value or the value at the time of each transaction, and the design depends on that choice. Common approaches: - Type 1: overwrite the old value, so history is lost. - Type 2: add a new row with valid-from and valid-to dates, preserving history. Example: if a salesperson moves region, Type 1 reassigns all past sales to the new region, while Type 2 keeps past sales in the old region. Choose based on what the business needs to report. Type 2 is more accurate historically but more complex to build and query. URL: https://zansa.co.za/answers/what-are-slowly-changing-dimensions-in-simple-terms ### How do partitioning and clustering speed up queries in BigQuery? Partitioning splits a table into segments, usually by date, so queries that filter on date read only the relevant segments. Clustering sorts data within a table by chosen columns, so filters on those columns scan less data. Both can reduce cost and improve speed. Good practice: - Partition large fact tables on a date column you filter by often. - Cluster on columns commonly used in filters or joins, such as customer or product ID. - Require a partition filter on very large tables to prevent accidental full scans. Small tables gain little from either. Check current documentation, as features and limits are updated over time. URL: https://zansa.co.za/answers/how-do-partitioning-and-clustering-speed-up-queries-in-bigquery ### What SQL skills does a small business need for reporting? For most reporting, the essentials are SELECT with filtering, joins, grouping and aggregation, date handling, and CASE expressions. Window functions and common table expressions are useful extras. With those, an analyst can answer most business questions from a well-modelled database. Learning order: - SELECT, WHERE, ORDER BY. - GROUP BY with SUM, COUNT, AVG. - JOINs and understanding row multiplication. - Dates and string functions. - CASE for categorising. - CTEs for readable, step-by-step queries. - Window functions for rankings and running totals. With a good BI tool, many staff can ask questions without any SQL, but at least one person who reads SQL helps greatly in checking numbers and debugging mismatches. URL: https://zansa.co.za/answers/what-sql-skills-does-a-small-business-need-for-reporting ### Is it safe to connect a BI tool directly to my production database? It is risky. Heavy analytical queries can slow your live application, and broad access can expose sensitive data. Use a read replica or a separate reporting database, a read-only account with limited permissions, and ideally run queries against modelled tables rather than raw ones. Safer set-up: - Replicate production data to a reporting database or warehouse. - Create a dedicated read-only user. - Restrict that user to the schemas or views needed. - Set query timeouts and limit concurrency. - Allow connections only from known addresses or through a secure tunnel. If you must connect directly for a small system, schedule heavy queries off-peak and monitor performance. Direct access is a short-term compromise, not a long-term design. URL: https://zansa.co.za/answers/is-it-safe-to-connect-a-bi-tool-directly-to-my-production-database ### How do I reconcile stock levels between my ERP, warehouse and online shop? Choose one system as the source of truth for stock, compare quantities per SKU across systems at the same point in time, and classify differences by cause, such as in-transit stock, reservations, returns or timing. Automate the comparison and review exceptions regularly. Typical causes of mismatch: - Different SKU codes or units of measure. - Reserved or allocated stock counted differently. - Returns processed late. - Sync delays between the shop and warehouse system. - Bundles and kits that decompose differently. Build a SKU mapping table first. Then snapshot stock daily so you can see when and where differences began. Fix the process that creates the difference, such as delayed receipting, not only the report. URL: https://zansa.co.za/answers/how-do-i-reconcile-stock-levels-between-my-erp-warehouse-and-online-shop ### How do I match bank transactions to invoices automatically? Match on reference numbers where available, then on amount and date proximity, and handle partial payments, batch payments and fees separately. Flag uncertain matches for human review rather than forcing them. Many accounting tools offer bank rules, and custom matching can be built in SQL. Matching tiers: - Exact: invoice reference appears in the bank description. - Strong: amount equals invoice total and date is within a few days. - Grouped: one payment covers several invoices, or the reverse. - Adjusted: payment net of fees or withholding. Keep a log of matches and rules. Always review unmatched items, as they often reveal real issues such as misapplied payments or duplicates. Aim to automate the routine majority, not every edge case. URL: https://zansa.co.za/answers/how-do-i-match-bank-transactions-to-invoices-automatically ### What does a good data refresh failure alert look like? A good alert says which pipeline failed, when, what the error was, which dashboards are affected, and who should act. It goes to a channel people actually watch, avoids noise from minor issues, and links to instructions for fixing common problems. Include: - Pipeline or table name and time of failure. - Error message or log link. - Impact: which reports show stale data. - Owner and severity. Avoid alert fatigue by separating critical failures from warnings, and by grouping repeated alerts. Also alert on silent failures, such as no new rows arriving, because a job that runs successfully but loads nothing is as harmful as one that crashes. URL: https://zansa.co.za/answers/what-does-a-good-data-refresh-failure-alert-look-like ### How do I test whether a dashboard number is correct? Trace it back: compare the dashboard figure with the same query run directly on the data, then with the source system's own report for the same period and filters. Check a sample of individual records and confirm any differences are explained by documented rules. Verification steps: - Confirm filters, date range and currency match. - Compare totals at several levels, such as month, product and customer. - Pick a few specific transactions and follow them through each stage. - Look for duplicated or dropped rows after joins. - Ask a subject-matter expert if the magnitude seems plausible. After fixing, add an automated test for that check so the same error cannot return unnoticed. URL: https://zansa.co.za/answers/how-do-i-test-whether-a-dashboard-number-is-correct ### What is the difference between a metric, a dimension and a measure? A metric or measure is a number you calculate, such as revenue or order count. A dimension is a category you group or filter by, such as product, region or month. Dashboards combine them, for example revenue by region by month. Terminology varies slightly between tools. Some use measure for the raw aggregation, such as a sum of amounts, and metric for the business-defined figure, such as gross margin percentage. What matters is that each metric has a written definition covering inclusion rules, date basis and calculation, and that it is implemented once. Dimensions should have consistent, clean values, since misspelt categories split totals across duplicate groups. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-metric-a-dimension-and-a-measure ### How should I name tables and columns in a data warehouse? Use consistent, descriptive, lowercase names with underscores, avoid abbreviations that outsiders will not understand, and follow a pattern that signals purpose, such as stg_ for staging and fct_ or dim_ for facts and dimensions. Consistency matters more than any particular style. Helpful conventions: - Singular or plural table names, but pick one. - Suffixes such as _id for keys, _at for timestamps, _date for dates. - Include units or currency in names when ambiguous, such as amount_usd. - Avoid spaces and special characters. - Never reuse a name with a different meaning. Write the conventions down. Clear naming makes self-service BI easier and cuts the questions people have to ask the data team. URL: https://zansa.co.za/answers/how-should-i-name-tables-and-columns-in-a-data-warehouse ### How can version control help with analytics code and dashboards? Version control, usually Git, records every change to SQL models and scripts, who made it and why, and lets you review changes before they go live and roll back mistakes. It makes analytics code auditable and enables several people to work safely together. Benefits: - Peer review of changes before deployment. - A clear history when a number suddenly changes. - Easy recovery from errors. - A way to separate development and production. Tools like dbt are designed around Git. Dashboards are harder to version, since many BI tools store them internally, so some teams export definitions or document them separately. At minimum, keep transformation logic in version control rather than only inside a BI tool. URL: https://zansa.co.za/answers/how-can-version-control-help-with-analytics-code-and-dashboards ### How do I handle missing values in business data? First understand why values are missing, since blank can mean unknown, not applicable or zero. Then choose a documented treatment: leave as null, fill with a sensible default, estimate, or exclude, and never silently turn missing into zero where that changes meaning. Options: - Fix at source: make the field required. - Fill using reliable rules, such as a default category. - Impute statistically for modelling, with the assumption stated. - Exclude rows from specific calculations and show how many were excluded. Measure missingness with data tests and track it over time. A sudden rise often signals a broken integration or process change rather than a real shift in the business. URL: https://zansa.co.za/answers/how-do-i-handle-missing-values-in-business-data ### How do I standardise product names and categories across systems? Build a product master table with a unique internal ID, and map each system's product codes and names to it. Define one category hierarchy owned by the business, and apply it through the mapping instead of relying on each system's own labels. Steps: - Export product lists from every system. - Match on SKU, barcode or supplier code first. - Resolve leftovers manually, and save the decisions in the mapping table. - Define standard attributes: name, category, brand, unit of measure. - Assign someone to maintain it when new products are added. Without this, category totals differ between shop, till and ERP reports, and trends are distorted by relabelling. URL: https://zansa.co.za/answers/how-do-i-standardise-product-names-and-categories-across-systems ### Why do totals change after I refresh my report? Totals change when source data is backdated, edited, deleted or loaded late, or when report logic or filters changed. Accounting entries posted to earlier periods and delayed card settlements are common causes. A change history and snapshots help explain what moved. Look for: - Late-arriving transactions assigned to past dates. - Credit notes or journals posted to prior months. - Incremental loads that missed updates. - Changes to mapping tables or definitions. - Time zone or cut-off differences. For reporting that must stay fixed, such as month-end packs, take a snapshot when the period closes. For live views, show a last-updated time and note that recent periods may still change. URL: https://zansa.co.za/answers/why-do-totals-change-after-i-refresh-my-report ### How do I explain data discrepancies to non-technical managers? Show a simple bridge from one number to the other, listing each reason for the difference with an amount, such as timing, fees, refunds and tax. Use plain language, avoid jargon, and make clear which figure is right for which purpose. A reconciliation table works well: - Start with the figure from system A. - Add or subtract each explained item. - Arrive at the figure from system B. - Show any remaining unexplained amount. Then state which number to use for which decision, for example the ledger for statutory reporting and the shop platform for operational sales. Managers lose trust when numbers vary without explanation, so documenting the bridge once and reusing it saves repeated debate. URL: https://zansa.co.za/answers/how-do-i-explain-data-discrepancies-to-non-technical-managers ### What should a Data Health Check for a small business cover? A Data Health Check should review which systems and spreadsheets hold your data, how they connect, where numbers do not match, data quality problems, key metric definitions, security and privacy risks, and what is realistically worth fixing first, ending in a prioritised plan. Typical areas: - Inventory of sources and who owns them. - Sample reconciliations between systems. - Data quality profiling: duplicates, gaps, formatting. - Review of existing reports and manual processes. - Access, backups and compliance considerations. - Recommendations with effort and value estimates. The outcome should be a clear, written view of the current state and options, so you can decide on investment with evidence rather than guesses. URL: https://zansa.co.za/answers/what-should-a-data-health-check-for-a-small-business-cover ### What does Zansa's fixed-fee Data Health Check involve for a small business? Zansa's Data Health Check is a fixed-fee audit, typically taking one to two weeks, of your systems, spreadsheets and the numbers that do not match. It follows a free intro call, and it produces a clear picture of what is wrong, why, and what to fix first. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The Health Check is the second step after the intro call and before any larger build. Guidance pricing for Health Checks starts from R25,000 (ZAR), and the exact fee is agreed before work begins. It suits companies where reports disagree, spreadsheets are fragile, or it is unclear which system to trust. Afterwards you can choose to proceed with a Foundation Build or not. URL: https://zansa.co.za/answers/what-does-zansa-s-fixed-fee-data-health-check-involve-for-a-small-business ### What is a Foundation Build in a data project? A Foundation Build is the phase that creates the core data setup: connecting your sources, loading them into a central warehouse, modelling the data, building first dashboards and documenting everything. At Zansa it typically takes four to eight weeks, with fixed price and timeline agreed in writing. At Zansa, the build follows the Data Health Check. Data lands daily in a cloud account owned by the client, usually Google BigQuery, and dashboards are typically in Metabase or Power BI. The client owns everything built, including code, data models, dashboards and documentation. Guidance pricing for builds is typically R80,000 to R250,000 (ZAR), depending on scope. Optional ongoing support can follow, from R12,000 per month. Exact scope and cost are confirmed in writing before starting. URL: https://zansa.co.za/answers/what-is-a-foundation-build-in-a-data-project ### Who owns the data models and dashboards when an agency builds them? Ownership depends on the contract, so agree it in writing before work starts. Ideally you own the cloud account, code, data models, dashboards and documentation, so you can switch provider or bring the work in-house without losing anything or paying to regain access. Warning signs of lock-in: - Data stored in the agency's own cloud account. - Dashboards built in tools licensed to the agency. - No documentation or access to code. - Ongoing fees required just to retain access. Ask for admin access to accounts, source code in a repository you control, and written documentation. At Zansa, data lives in a cloud account owned by the client and the client owns all that is built, which keeps switching and handover simple. URL: https://zansa.co.za/answers/who-owns-the-data-models-and-dashboards-when-an-agency-builds-them ### Why does Zansa usually use Metabase for dashboards? Zansa usually builds dashboards in Metabase because it has no per-user licence fees, which suits small and mid-sized companies wanting many staff to see data. It is easy for non-technical people to explore, and the data models underneath stay tool-independent. Zansa also builds in Power BI when the team already uses it. The data models underneath are the same either way, so the choice of front end is about what your team prefers and already pays for, not about the quality of the data foundation. Because the client owns the cloud data account and everything built, switching dashboard tools later does not mean starting again. Which is better for you depends on user numbers, existing Microsoft use and how you plan to share data. URL: https://zansa.co.za/answers/why-does-zansa-usually-use-metabase-for-dashboards ### Which data sources can Zansa connect for a small company? Zansa works with Excel and CSV files, Xero, Sage, Shopify and SQL databases, loading them daily into a central warehouse owned by the client. Other sources may be possible, and compatibility is confirmed during the Data Health Check. Whether a given system can connect usually depends on if it offers an API or file exports. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. It serves sectors including retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. Whether a particular system can be connected depends on what access it offers, such as an API or exports, so the audit step checks this up front. Data is typically stored in Google BigQuery in the client's own account. URL: https://zansa.co.za/answers/which-data-sources-can-zansa-connect-for-a-small-company ### How does Zansa handle privacy laws such as POPIA and GDPR? Zansa designs its work with privacy law in mind, including POPIA and GDPR. Data lives in a cloud account owned by the client, and AI tools used to help write code never receive personal or client-identifiable data. Any machine learning models run inside the client's own account. Keeping data in the client's account means the client controls access, location and retention. Zansa is based in South Africa and works remotely with companies around the world. This summary is not legal advice. Your obligations depend on the data you hold and where your customers are, so you should confirm compliance requirements with a qualified adviser. A good data setup helps by limiting who can see personal data and by documenting where it flows. URL: https://zansa.co.za/answers/how-does-zansa-handle-privacy-laws-such-as-popia-and-gdpr ### Does Zansa use AI tools, and what happens to my data? Zansa uses AI tools to help write code, but these tools never receive personal or client-identifiable data. Any machine learning models are run inside the client's own cloud account, so your data does not leave your environment for modelling purposes. This separation matters for privacy laws such as POPIA and GDPR, and for commercial confidentiality. Code assistance works from schemas, descriptions and synthetic or non-identifying examples rather than real customer records. When evaluating any provider that uses AI, ask what data is shared with third-party tools, where it is processed, whether it is retained, and who owns the outputs. Get the answers in writing as part of your contract. URL: https://zansa.co.za/answers/does-zansa-use-ai-tools-and-what-happens-to-my-data ### How much does it cost to set up a data warehouse and dashboards for a small business? Costs vary widely with scope, number of sources and complexity. As a guide, Zansa prices Health Checks from R25,000 and builds typically between R80,000 and R250,000 (ZAR), with optional support from R12,000 per month. Other providers will differ, so compare scope carefully. Cost drivers: - Number and quality of data sources. - Complexity of business logic and reconciliation. - Number of dashboards and users. - Ongoing support needs. - Cloud running costs, which are separate and depend on usage. Ask for a fixed price and timeline in writing, and clarify who owns the outputs. Beware of low quotes that exclude data cleaning, which is often the largest part of the work. URL: https://zansa.co.za/answers/how-much-does-it-cost-to-set-up-a-data-warehouse-and-dashboards-for-a-small-busi ### What is the typical process for starting a data project with Zansa? Zansa starts with a free intro call, then a fixed-fee Data Health Check lasting one to two weeks, then a Foundation Build typically taking four to eight weeks with fixed price and timeline agreed in writing, followed by optional ongoing support. Each stage is a decision point, so you can stop after the audit if it does not make sense to continue. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The name comes from the Japanese 残差 (zansa), meaning residual: the part left over that the usual explanations do not account for, which is a fitting image for numbers that do not match. URL: https://zansa.co.za/answers/what-is-the-typical-process-for-starting-a-data-project-with-zansa ### What does the name Zansa mean and where does it come from? Zansa comes from the Japanese word 残差 (zansa), which means residual. In statistics, a residual is the difference between an observed value and what was expected, which fits a team focused on explaining numbers that do not match. Zansa applies that idea to small and mid-sized companies whose systems disagree. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. A residual is the leftover amount after you account for everything you can explain. In reconciliation work, the aim is the same: explain each difference between systems until only a small, understood remainder is left. URL: https://zansa.co.za/answers/what-does-the-name-zansa-mean-and-where-does-it-come-from ### Which industries benefit most from a unified data model? Any business with data in several systems benefits, but the payoff is clearest where operations generate lots of transactions, such as retail and e-commerce, manufacturing, logistics, healthcare, property, hospitality and professional services. Each has recurring needs around stock, time, utilisation, occupancy or margin. Examples of why it helps: - Retail and e-commerce: unify shop, till, stock and accounting. - Manufacturing: link production, materials and costing. - Logistics: tie routes, fuel and delivery performance to revenue. - Property: combine rent rolls, maintenance and accounts. - Hospitality: link bookings, covers and costs. - Professional services: connect time, projects and billing. Zansa works across these sectors for small and mid-sized companies. The common thread is multiple systems that each tell part of the story. URL: https://zansa.co.za/answers/which-industries-benefit-most-from-a-unified-data-model ### How can a retailer combine online and in-store sales into one view? Load sales from both channels into a central database, map products and customers to common IDs, align dates and currency, and model a single sales fact table with a channel field. Then reconcile each channel to payments and accounting totals. Key steps: - Product mapping: shop SKUs and till codes need a common product master. - Treat returns consistently across channels. - Align tax, discounts and gift card handling. - Standardise time zones and trading-day cut-offs. - Decide how to identify customers across channels, with consent and privacy in mind. Once unified, you can compare channel margins, stock turn and promotions. Zansa works with Shopify, Excel and CSV exports and Xero for retail and e-commerce clients, among others. URL: https://zansa.co.za/answers/how-can-a-retailer-combine-online-and-in-store-sales-into-one-view ### How do I measure true profit when revenue and costs sit in different systems? Bring sales, product costs, fees, shipping, refunds and overheads into one model, using a consistent period and allocation rules, then calculate margin at the level you need, such as product, customer or channel. Document each allocation rule so finance can validate it. Common data sources: - Revenue from shop or invoicing. - Cost of goods from ERP, supplier invoices or accounting. - Payment and marketplace fees from payouts. - Shipping costs from carriers. - Overheads from the ledger. Decide how to allocate shared costs and keep rules simple and consistent. Compare the total of your model with the accounting profit and loss and explain differences. Margin by product or customer is often where surprises appear. URL: https://zansa.co.za/answers/how-do-i-measure-true-profit-when-revenue-and-costs-sit-in-different-systems ### Why is my Excel pivot table giving different totals from my accounting report? Usually because of different filters, dates, tax treatment, duplicated rows, text-formatted numbers, hidden rows or stale data. Check that the date range, account selection and tax basis match, then compare row counts and look for duplicates or missing lines in the export. Checklist: - Does the export include all accounts and statuses, such as drafts and voided items? - Are credit notes shown as negatives? - Are numbers stored as text and being skipped? - Is the pivot refreshed, and does its source range include all rows? - Is the report cash or accrual basis? Break totals down by month and account to find where the gap begins. Then fix the cause rather than adjusting with a manual plug figure. URL: https://zansa.co.za/answers/why-is-my-excel-pivot-table-giving-different-totals-from-my-accounting-report ### What is a data gateway and when do I need one? A data gateway is a piece of software that securely connects a cloud BI service to data held on your own network, such as an on-premises database. You need one when a cloud tool must refresh from sources that are not reachable from the internet. Power BI, for example, uses an on-premises data gateway for scheduled refresh from internal sources. Considerations: - It must be installed on an always-on machine. - If that machine is off, refreshes fail. - It needs updates and monitoring. An alternative is to copy on-premises data to a cloud warehouse on a schedule and connect the BI tool to that, which often simplifies refresh and security and removes dependence on a single office machine. URL: https://zansa.co.za/answers/what-is-a-data-gateway-and-when-do-i-need-one ### How do I decide between building integrations myself and buying a connector tool? Buy a connector tool when sources are common, the vendor supports them well, and you value low maintenance. Build custom pipelines when connectors are missing, too costly at your volume, or you need control. Weigh licence cost against the engineering time to maintain APIs. Factors: - Number of sources and how standard they are. - Data volume, since some connector tools price by rows or usage. - Reliability and support when APIs change. - Customisation needs. - Internal skills to maintain code. A hybrid is common: managed connectors for popular systems and small custom scripts for the rest. Whichever route, keep raw data in your own warehouse so you are not locked into a single tool. URL: https://zansa.co.za/answers/how-do-i-decide-between-building-integrations-myself-and-buying-a-connector-tool ### Is Airbyte, Fivetran or a custom script better for loading data? It depends on your sources, budget and skills. Managed services like Fivetran reduce maintenance but are typically priced by usage. Airbyte is open-source and offers a managed option, with more flexibility. Custom scripts offer total control but need ongoing upkeep. Questions to ask: - Does the tool support all your sources and the specific objects you need? - How is it priced, and how does cost scale with your data? - Who maintains it when a source API changes? - Can you self-host if data residency matters? Check current features and pricing, since they change. For a handful of common sources and small volumes, either a connector service or a few well-written scripts can work. Prioritise monitoring and clear ownership over the specific tool. URL: https://zansa.co.za/answers/is-airbyte-fivetran-or-a-custom-script-better-for-loading-data ### How should I schedule and orchestrate data jobs? Use a scheduler that runs jobs in the right order, retries failures and alerts you. Simple needs can be met with cron-style schedules or the cloud provider's scheduler, while more complex dependencies suit orchestration tools such as Airflow or Dagster. Core needs: - Dependencies: transformations should run only after loads finish. - Retries with limits for temporary failures. - Logging and alerts. - Idempotency, meaning rerunning a job does not duplicate data. - Sensible timing, such as after the source system closes its day. For a small number of daily jobs, a simple scheduler is enough. Add heavier orchestration only when dependencies become hard to manage by hand. URL: https://zansa.co.za/answers/how-should-i-schedule-and-orchestrate-data-jobs ### How do I deal with duplicated rows after loading data? Identify the cause first: repeated loads, joins that multiply rows, or duplicates in the source. Then fix it at the right layer by using unique keys, upserts or deduplication logic, and add a uniqueness test so it is caught automatically in future. Diagnosis steps: - Count rows by key and find keys with more than one row. - Check load logs for repeated runs. - Examine joins for many-to-many relationships. Fixes: - Make loads idempotent. - Deduplicate using a rule, such as keeping the latest record per ID by timestamp. - Correct the source if duplicates are entered by staff. Never just delete duplicates in a report. Without fixing the cause, they return on the next refresh. URL: https://zansa.co.za/answers/how-do-i-deal-with-duplicated-rows-after-loading-data ### What are NULLs and why do they break calculations? NULL means a value is unknown or missing, not zero or blank. In SQL, calculations involving NULL usually return NULL, and comparisons with NULL do not behave like normal equality, so totals, filters and joins can silently drop or distort rows. Common effects: - Adding a NULL to a number gives NULL. - Aggregates like SUM and AVG ignore NULLs, which can change averages. - WHERE column = value excludes NULL rows; you need IS NULL. - Joins on NULL keys do not match. Handle them deliberately using COALESCE where a default is appropriate, and use tests to flag unexpected NULLs. Decide per field whether missing should be treated as zero, unknown or an error. URL: https://zansa.co.za/answers/what-are-nulls-and-why-do-they-break-calculations ### How do I give clients or franchisees their own dashboard securely? Use a BI setup that supports row-level security or per-tenant filtering, ideally enforced in the database, so each external user sees only their own data. Test thoroughly, and confirm licensing covers external users, since that often differs from internal rules. Options: - Separate dashboards per client, simple but laborious to maintain. - One dashboard with user-based filters or signed embedding tokens. - A separate dataset per client for strict isolation. Safeguards: - Do not rely on hidden front-end filters alone. - Test by trying to alter URL parameters. - Log access. - Check the privacy implications if personal data is involved. Plan for how new clients are onboarded so permissions are set up consistently. URL: https://zansa.co.za/answers/how-do-i-give-clients-or-franchisees-their-own-dashboard-securely ### What is the difference between a report and a dashboard? A dashboard gives a quick, interactive overview of key metrics for monitoring and decisions, while a report is usually a more detailed, structured document or table for a specific period or purpose. Dashboards are for scanning, reports are for reading and record-keeping. Dashboards: live or frequently refreshed, visual, filterable, focused on a few key questions. Reports: often static or scheduled, detailed, formatted for printing or sharing, such as a monthly management pack. Many businesses need both. A good pattern is a dashboard for daily monitoring that links to detailed tables for investigation, and a fixed monthly report produced from the same underlying data so the numbers agree. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-report-and-a-dashboard ### How do I measure whether a BI project is delivering value? Tie it to specific outcomes before you start, such as hours saved on reporting, fewer disputes over numbers, faster month-end close, or decisions made with data, then check these after launch. Dashboard views alone do not show value unless they lead to action. Ways to assess: - Time spent producing recurring reports before and after. - Number of data corrections or reconciliations needed. - Speed to answer ad hoc questions. - Decisions or actions traceable to insights, such as pricing, stock or staffing changes. Avoid inventing precise return figures. Use your own baseline measurements and honest estimates. If a dashboard is never used, find out why, because the issue may be relevance or trust rather than technology. URL: https://zansa.co.za/answers/how-do-i-measure-whether-a-bi-project-is-delivering-value ### How do I move from spreadsheets to a proper data setup without disrupting the business? Do it in stages: inventory current spreadsheets, load their sources into a database alongside existing processes, rebuild one important report on the new setup, reconcile it with the old one, and then retire the spreadsheet only when users trust the new version. Staged approach: - Identify the reports that matter most and cause most pain. - Automate data collection first, keeping familiar outputs. - Run old and new in parallel for a period. - Document definitions as you go. - Train users and keep Excel access to the new data where helpful. Zansa's process follows a similar logic: a free intro call, a Data Health Check, then a Foundation Build, so the move is planned and the business keeps running throughout. URL: https://zansa.co.za/answers/how-do-i-move-from-spreadsheets-to-a-proper-data-setup-without-disrupting-the-bu ### Can I use BigQuery with Power BI and Metabase at the same time? Yes. BigQuery is a standard SQL warehouse, so multiple BI tools can connect to the same modelled tables simultaneously. This is a strength of keeping logic in the warehouse: different teams can use different front ends while seeing identical numbers. Considerations: - Authentication: use service accounts or user credentials with appropriate roles. - Cost: each tool's queries consume BigQuery resources, so monitor usage. - Consistency: define metrics in warehouse models, not separately in each tool. Zansa typically builds the data models in the client's BigQuery account and then delivers dashboards in Metabase or Power BI. Because the underlying models are the same, the front-end choice can change later without redoing the foundations. URL: https://zansa.co.za/answers/can-i-use-bigquery-with-power-bi-and-metabase-at-the-same-time ### What does a data engineer do compared with a data analyst? A data engineer builds and maintains the pipelines, storage and models that make data reliable and available. A data analyst uses that data to answer business questions, build dashboards and explain results. Small companies often need both skills, sometimes in one person or an external team. Data engineer: extraction, loading, transformation, orchestration, testing, performance, security. Data analyst: querying, visualisation, interpretation, recommending actions. Data scientist: statistical modelling and machine learning for prediction or classification. When the foundation is weak, analysts spend most of their time cleaning and reconciling. Investing in solid data engineering first usually makes later analysis faster and more trustworthy. Zansa combines analytics and data science work for small and mid-sized companies. URL: https://zansa.co.za/answers/what-does-a-data-engineer-do-compared-with-a-data-analyst ### Do small companies need machine learning, or is BI enough? Most small companies get far more value from reliable BI first. Machine learning needs clean, consistent historical data and a clear question, such as forecasting demand or predicting churn. Without a solid data foundation, models tend to be unreliable and hard to trust. Sensible order: - Get data integrated and trustworthy. - Build dashboards that reveal current performance. - Then consider forecasting or classification where a specific decision would benefit. Models also need monitoring, because patterns change. Where personal data is involved, run models in a secure environment you control. Zansa runs ML models inside the client's own cloud account, but recommends the data foundation come first. URL: https://zansa.co.za/answers/do-small-companies-need-machine-learning-or-is-bi-enough ### What happens to my dashboards if the person who built them leaves? Without documentation and shared access, they may become unmaintainable and eventually untrusted. Reduce the risk by keeping logic in version control, using shared accounts and ownership, documenting metrics and refresh processes, and avoiding hidden formulas in personal files or laptops. Resilience checklist: - Data and dashboards in organisation-owned accounts, not personal ones. - Credentials in a shared secrets store. - Written documentation of sources, models and schedules. - At least two people who understand the setup. - A handover process for leavers. This applies equally to staff and external providers. Insist that all code, models, dashboards and documentation are yours, as Zansa's approach provides, so continuity never depends on one supplier or individual. URL: https://zansa.co.za/answers/what-happens-to-my-dashboards-if-the-person-who-built-them-leaves ## Logistics ### How do I calculate on-time delivery rate in logistics? On-time delivery rate is the number of deliveries completed within the agreed delivery window divided by total deliveries in the period, multiplied by 100. Deliveries arriving early or late outside the window count as not on time, unless your agreement says otherwise. It matters because it is the main promise a logistics customer experiences. Commonly misread: - Measuring against the planned time set by your own dispatchers rather than the customer's requested window. - Using the time the vehicle arrived instead of the time the goods were handed over. - Hiding customer-caused delays by excluding them without a clear rule. Record the reason and report both views. - Using a single average across all customers when contracts differ. Decide the clock-start, clock-stop and window rules in writing and apply them to every delivery. URL: https://zansa.co.za/answers/how-do-i-calculate-on-time-delivery-rate-in-logistics ### What is the difference between on-time delivery and OTIF? On-time delivery measures only whether goods arrived within the agreed window. OTIF, on time in full, also requires that the complete ordered quantity arrived. A delivery counts as OTIF only if it passes both tests, so OTIF is never higher than on-time delivery. Formula in words: deliveries both on time and complete, divided by total deliveries. It matters because a load that arrives on time but short still causes the customer a problem, and may trigger penalties or disputes. Commonly misread: - Customers define in full differently, such as by line, case or order value. Agree it in advance. - Treating damaged goods as in full. - Mixing OTIF with on-time delivery in reports, so figures cannot be compared. Track the reason for each failure to see whether late or short deliveries are the larger problem. URL: https://zansa.co.za/answers/what-is-the-difference-between-on-time-delivery-and-otif ### How do I calculate cost per drop for a delivery operation? Cost per drop is the total cost of running a route or fleet in a period divided by the number of drops delivered. Total cost includes fuel, driver wages, vehicle finance or depreciation, maintenance, tyres, tolls, insurance and a share of overhead. Example: a route costing 12,000 currency units in a week with 400 drops gives 30 per drop. It matters because pricing, route design and customer profitability depend on it. Commonly misread: - Counting only fuel and driver cost, so the figure looks lower than it is. - Treating all drops as equal; a drop of one parcel and a drop of ten pallets have different costs. - Ignoring failed deliveries and returns trips, which add cost without a completed drop. Track it by route, vehicle and customer, and consider weight or volume as a second measure. URL: https://zansa.co.za/answers/how-do-i-calculate-cost-per-drop-for-a-delivery-operation ### How do I measure fuel consumption per kilometre for a vehicle fleet? Fuel consumption is litres of fuel used divided by kilometres driven, often shown as litres per 100 km. Use fuel purchase or tank records and odometer or tracking distance for the same period, and compare each vehicle against its own history. The inverse, kilometres per litre, is also common. Pick one and stay consistent. It matters because fuel is a large and controllable fleet cost, and sudden changes can indicate mechanical faults, driver behaviour or fuel loss. Commonly misread: - Using fill-ups that do not match the distance period, as partial fills distort individual figures. Use totals over a longer window. - Comparing loaded and empty vehicles or different vehicle types. - Ignoring idling and stationary engine hours. Match fuel card data to tracking data to check each fill against vehicle location. URL: https://zansa.co.za/answers/how-do-i-measure-fuel-consumption-per-kilometre-for-a-vehicle-fleet ### How do I calculate vehicle utilisation in a transport business? Vehicle utilisation is the time or distance a vehicle spends productively on revenue work divided by the time or distance it was available, multiplied by 100. It can also be measured as load carried divided by capacity. State which basis you use. It matters because trucks cost money whether or not they are earning: finance, insurance and driver time continue when parked. Commonly misread: - A vehicle that is moving is not necessarily productive. Empty running counts as movement but not revenue. - Combining time-based and load-based figures. - Counting maintenance days as available time without separating them out. Use three separate views: time utilisation, distance with revenue, and load utilisation. A high figure on one with a low figure on another shows where the loss is. URL: https://zansa.co.za/answers/how-do-i-calculate-vehicle-utilisation-in-a-transport-business ### How do I work out route profitability for a delivery or haulage business? Route profitability is the revenue earned on a route minus all costs of running it, over the same period. Costs include fuel, driver time, vehicle costs, tolls, maintenance and a fair share of overhead. Dividing the profit by revenue gives a route margin. It matters because some routes quietly lose money while the fleet overall appears profitable. Commonly misread: - Using revenue per route without matching the actual costs of the vehicles and drivers used. - Leaving out the return leg when it runs empty. - Ignoring waiting time at customer sites. - Spreading overhead evenly, which can make short routes look worse or better than they are. You need billing data matched to tracking and fuel data per trip. Review route profit monthly and check whether poor routes can be repriced, re-sequenced or dropped. URL: https://zansa.co.za/answers/how-do-i-work-out-route-profitability-for-a-delivery-or-haulage-business ### What are unbilled loads and how do I find them? Unbilled loads are deliveries that have been completed but not yet invoiced. Find them by comparing completed trips or proof-of-delivery records against invoices issued, and listing every trip that has no matching invoice line. It matters because every unbilled load is earned revenue not yet collected, and old ones risk never being billed. Common causes: - Missing or late proof-of-delivery documents. - Extra charges, such as waiting time or tolls, never passed on. - Trips arranged informally, outside the billing system. - Rate disputes holding up invoices. Measure the count, value and age of unbilled loads weekly, and report days from delivery to invoice. A reconciliation key such as trip or waybill number is needed on both sides. URL: https://zansa.co.za/answers/what-are-unbilled-loads-and-how-do-i-find-them ### What is empty running or deadhead in logistics and how is it measured? Empty running, or deadhead, is distance driven without a load. Measure it as empty kilometres divided by total kilometres, multiplied by 100. It shows how much driving earns no revenue but still consumes fuel, time and wear. It matters because return legs and positioning moves can quietly take a large share of cost. Commonly misread: - Some empty running is unavoidable, such as returning from a one-way delivery. The aim is to reduce the avoidable part. - Counting a partly loaded return as fully loaded or fully empty. - Measuring by distance only, when empty time waiting is also a loss. Capture load status in tracking or trip records. Then look at backload opportunities, route clustering and where pick-ups can be combined to reduce empty kilometres. URL: https://zansa.co.za/answers/what-is-empty-running-or-deadhead-in-logistics-and-how-is-it-measured ### What is cost per kilometre and how should a fleet use it? Cost per kilometre is the total running cost of a vehicle or fleet in a period divided by kilometres driven. Include fuel, driver cost, maintenance, tyres, insurance, finance or depreciation, and tolls, not just fuel. It matters as a baseline for quoting and for spotting vehicles that cost more than similar ones. Commonly misread: - Using only variable costs, which understates the cost of idle or low-mileage vehicles. - Comparing different vehicle types. - Treating a falling cost per km as success when the fleet is simply driving further with empty loads. Read it with cost per drop and revenue per kilometre. If cost per km is low but cost per drop is high, vehicles may be travelling long distances for few deliveries. URL: https://zansa.co.za/answers/what-is-cost-per-kilometre-and-how-should-a-fleet-use-it ### How do I measure first-attempt delivery success rate for a courier fleet? First-attempt delivery success is the number of deliveries completed on the first attempt divided by total deliveries attempted, multiplied by 100. It shows how often a driver can hand over goods without a failed visit or return trip. It matters because every failed attempt adds cost: extra kilometres, time, re-scheduling and sometimes customer dissatisfaction. Commonly misread: - Counting a delivery to a neighbour or safe place as the same as a signed handover. - Not recording the failure reason, so you cannot tell whether the cause is address quality, customer absence or access problems. Capture reason codes such as no access, customer not available, wrong address and refused. Cost per successful drop includes the cost of failed attempts, so track both measures together. URL: https://zansa.co.za/answers/how-do-i-measure-first-attempt-delivery-success-rate-for-a-courier-fleet ### What is load factor or fill rate for a transport fleet? Load factor is the load actually carried divided by the vehicle's capacity, by weight, volume or pallet positions, multiplied by 100. A truck carrying 8 tonnes on a 10 tonne capacity has an 80 percent load factor. It matters because carrying air or partly filled loads raises cost per unit moved. Commonly misread: - A vehicle may be full by volume but under weight limit, or the other way round. State which constraint you measure. - Averaging outbound and return legs hides the empty return. - Chasing 100 percent can cause overloading, which is illegal and unsafe. Use load factor alongside cost per drop and empty running. Together they show whether you can combine loads, change vehicle sizes or reprice small shipments. URL: https://zansa.co.za/answers/what-is-load-factor-or-fill-rate-for-a-transport-fleet ### How do maintenance costs per kilometre help manage a delivery fleet? Maintenance cost per kilometre is the total repair, service and tyre cost of a vehicle divided by kilometres driven. It helps identify vehicles that cost more to keep running than they should, and supports replacement decisions. Compare each vehicle with similar ones and with its own history. A rising trend as a vehicle ages is normal, but a sharp rise points to a specific fault, driver issue or poor maintenance. Commonly misread: - Using a short period that includes a one-off major repair. - Leaving out the cost of downtime and replacement hire vehicles. - Comparing vehicles used on gravel roads with those on highways. Combine it with availability, since a cheap-to-maintain vehicle that is often off the road still costs money in missed deliveries. URL: https://zansa.co.za/answers/how-do-maintenance-costs-per-kilometre-help-manage-a-delivery-fleet ### What is a daily exceptions list in logistics operations? A daily exceptions list is a short report each morning of the items that need action: late or failed deliveries, loads delivered but not invoiced, vehicles with abnormal fuel use, and trips with missing documents. It shows problems rather than totals. It matters because averages hide the individual cases that need someone to act today, such as an invoice about to age or a customer with repeated late drops. A good list is: - Short and prioritised by value or urgency. - Owned, with each item assigned to a person. - Cleared daily, so old items do not pile up. Zansa builds a daily exceptions list as part of its logistics dashboards, drawn from tracking, fuel and billing data. URL: https://zansa.co.za/answers/what-is-a-daily-exceptions-list-in-logistics-operations ### What does Zansa's logistics dashboard for small transport companies actually include? Zansa's logistics dashboard combines tracking, fuel and billing data to show on-time delivery and cost per drop by route, plus a daily exceptions list. It is designed to show which routes and customers earn money and which problems need action today. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process is a free intro call, a fixed-fee Data Health Check of one to two weeks, and then a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Support is optional afterwards. Dashboards are usually in Metabase or Power BI, on a cloud account owned by the client, usually Google BigQuery. The client owns everything built. Guidance pricing is Health Checks from R25,000 and builds typically R80,000 to R250,000. URL: https://zansa.co.za/answers/what-does-zansa-s-logistics-dashboard-for-small-transport-companies-actually-inc ## Manufacturing ### What is OEE in manufacturing and how is it calculated? OEE, overall equipment effectiveness, is availability multiplied by performance multiplied by quality. Availability is run time divided by planned production time, performance is actual output divided by ideal output at rated speed, and quality is good units divided by total units. Example: 90 percent availability, 95 percent performance and 98 percent quality gives 0.90 x 0.95 x 0.98, about 83.8 percent. It matters because it shows how much of your scheduled time produces good product at full speed, and splits losses into stoppages, slow running and defects. Commonly misread: - Comparing OEE between machines or plants that define planned time or ideal speed differently. - Chasing one number and ignoring which of the three losses is the largest. - Using inflated ideal speeds, which makes performance look worse than reality. Fix your definitions first, then compare each line only against its own history. URL: https://zansa.co.za/answers/what-is-oee-in-manufacturing-and-how-is-it-calculated ### What are the three components of OEE and how do I find the biggest loss? The three components are availability (lost to stoppages), performance (lost to slow running and small stops) and quality (lost to scrap and rework). Calculate each as a percentage; the component furthest from 100 percent is usually where the biggest improvement lies. Formulas in words: - Availability: run time divided by planned production time. - Performance: (ideal cycle time x total units) divided by run time. - Quality: good units divided by total units. It matters because the total OEE number says little about where to act. Two lines with the same OEE can need entirely different fixes. Commonly misread: - Treating a low performance figure as an operator problem when it comes from an unrealistic ideal rate. - Missing micro-stops, which are rarely recorded by hand. Break losses down by reason code and rank them by lost time. URL: https://zansa.co.za/answers/what-are-the-three-components-of-oee-and-how-do-i-find-the-biggest-loss ### What is first pass yield and how does it differ from final yield? First pass yield (FPY) is the percentage of units that are made correctly the first time, with no rework or repair. Divide good units that passed first time by total units started. Final yield counts units that were eventually good, including reworked ones. Example: 1,000 units started, 900 pass first time, 60 are reworked and then pass, 40 are scrapped. FPY is 90 percent, while final yield is 96 percent. It matters because rework costs labour, time and capacity that final yield hides. Commonly misread: - Reporting only final yield and believing the process is healthier than it is. - Multiplying yields across process steps without checking that each step uses the same definition. - Counting units at different points in the process. Track both figures and the gap between them. URL: https://zansa.co.za/answers/what-is-first-pass-yield-and-how-does-it-differ-from-final-yield ### How do I calculate scrap rate and put a cost on it? Scrap rate is the units (or material weight) scrapped divided by total units (or material) produced or put into the process, multiplied by 100. To cost it, multiply scrapped units by the cost accumulated in each unit up to the point it was scrapped. It matters because a unit scrapped late in the process has absorbed more material, labour and machine time than one scrapped early. Commonly misread: - Costing scrap at raw material value only, ignoring conversion cost. - Mixing units and weight or measure bases between products. - Excluding reworked items, which consume cost without being scrapped. - Failing to separate normal, expected process loss from abnormal scrap. Record a reason code, product, machine and shift for each scrap event so that causes can be ranked by cost, not just by count. URL: https://zansa.co.za/answers/how-do-i-calculate-scrap-rate-and-put-a-cost-on-it ### How should a factory measure and categorise machine downtime? Measure downtime as the time a machine is scheduled to run but is not producing, and record every event with a start time, end time and reason code. Group causes into planned stops, such as changeovers and maintenance, and unplanned stops, such as breakdowns and material shortages. It matters because downtime is the first loss in OEE and the easiest to reduce once its causes are visible. Commonly misread: - Using a generic reason like Other for most events, which removes any chance of analysis. - Recording only long stops while many short stops add up to more lost time. - Mixing planned and unplanned downtime so maintenance planning looks like failure. Keep the reason list short, around ten to fifteen codes, and rank causes by total minutes lost, not by number of events. Mean time between failures and mean time to repair add useful detail. URL: https://zansa.co.za/answers/how-should-a-factory-measure-and-categorise-machine-downtime ### How do I calculate cost per unit in a small manufacturing business? Cost per unit is total production cost for a period divided by good units produced in that period. Total cost includes direct materials, direct labour and an allocated share of manufacturing overhead such as machine running costs, rent and supervision. Formula in words: (materials + direct labour + allocated overhead) divided by good units. It matters because pricing, margin and make-or-buy decisions depend on it. Commonly misread: - Dividing by units started rather than good units, which hides scrap cost. - Allocating overhead evenly by units when products use machine time very differently. - Using an old standard cost long after material prices or yields have changed. - Leaving out the cost of rework and downtime. Compare actual cost per unit to standard cost each period, and investigate the variances in materials, labour and overhead separately. URL: https://zansa.co.za/answers/how-do-i-calculate-cost-per-unit-in-a-small-manufacturing-business ### What is work-in-progress in manufacturing and how do I measure it? Work-in-progress (WIP) is the value of partly finished goods still in production, valued at the material, labour and overhead cost incurred so far. Measure it as opening WIP plus costs added, minus the cost of goods completed, per period. It matters because WIP ties up cash, signals bottlenecks, and appears in the balance sheet. Commonly misread: - A rising WIP figure is read as busy production when it may mean jobs are stuck waiting for a bottleneck. - Valuing WIP at full cost when it is only partly complete. - Not tracking WIP age, so old jobs hide within the total. Useful companions are WIP count by stage, average time in WIP, and the age of the oldest batch. If WIP is only counted at month-end, you will miss how it builds up during the month. URL: https://zansa.co.za/answers/what-is-work-in-progress-in-manufacturing-and-how-do-i-measure-it ### How do I use shift data to compare production performance fairly? Compare shifts using rates that account for scheduled time and product mix, such as good units per scheduled hour, OEE, scrap rate and downtime minutes by reason. Raw output totals alone mislead because shifts may differ in length, staffing, product run and breakdowns. It matters because shift differences often reveal training, handover, maintenance or material-availability problems that machine data does not show. Commonly misread: - Blaming a shift for losses caused by upstream material shortages or equipment condition. - Comparing a shift running simple products to one running complex changeovers. - Trusting hand-written shift sheets that are filled in at the end of the shift from memory. Use consistent codes, record events as they happen where possible, and review several weeks of data before drawing conclusions about any shift. URL: https://zansa.co.za/answers/how-do-i-use-shift-data-to-compare-production-performance-fairly ### What is the difference between cycle time and takt time? Cycle time is how long it actually takes to complete one unit at a process step. Takt time is the pace customer demand requires: available production time divided by customer demand. If cycle time exceeds takt time, you cannot meet demand at that rate. Example: 450 available minutes and demand of 225 units gives a takt time of 2 minutes per unit. A step with a 2.5 minute cycle time becomes a bottleneck. It matters because it shows where capacity is short relative to demand, not just where a machine is slow. Commonly misread: - Treating takt time as a machine speed rather than a demand-driven target. - Using a nominal shift length without deducting breaks and planned stops. - Measuring cycle time from a few best-case observations. Recalculate takt whenever demand or working hours change. URL: https://zansa.co.za/answers/what-is-the-difference-between-cycle-time-and-takt-time ### How do I measure production schedule adherence and on-time production? Schedule adherence is the percentage of planned production orders completed in the planned quantity within the planned time window. Divide orders completed as scheduled by total orders scheduled for the period, then multiply by 100. It matters because late or short production breaks delivery promises and drives expediting and overtime costs. Commonly misread: - Counting an order as on time when it finished on time but short in quantity. - Letting the schedule be quietly rewritten, so plans always appear to be met. Measure against the schedule frozen at a set time. - Ignoring the reason for misses, such as material shortages, breakdowns, or planning errors. Track it with reason codes and compare to on-time-in-full delivery to customers, as good production adherence does not guarantee good delivery. URL: https://zansa.co.za/answers/how-do-i-measure-production-schedule-adherence-and-on-time-production ### What are MTBF and MTTR and how are they used in maintenance reporting? MTBF, mean time between failures, is operating time divided by the number of failures. MTTR, mean time to repair, is total repair time divided by the number of repairs. Together they show how often equipment fails and how quickly it is restored. Example: 400 operating hours with four failures gives an MTBF of 100 hours. If those repairs took 12 hours in total, MTTR is three hours. It matters because it separates reliability problems from response and repair problems. Commonly misread: - Averages hide a single very long breakdown. Look at the distribution as well. - Counting planned maintenance as failure. - Mixing different machine types into one average. Track both per machine, and combine with the cost of downtime to prioritise which assets deserve preventive maintenance first. URL: https://zansa.co.za/answers/what-are-mtbf-and-mttr-and-how-are-they-used-in-maintenance-reporting ### How do I measure labour productivity and efficiency on a production line? Labour productivity is output divided by labour hours, for example good units per direct labour hour. Labour efficiency compares standard hours earned (units produced multiplied by standard hours per unit) with actual hours worked, expressed as a percentage. It matters because labour is often a major controllable cost and the clearest lever for cost per unit. Commonly misread: - Comparing products with different standard times in one units-per-hour figure. - Counting hours paid rather than hours worked on the line, or the reverse, without being consistent. - Blaming workers when output was limited by machine downtime or missing material. Read labour efficiency alongside downtime and material availability. Record hours by line and by product so that mix changes do not appear as performance changes. URL: https://zansa.co.za/answers/how-do-i-measure-labour-productivity-and-efficiency-on-a-production-line ### How do I compare standard cost with actual cost in manufacturing? Compare the expected cost of each unit, set from standard material quantities, labour hours and rates, with what you actually spent. The difference is a variance, split into price, usage or efficiency, and volume or overhead effects so each cause is visible. Formulas in words: - Material price variance: (actual price minus standard price) x quantity bought. - Material usage variance: (actual quantity minus standard quantity) x standard price. - Labour efficiency variance: (actual hours minus standard hours) x standard rate. It matters because a total variance alone does not say what to fix. Commonly misread: - Treating a favourable variance as good when it came from poor quality material or skipped steps. - Leaving standards unchanged for years. Review standards at least once a year. URL: https://zansa.co.za/answers/how-do-i-compare-standard-cost-with-actual-cost-in-manufacturing ### What does Zansa's manufacturing dashboard cover and what data does it need? Zansa's manufacturing dashboards join shift sheets and ERP data at batch level to report OEE, yield and a cost-per-unit model. The input data is usually production records from the ERP or spreadsheets, plus shift sheets with downtime and scrap entries. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process starts with a free intro call, then a fixed-fee Data Health Check of one to two weeks to see what data exists and how reliable it is. A Foundation Build, typically four to eight weeks, follows at a fixed price and timeline agreed in writing. Dashboards are usually in Metabase or Power BI. Data sits in a cloud account owned by the client, usually Google BigQuery, and the client owns everything built. URL: https://zansa.co.za/answers/what-does-zansa-s-manufacturing-dashboard-cover-and-what-data-does-it-need ### Why does joining shift sheets to ERP data at batch level improve manufacturing reporting? Shift sheets record what happened on the floor, such as downtime and scrap, while the ERP records materials, orders and costs. Joining them on batch number lets you see the true cost, yield and downtime of each batch rather than estimates built from separate reports. Without a common key, shift losses cannot be tied to the product, order or material that caused them, and cost per unit is an average rather than a measurement. Typical obstacles: - Batch numbers typed inconsistently on paper sheets. - Shifts that span several batches. - Units of measure that differ between systems. Zansa joins shift sheets and ERP records at batch level as part of its manufacturing dashboards. A Data Health Check first assesses how clean the keys are, so the Foundation Build can rely on them. URL: https://zansa.co.za/answers/why-does-joining-shift-sheets-to-erp-data-at-batch-level-improve-manufacturing-r ## Privacy, security and ownership ### Can Zansa work with healthcare businesses that hold sensitive data? Healthcare is one of the sectors Zansa works in. Because health data is sensitive, Zansa signs an operator agreement before touching any data, minimises or de-identifies personal information for reporting, and limits and logs access. Specific legal requirements for your situation are for you and your adviser to confirm. Zansa is designed for privacy law including POPIA and GDPR. Zansa does not claim any certification, so if you need a specific framework, discuss it on the intro call. In practice, reporting often needs counts and trends rather than patient names, so identifiers can usually be removed or masked before the data reaches dashboards. AI tools used for writing code never receive personal or client-identifiable data. URL: https://zansa.co.za/answers/can-zansa-work-with-healthcare-businesses-that-hold-sensitive-data ### Does Zansa sign an agreement before it accesses any client data? Yes. Zansa signs an operator agreement with the client before any data is touched. This sets out that Zansa processes personal information only on the client's instructions. It is part of Zansa's approach of designing for privacy laws including POPIA and GDPR. The agreement comes before read access is granted or any data is loaded. It is separate from the written scope, price and timeline, which are also agreed before a build begins. If you have your own template or legal requirements, raise them on the intro call. This is general information and not legal advice, so have your adviser review any agreement. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/does-zansa-sign-an-agreement-before-it-accesses-any-client-data ### How does Zansa protect personal information in the reports it builds? Zansa minimises or de-identifies personal information for reporting, limits who has access, and logs that access. Most dashboards need totals and trends rather than names, so identifiers can often be removed or masked before the data reaches the reporting layer. The approach follows common privacy principles: only use the data the purpose requires, and restrict who can see it. In practice that means asking, for each report, whether a person's identity is genuinely needed. Where it is not, it is left out. Where it is, access is restricted to those who need it. Zansa is designed for privacy law including POPIA and GDPR, but this is not legal advice for your circumstances. URL: https://zansa.co.za/answers/how-does-zansa-protect-personal-information-in-the-reports-it-builds ### Does Zansa use AI tools on my business or customer data? No, not without written approval. AI tools used by Zansa for writing code never receive personal or client-identifiable data. Zansa does not use AI on client data unless the client approves it in writing, and any machine-learning models run inside the client's own cloud account. This separates two uses of AI. Helping a developer write code is different from feeding your records to a model. The first is done without your personal or identifiable data; the second only happens with your written agreement. Forecasting models are trained and run in your own account, so your data stays in your environment. If you have strict rules on AI use, state them on the intro call. URL: https://zansa.co.za/answers/does-zansa-use-ai-tools-on-my-business-or-customer-data ### What is GDPR and does it apply to a small business? GDPR is the European Union's General Data Protection Regulation. It governs how personal data of people in the EU is collected and used, and applies to organisations of any size, including small businesses, that offer goods or services to people there or monitor their behaviour. This is general information, not legal advice. Core principles include lawfulness, purpose limitation, data minimisation, accuracy, storage limitation, security and accountability. Individuals have rights such as access and erasure. Small organisations get some relief on record-keeping in certain cases, but not an exemption from the core principles. Whether GDPR applies to you depends on where your customers are and what you do with their data, so check with a qualified adviser. URL: https://zansa.co.za/answers/what-is-gdpr-and-does-it-apply-to-a-small-business ### Does GDPR apply to my business if it is based outside the EU? It can. GDPR has extraterritorial reach: a business outside the EU may still be covered if it offers goods or services to people in the EU or monitors their behaviour there. Location of the business alone does not decide it. This is general information, not legal advice. For example, an online shop outside Europe that markets to and ships to EU customers is likely to be in scope. A business that only serves local customers and has an occasional EU visitor usually is not. If you are covered, you may need to consider a representative in the EU in some cases. A qualified adviser can confirm how the rules apply to your facts. URL: https://zansa.co.za/answers/does-gdpr-apply-to-my-business-if-it-is-based-outside-the-eu ### What is the difference between UK GDPR and EU GDPR? UK GDPR is the version of the regulation that applies in the United Kingdom after Brexit, alongside the Data Protection Act 2018. It is very similar in principles and rights to EU GDPR, but it is a separate legal regime, with the UK regulator being the Information Commissioner's Office. This is general information, not legal advice. A business dealing with both UK and EU individuals may need to consider both regimes. Rules for transferring personal data internationally are also set separately in each. Because laws and adequacy arrangements change, check current guidance from the ICO and the relevant EU authorities, or ask a qualified adviser, rather than relying on a general summary. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-the-difference-between-uk-gdpr-and-eu-gdpr ### What is POPIA and who must comply with it? POPIA is South Africa's Protection of Personal Information Act. It sets conditions for lawfully processing personal information, and applies to responsible parties, meaning organisations that decide why and how personal information is processed, in most private and public bodies. This is general information, not legal advice. POPIA covers information about identifiable people and, in South Africa, also identifiable juristic persons such as companies. Its conditions include accountability, purpose specification, minimal processing, security safeguards and openness. The regulator is the Information Regulator. Small businesses are generally not exempt, so if you handle customer, staff or supplier information in South Africa, check your obligations with a qualified adviser. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-popia-and-who-must-comply-with-it ### What is an operator under POPIA and what must the agreement say? Under POPIA, an operator is a party that processes personal information on behalf of a responsible party under a mandate, without coming under its direct authority. POPIA expects a written contract requiring the operator to keep the information secure and to treat it confidentially. This is general information, not legal advice. A data analytics provider is typically an operator when it handles your customers' personal information on your instructions. The responsible party remains accountable. A sound operator agreement usually covers purpose, security measures, confidentiality, notification of breaches, return or deletion of data, and limits on onward sharing. Have your adviser review the precise wording for your situation. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-an-operator-under-popia-and-what-must-the-agreement-say ### What is CCPA and CPRA and does it affect small businesses? CCPA is the California Consumer Privacy Act, amended by the California Privacy Rights Act (CPRA). It gives California residents rights over their personal information. It applies only to businesses that meet certain thresholds, so many small businesses are not covered, but some are. This is general information, not legal advice. The thresholds relate to factors such as annual revenue, the volume of personal information handled, and income from selling or sharing it. They are updated from time to time. If you do fall in scope, rights typically include knowing what is collected, deleting it and opting out of sale or sharing. Check the current thresholds with the California authorities or a qualified adviser before concluding that you are exempt. URL: https://zansa.co.za/answers/what-is-ccpa-and-cpra-and-does-it-affect-small-businesses ### Does HIPAA apply to a business that analyses health data? HIPAA is a US law that applies to covered entities, such as healthcare providers and health plans, and to their business associates who handle protected health information for them. An analytics provider working with such data may be a business associate. This is general information, not legal advice. Where a business associate relationship exists, HIPAA expects a written business associate agreement and appropriate safeguards. HIPAA applies to US-regulated health information and does not cover all health data everywhere. Other countries have their own rules. Whether HIPAA applies to you depends on who you are and whom you serve, so get advice before assuming it does or does not. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/does-hipaa-apply-to-a-business-that-analyses-health-data ### What is data residency and why does it matter for compliance? Data residency refers to the country or region where data is physically stored. It matters because some laws, contracts or customers restrict where personal or regulated data may be kept or transferred. It is a different, though related, matter from who can access the data. This is general information, not legal advice. Cloud providers usually let you choose a region when setting up an account. Choosing a region in a suitable jurisdiction can simplify compliance, but it does not on its own settle transfer rules if people elsewhere can access the data. Check your contracts and the laws that apply to your data, and record where each system stores it. A qualified adviser can help with specific requirements. URL: https://zansa.co.za/answers/what-is-data-residency-and-why-does-it-matter-for-compliance ### Can I send personal data to another country for analytics work? Often yes, but many privacy laws, including GDPR, UK GDPR and POPIA, place conditions on cross-border transfers, such as adequate protection in the destination or suitable contractual safeguards. Remote access from abroad can count as a transfer. This is general information, not legal advice. Before sending or exposing personal data across borders, consider: - Which laws apply to the data subjects - Whether the destination is recognised as adequate or whether contractual safeguards are needed - Whether you can avoid the transfer, for example by de-identifying data first Rules change over time, so confirm current requirements with a qualified adviser. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/can-i-send-personal-data-to-another-country-for-analytics-work ### What is a data processing agreement and when do I need one? A data processing agreement is a contract between a data controller and a processor, also called an operator in POPIA, that governs how the provider handles personal data on your behalf. You generally need one whenever an outside party processes personal data for you. This is general information, not legal advice. GDPR requires such a contract between controller and processor. POPIA expects a written contract with operators. Typical contents include the subject and duration, types of data, instructions, confidentiality, security measures, help with data subject requests, breach notification and return or deletion at the end. Use a template reviewed by a qualified adviser, and sign it before any data is shared. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-a-data-processing-agreement-and-when-do-i-need-one ### What is the difference between a data controller and a data processor? A controller decides why and how personal data is processed, while a processor handles it on the controller's behalf and on its instructions. The same roles exist under POPIA as responsible party and operator. The controller carries the main accountability. This is general information, not legal advice. A retailer deciding to analyse its customer purchases is the controller. The analytics provider doing the work under its instructions is the processor. Roles depend on facts, not labels. A provider that starts using the data for its own purposes may become a controller as well. That is why a written agreement should state who decides what. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-the-difference-between-a-data-controller-and-a-data-processor ### What is pseudonymisation and how does it differ from anonymisation? Pseudonymisation replaces identifying details, such as names, with codes, while a separate key can still link the data back to a person. Anonymisation removes identifiability so that the data can no longer reasonably be linked to anyone. Pseudonymised data is generally still personal data under GDPR. This is general information, not legal advice. Pseudonymisation reduces risk and is a recognised safeguard, but does not take data outside privacy law. True anonymisation is hard, because combinations of ordinary fields, such as postcode, age and date, can sometimes identify a person. Keep any re-identification key separate and restricted, and test whether anonymised data could realistically be linked back. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-pseudonymisation-and-how-does-it-differ-from-anonymisation ### How can a small business de-identify data before building reports? Remove or mask direct identifiers such as names, email addresses, phone numbers and ID numbers, replace them with codes where you still need to group records, and generalise details like exact dates of birth or full addresses. Dashboards usually need totals and trends rather than individuals. A practical approach: - Decide what each report needs to show - Drop every field not needed for that purpose - Replace identifiers with consistent codes - Generalise precise details, such as age bands instead of birth dates - Keep any key that could reverse the process separate and restricted Test whether combined fields could still identify someone, particularly in small groups. URL: https://zansa.co.za/answers/how-can-a-small-business-de-identify-data-before-building-reports ### What does data minimisation mean in practice for analytics projects? Data minimisation means collecting and using only the personal data you actually need for a stated purpose. In analytics, that means not copying whole customer tables into reports when a handful of fields would answer the question, and removing identifiers once they are no longer required. It is a principle in GDPR and a similar idea appears in POPIA's minimality and purpose conditions. It also lowers your risk, because data you do not hold cannot be leaked. Before a project, list each field you plan to use and what decision it supports. If you cannot name one, leave it out. Revisit the list when reports change. This is general information, not legal advice. URL: https://zansa.co.za/answers/what-does-data-minimisation-mean-in-practice-for-analytics-projects ### What is access control and why does it matter for company data? Access control decides who can see or change which data and systems. It matters because most data incidents involve someone having more access than they needed, whether through a mistake, a stolen password or misuse. Good access control limits the damage any one account can do. It combines authentication, which proves who someone is, with authorisation, which defines what they may do. Practical steps include individual accounts rather than shared logins, role-based permissions, regular review of who has access, and prompt removal when people leave. Apply it to dashboards, databases, spreadsheets and cloud storage alike. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-access-control-and-why-does-it-matter-for-company-data ### What is the principle of least privilege and how do I apply it? Least privilege means giving each person or system only the access needed to do their job, and nothing more. Apply it by starting from no access, granting specific permissions for defined roles, preferring read-only access where editing is not needed, and reviewing permissions regularly. Practical steps: - Define roles, such as finance viewer or sales viewer, rather than granting access individually - Use read-only access for tools that only need to read - Grant time-limited access for temporary needs - Review access every few months and after staff changes It limits the harm from mistakes and compromised accounts. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-the-principle-of-least-privilege-and-how-do-i-apply-it ### Why is multi-factor authentication important for business data systems? Multi-factor authentication requires a second proof of identity, such as an app code or security key, in addition to a password. It matters because stolen or guessed passwords are a common way in, and a second factor stops many of those attempts from succeeding. Turn it on first for email, cloud accounts, accounting software, data warehouses and administrator accounts, since these give the broadest access. App-based or hardware-key methods are generally stronger than text messages. It is not a complete defence, so combine it with least privilege, monitoring and prompt removal of leavers. Many cloud and accounting platforms offer it at no extra cost. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/why-is-multi-factor-authentication-important-for-business-data-systems ### What is audit logging and what should a small business log? Audit logging records who accessed or changed what data, and when. It lets you investigate incidents, show accountability and spot unusual behaviour. At a minimum, log sign-ins, access to sensitive datasets, permission changes and exports of data. Check how this applies to you with a qualified adviser. Many cloud platforms provide logs that only need switching on and retaining. Decide how long to keep them, who can read them, and who reviews them, because logs nobody checks give limited value. Protect logs from tampering, and avoid recording personal data in them unnecessarily. Logging supports accountability under laws such as GDPR and POPIA, though it does not alone ensure compliance. URL: https://zansa.co.za/answers/what-is-audit-logging-and-what-should-a-small-business-log ### Why does encryption matter and what should be encrypted? Encryption scrambles data so it cannot be read without the key. It matters because it protects information if a device is lost or a storage system is accessed improperly. Encrypt data in transit, when it moves over networks, and at rest, when it is stored. Most reputable cloud services encrypt stored data by default and use secure connections, but confirm this rather than assume it. Also consider laptops, phones, USB drives and backups. Encryption protects mainly against particular risks. It does not help if an authorised account is compromised, so pair it with access controls and strong authentication. Manage keys carefully, since losing them can mean losing the data. URL: https://zansa.co.za/answers/why-does-encryption-matter-and-what-should-be-encrypted ### How should a small business back up its important data? Keep multiple copies of important data, at least one stored separately from the main system, and test that you can actually restore from them. A common rule of thumb is three copies, on two types of storage, with one offsite. Backups must also be protected from unauthorised access. Decide how much data you could afford to lose and how long you could be without it, then set backup frequency and recovery time to match. Remember that backups contain personal data too, so they need encryption, access control and a retention period. Cloud services often have built-in backup or recovery features, but check what is covered, since some only protect against platform failure and not accidental deletion. URL: https://zansa.co.za/answers/how-should-a-small-business-back-up-its-important-data ### How long should a business keep personal data? Keep personal data only as long as you need it for the purpose you collected it for, or as long as the law requires you to keep it. There is no single period for everyone, because retention depends on the type of data and the laws that apply. This is general information, not legal advice. Good practice is to set a written retention schedule that lists each data type, the reason for keeping it, and when it is deleted or anonymised. Tax, employment and sector rules often require specific minimum periods for certain records. GDPR calls this storage limitation, and POPIA has a similar retention principle. Include backups and copies in spreadsheets when you delete. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/how-long-should-a-business-keep-personal-data ### What should a small business do first if it suffers a data breach? Contain it, record what happened, assess what personal data was affected, and check whether you must notify a regulator or affected individuals. Act quickly, because some laws set short deadlines. This is general information, not legal advice. Check how this applies to you with a qualified adviser. Immediate steps usually include: - Stop the exposure, for example by resetting passwords or revoking access - Preserve logs and evidence - Work out what data and which people are involved - Inform your operators or processors where relevant - Decide on notification with legal advice Under GDPR, notifying the supervisory authority is generally due within 72 hours of becoming aware, where required. POPIA requires notification as soon as reasonably possible. Confirm current rules locally. URL: https://zansa.co.za/answers/what-should-a-small-business-do-first-if-it-suffers-a-data-breach ### How soon must a data breach be reported under GDPR and POPIA? Under GDPR, a controller must generally notify the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of a breach, unless it is unlikely to risk individuals. POPIA requires notification to the regulator and affected people as soon as reasonably possible. This is general information, not legal advice. Not every incident is notifiable, and the threshold depends on the risk to individuals. You should still record every breach internally. Processors and operators are normally expected to tell the controller or responsible party promptly, so make sure your agreements state how quickly. Take legal advice quickly when an incident happens, because deadlines run from awareness. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/how-soon-must-a-data-breach-be-reported-under-gdpr-and-popia ### What should I check before choosing a data or analytics vendor? Check where they store data, who can access it, whether they will sign a processing or operator agreement, how they secure and log access, who owns the outputs, and what happens when you leave. Also ask which sub-processors and AI tools they use. This is general information, not legal advice. A quick due diligence list: - Written data processing or operator agreement - Data location and your control over it - Access controls, logging and encryption - Breach notification commitments - Use of AI tools and sub-processors - Ownership of code, models and dashboards - Exit and data return terms Ask for answers in writing, and be cautious of vague responses. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-should-i-check-before-choosing-a-data-or-analytics-vendor ### What security questions should I ask a vendor that will handle my data? Ask how they authenticate users, whether multi-factor authentication is required, how access is limited and logged, how data is encrypted, how incidents are handled, which staff or sub-processors can see your data, and how data is returned or deleted at the end of the contract. Also ask whether the vendor stores your data in their own environment or yours, because data held in your own account is easier to control and withdraw. Request details of any certifications they claim, and verify them rather than relying on a logo. Where answers are unclear, ask for them in writing, and consider whether the risks are acceptable for the sensitivity of the data. URL: https://zansa.co.za/answers/what-security-questions-should-i-ask-a-vendor-that-will-handle-my-data ### Can I put customer personal data into ChatGPT or other AI tools? It is risky and often inadvisable. Putting personal data into a public AI tool can mean sharing it with a third party without a lawful basis, an agreement or control over how it is retained. Check the provider's terms, your privacy notice and the law first. This is general information, not legal advice. Safer habits include: - Never paste identifiable customer or staff data into consumer AI tools - Remove or mask identifiers first - Use business-grade offerings with written data terms where AI use is justified - Set a short written staff policy on AI use Analytics providers should also be able to say clearly what AI tools they use and what data those tools see. URL: https://zansa.co.za/answers/can-i-put-customer-personal-data-into-chatgpt-or-other-ai-tools ### How can a company use AI tools safely when personal data is involved? Use AI only where there is a lawful purpose, minimise or de-identify the data first, choose providers with written data protection terms, keep processing inside an environment you control where possible, and keep a human accountable for decisions. Write down your rules. This is general information, not legal advice. Consider whether you need personal data at all. Many forecasting and pattern-finding tasks work on aggregated or de-identified figures. Running models inside your own cloud account avoids sending data to outside services. Where outside AI services are used, check retention, training use, location and sub-processors. Some privacy laws also restrict automated decisions that significantly affect people, so check that before deploying. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/how-can-a-company-use-ai-tools-safely-when-personal-data-is-involved ### Who owns the data when a company hires an outside analytics provider? The client should own its data, and the contract should say so. A provider handling personal data as a processor or operator acts on the client's instructions and should not claim ownership or use the data for its own purposes. This is general information, not legal advice. Make sure the agreement covers data ownership, permitted use, return or deletion at the end, and who owns derived items such as models and reports. Do not assume; some providers claim rights over outputs or aggregated data. Storing data in a cloud account you own makes ownership practical as well as legal, since you hold the keys and can revoke access. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/who-owns-the-data-when-a-company-hires-an-outside-analytics-provider ### Who owns the dashboards and models an analytics provider builds for me? It depends on the contract. Without clear terms, ownership of code, models and dashboards can be disputed. Insist on written terms stating that you own the deliverables, including documentation, and that this applies from the start rather than only after final payment. This is general information, not legal advice. Questions to settle: - Do you own the code and data models, or only a licence? - Is documentation included? - Can another provider maintain the work? - Does the provider reuse your work for others? Also check whether dashboards depend on a tool with per-user licensing, because ownership of the work does not remove licence costs. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/who-owns-the-dashboards-and-models-an-analytics-provider-builds-for-me ### What should an exit and handover plan include when ending a data project? Include revoking the provider's access, returning or deleting any data they hold, confirming ownership of code, models and dashboards, receiving documentation, and a handover session. Also rotate any shared credentials and update your records of who processes your data. Check how this applies to you with a qualified adviser. A checklist: - Remove accounts and API keys - Obtain written confirmation of deletion of any copies - Collect documentation and credentials for tools you own - Check that scheduled jobs and dashboards still run - Update your processor register and privacy notice if needed Plan this at the start by putting exit terms in the contract. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-should-an-exit-and-handover-plan-include-when-ending-a-data-project ### Who is liable if my analytics provider causes a data breach? Generally the business that decides how the data is used, the controller or responsible party, stays accountable to regulators and individuals, even when a provider caused the problem. Providers can also face direct obligations. Contract terms decide how loss is shared between you. This is general information, not legal advice. Because you remain accountable, choose providers carefully and sign a written processing or operator agreement setting out security duties, breach notification and any indemnity or liability limits. Insurance, including cyber cover, may also help. How liability falls in a specific case depends on the laws, the contract and the facts, so a qualified adviser should review it. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/who-is-liable-if-my-analytics-provider-causes-a-data-breach ### Do I need a data protection officer for a small business? Not necessarily. GDPR requires a data protection officer only in particular cases, such as public authorities or organisations whose core activities involve large-scale monitoring or large-scale sensitive data. Under POPIA, an information officer role exists and is by default the head of the organisation. This is general information, not legal advice. Even without a formal DPO, someone should own privacy in the business, handle requests and be the contact for regulators and providers. In South Africa, the information officer needs to be registered with the Information Regulator. Check whether you meet the criteria that apply in your country, and ask a qualified adviser if you are unsure. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/do-i-need-a-data-protection-officer-for-a-small-business ### What are data subject access requests and how should a small firm handle them? A data subject access request is when an individual asks what personal data you hold about them and how you use it. Laws such as GDPR and POPIA give people this right. Handle it by verifying identity, searching your systems, and responding within the legal deadline. This is general information, not legal advice. GDPR generally expects a response within one month, with limited extensions. Check the deadline in your jurisdiction. Preparation helps: keep an up-to-date list of where personal data lives, including spreadsheets and exports, so that you can find it. Reporting and analytics copies count too, which is another reason to keep personal data out of them where possible. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-are-data-subject-access-requests-and-how-should-a-small-firm-handle-them ### What counts as personal data or personal information? Personal data is any information relating to an identified or identifiable person, such as names, email addresses, phone numbers, ID numbers, online identifiers, location data and purchase histories linked to a person. POPIA uses the term personal information and also covers identifiable companies. This is general information, not legal advice. Information can be identifying even without a name, if it can reasonably be linked to someone, for example by combining postcode, age and job title. Some categories are more sensitive, such as health, biometric, religious and children's data, and carry stricter rules. When unsure whether a field is personal, treat it as personal until you have checked. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-counts-as-personal-data-or-personal-information ### Is health data treated differently under privacy law? Yes. Health data is generally treated as a special or sensitive category and attracts stricter conditions under GDPR and POPIA, and sector rules such as HIPAA in the US. Processing usually needs a specific legal basis or condition and stronger safeguards. This is general information, not legal advice. Extra care typically means tighter access control, encryption, de-identification for reporting, and clear documentation of why the data is needed. Businesses that only need counts or trends, such as appointments per month, can often work without identifiers. Because rules vary by country and role, confirm the requirements that apply to you with a qualified adviser. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/is-health-data-treated-differently-under-privacy-law ### Can I use customer data for analytics without asking for consent? Sometimes. Consent is only one legal basis. Under GDPR, other bases such as legitimate interests or contract performance can apply, and POPIA has its own grounds for processing. The purpose must be compatible with why the data was collected, and people should be told. This is general information, not legal advice. Using customer data to understand your own business, such as sales trends, often fits within the original purpose, particularly if your privacy notice explains it. Using it for something very different, or for marketing profiling, may need new consent. Document your reasoning and minimise the data you use. Ask a qualified adviser if you rely on a judgement call. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/can-i-use-customer-data-for-analytics-without-asking-for-consent ### What is a privacy impact assessment and when is one needed? A privacy impact assessment, called a data protection impact assessment under GDPR, is a structured review of the privacy risks of a planned use of data. GDPR requires one when processing is likely to result in high risk to individuals, such as large-scale sensitive data. This is general information, not legal advice. Even where not mandatory, a short assessment is useful before connecting systems or adopting new tools. It asks what data is used, why, who can see it, what could go wrong and how to reduce the risk. The result may be to drop fields, add controls or choose a different approach. Keep it on file as evidence of accountability. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-a-privacy-impact-assessment-and-when-is-one-needed ### Is it safe to share spreadsheets with personal data by email? It is risky. Emailed spreadsheets are copied, forwarded and stored in many inboxes, with no easy way to control or withdraw access. If they contain personal data, a misdirected email can itself be a data breach. Prefer secure shared locations with access control, and remove unneeded personal columns. Safer habits: - Share links in a controlled system, with named users and expiry - Remove or mask identifiers before sharing - Password-protect and encrypt files where email is unavoidable, sending the password separately - Keep one master copy rather than many versions Spreadsheets are also hard to audit, which is a reason to move reporting into governed systems. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/is-it-safe-to-share-spreadsheets-with-personal-data-by-email ### Why are shared passwords a problem for data security? Shared passwords remove accountability, because you cannot tell who did what, and they are hard to revoke when someone leaves without changing the password for everyone. They also tend to spread to places they should not be. Individual accounts with multi-factor authentication are safer. If a team needs shared access, use a password manager that supports sharing, or give each person their own account with appropriate permissions. For service accounts used by software, store credentials in a secrets manager rather than in documents or code, and rotate them when staff or providers change. Individual accounts also make audit logs meaningful. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/why-are-shared-passwords-a-problem-for-data-security ### Should staff and payroll data be included in company dashboards? Only where there is a clear need, and with tighter access than for general reports. Staff and payroll data is personal and often sensitive, so show aggregates where possible and restrict individual-level views to the few people who need them, such as HR or finance leads. Staff data is covered by privacy laws too, and employees have rights. Consider the purpose of each metric and whether you can use totals or averages by team instead of named individuals. Small groups can accidentally reveal individuals, so avoid showing figures for teams of one or two. Log access to these views and review who has it regularly. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/should-staff-and-payroll-data-be-included-in-company-dashboards ### What is a sub-processor and why should I ask about them? A sub-processor is another company that your provider uses to help process your data, such as a hosting or software service. You should ask because your data may flow to parties you have not vetted, and under GDPR a processor generally needs your authorisation to engage one. This is general information, not legal advice. Ask for a list of sub-processors, where they are located, and how you will be told about changes. The provider should bind them to equivalent data protection terms and remains responsible for them. AI tools and cloud services count as sub-processors if they handle your personal data, so include them in your questions. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-is-a-sub-processor-and-why-should-i-ask-about-them ### Is it better to keep company data in my own cloud account or the provider's? Keeping data in a cloud account you own generally gives you more control, because you hold the keys, choose the settings, see the logs and can revoke the provider's access at any time. Provider-hosted data depends on their security and their willingness to hand it back. A client-owned account also makes exit simpler, since the data never needed to be returned. It makes responsibilities clearer too: you set the region, access policy and retention, and the provider works inside your environment. The trade-off is that you carry the account's billing and need basic administration. Either model needs a written agreement and sensible access controls. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/is-it-better-to-keep-company-data-in-my-own-cloud-account-or-the-provider-s ### What should I do when an employee or contractor with data access leaves? Remove their access on or before their last day, covering email, cloud accounts, databases, dashboards and shared drives. Rotate any shared credentials or keys they knew, retrieve devices, and review logs for unusual activity. Prompt removal is one of the cheapest security measures available. Keep a current list of who has access to what, so nothing is missed. Include outside providers and contractors, whose access often outlasts the project. Remind leavers in writing of continuing confidentiality duties. Individual accounts, rather than shared ones, make this far easier, since you remove one account without disrupting others. Rules differ by country and sector, so check how this applies to you with a qualified adviser. URL: https://zansa.co.za/answers/what-should-i-do-when-an-employee-or-contractor-with-data-access-leaves ### How do I keep a record of which vendors process my personal data? Keep a simple register listing each vendor, what personal data they process, why, where it is stored, what agreement covers them, and who in your business owns the relationship. A spreadsheet is enough for a small business, provided it is kept up to date. GDPR expects controllers and processors to keep records of processing in many cases, and POPIA expects you to know what you hold and why. A vendor register supports both, and makes breach response and access requests faster. Review it at least annually and whenever you add or remove a tool. Include analytics, email, accounting, payment and AI tools. This is general information, not legal advice. URL: https://zansa.co.za/answers/how-do-i-keep-a-record-of-which-vendors-process-my-personal-data ### Does Zansa sign an operator agreement and limit access when it handles my data? Yes. Zansa signs an operator agreement before any data is touched, minimises or de-identifies personal information for reporting, and limits and logs access. Data stays in a cloud account you own, access is revocable at any time, and nothing is stored in a Zansa account. These measures are designed with privacy law including POPIA and GDPR in mind. They do not replace your own legal review, and Zansa claims no certifications. AI tools used by Zansa for writing code never receive personal or client-identifiable data, and AI is not used on client data without written approval. Machine-learning models run in your account. Ask for the agreement text on the intro call if you want your adviser to review it first. URL: https://zansa.co.za/answers/does-zansa-sign-an-operator-agreement-and-limit-access-when-it-handles-my-data ### What is a data retention schedule and how do I create one? A data retention schedule is a written list of each type of data you hold, why you keep it, how long you keep it, and what happens at the end, such as deletion or anonymisation. Create one by listing your data, checking legal minimum periods, and assigning an owner. Start with the categories that matter most, such as customer records, invoices, staff files and marketing lists. For each, note the legal or business reason for keeping it and the period. Tax and employment rules often set minimums, so check them locally. Remember copies: spreadsheets, exports, reports and backups all need to be covered. Review the schedule yearly. This is general information, not legal advice. URL: https://zansa.co.za/answers/what-is-a-data-retention-schedule-and-how-do-i-create-one ## Professional services ### What is billable utilisation and how do I calculate it? Billable utilisation is billable hours divided by available hours, multiplied by 100, for a person or team. It shows how much of paid capacity is spent on work that can be charged to clients. Example: 120 billable hours out of 160 available gives 75 percent. It matters because time is the stock in trade of a professional services firm. Commonly misread: - Using different denominators: total hours worked, contracted hours, or hours after leave and holidays. State which you use. - Treating 100 percent as the target. Staff also need time for training, sales, admin and management. - Counting hours billed but later written off. Set separate expectations by role, since partners and juniors spend different shares of time on client work. URL: https://zansa.co.za/answers/what-is-billable-utilisation-and-how-do-i-calculate-it ### What is work in progress (WIP) in a professional services firm? WIP is the value of work done for clients that has not yet been invoiced. It is usually calculated as recorded time multiplied by the billing rate (and expenses incurred) for the period, less amounts already billed. It matters because WIP is earned revenue that has not yet turned into a debt owed to you. It sits between work and cash. Commonly misread: - Treating all WIP as collectable. Old WIP is often written down. - Mixing recorded time at standard rates with value expected to be billed. - Ignoring the age of WIP. A large balance that is mostly recent is different from one that is mostly old. Report WIP by client, project and age. Days of WIP is WIP divided by average daily billing, which shows how long work waits before invoicing. URL: https://zansa.co.za/answers/what-is-work-in-progress-wip-in-a-professional-services-firm ### What is realisation rate and how is it calculated? Realisation rate is the revenue you actually bill (or collect) divided by the value of time recorded at standard rates, multiplied by 100. It shows how much of the time you put in is turned into fees. Example: 80,000 billed for work recorded at a standard value of 100,000 gives 80 percent. It matters because gaps arise from discounts, write-offs, fixed-fee overruns and unbilled time. Commonly misread: - Billing realisation (billed over recorded value) differs from collection realisation (collected over recorded value). Label which you report. - Using outdated standard rates. - Reading an overall rate without separating clients and services. Break it down by client, fee type and team to find where value leaks, and check write-off reasons. URL: https://zansa.co.za/answers/what-is-realisation-rate-and-how-is-it-calculated ### What are lock-up days and how are they calculated? Lock-up days measure how long cash is tied up between doing work and receiving payment. It equals WIP days plus debtor days. WIP days is unbilled work divided by billings, and debtor days is receivables divided by billings, each multiplied by the days in the period. Formula in words: (unbilled WIP + trade receivables) divided by billings in the period, multiplied by days in the period. Example: 60,000 WIP and 90,000 receivables against 300,000 billed over 90 days gives (150,000 / 300,000) x 90 = 45 lock-up days. It matters because long lock-up means the firm funds its clients' work. Commonly misread: - Using different billing periods for WIP and receivables. - Looking at the total without seeing whether the delay is in invoicing or in collection. Track the two parts separately. URL: https://zansa.co.za/answers/what-are-lock-up-days-and-how-are-they-calculated ### How do I measure client profitability in a services firm? Client profitability is the fees earned from a client minus the cost of the time and expenses spent serving them, as a currency amount and as a percentage of fees. Cost of time uses a cost rate per hour for each person, not the billing rate. Formula in words: (fees collected or billed minus direct staff cost of time minus direct expenses) divided by fees. It matters because some clients take far more effort than their fees justify, while others carry the firm. Commonly misread: - Using billing rates for cost, which hides real margin. - Ignoring time that was not recorded, or write-offs. - Leaving out the cost of slow payment. - Ignoring strategic value such as referrals, which can be noted separately. Review regularly and use the result for repricing, scoping and, if needed, ending unprofitable relationships. This requires timesheets, billing and ledger data reconciled. URL: https://zansa.co.za/answers/how-do-i-measure-client-profitability-in-a-services-firm ### How should a professional services firm track its sales pipeline? Track each opportunity with stage, expected fee, expected start date and a probability of winning, then calculate the weighted value as fee times probability. Report the pipeline by stage, age and expected timing against your forecast capacity. It matters because services firms need to see future work early enough to hire or redeploy staff. Useful measures: - Pipeline value by stage. - Win rate: opportunities won divided by those decided. - Average time in each stage. Commonly misread: - Treating the total pipeline as revenue when only a share will be won. - Probabilities set by guesswork. Calibrate them using your own history of wins and losses. - Not removing stale opportunities. Compare weighted pipeline timing with team capacity to spot gaps and bottlenecks. URL: https://zansa.co.za/answers/how-should-a-professional-services-firm-track-its-sales-pipeline ### What is effective hourly rate and how does it differ from the standard rate? Effective hourly rate is the revenue earned on a piece of work divided by the total hours spent on it, whether billable or not. It differs from the standard rate, which is the list price per hour. The effective rate shows what the work really earned per hour. Example: a fixed fee of 10,000 for 125 hours of work gives an effective rate of 80 per hour, regardless of a standard rate of 120. It matters most for fixed-fee and capped work, where overruns reduce the real rate. Commonly misread: - Excluding unrecorded or unbilled time. - Comparing with the standard rate and calling the gap a discount, when it also reflects scope creep. - Using revenue billed rather than collected. Compare effective rate with your cost rate per hour to see whether the work was profitable. URL: https://zansa.co.za/answers/what-is-effective-hourly-rate-and-how-does-it-differ-from-the-standard-rate ### What is the difference between debtor days and lock-up days? Debtor days measures only the time from invoice to payment: receivables divided by billings, times the days in the period. Lock-up days adds the unbilled work waiting to be invoiced, so it covers the full time from doing the work to getting paid. Lock-up days equal WIP days plus debtor days. It matters because a firm can collect invoices quickly and still be cash-starved if it invoices late. Commonly misread: - Reporting debtor days alone, so slow invoicing stays invisible. - Mixing different billing figures when calculating the two parts. - Averaging across clients, which hides one client with a very large unpaid balance. Track both parts monthly and view the oldest balances by client. Improving invoicing frequency often cuts lock-up faster than chasing payment. URL: https://zansa.co.za/answers/what-is-the-difference-between-debtor-days-and-lock-up-days ### How do write-offs and write-downs affect services firm performance? Write-offs reduce the value of WIP or invoices you will not recover, and write-downs reduce billed value below recorded time value. Both lower realisation and profit, and indicate scoping, pricing or quality problems that need fixing. Formula in words: write-off rate equals value written off divided by value recorded or billed, multiplied by 100. It matters because a small percentage across many projects adds up to a large figure. Commonly misread: - Treating write-offs as unavoidable cost rather than a signal about estimates, scope control or staff training. - Failing to record the reason, so patterns cannot be found. - Looking only at the firm total and not by client, project manager or work type. Record reason codes and review them regularly with project owners. URL: https://zansa.co.za/answers/how-do-write-offs-and-write-downs-affect-services-firm-performance ### What is revenue per fee earner and how is leverage calculated? Revenue per fee earner is total fee revenue divided by the number of fee earners, ideally in full-time equivalents. Leverage is the ratio of junior staff to senior staff (or partners) working on the same work. Both describe how the firm uses its people. It matters because firms with higher leverage can earn more per senior person, if juniors are well supervised and fully used. Commonly misread: - Counting part-time staff as full-time. - Reading revenue per head without costs; high revenue with high pay may not earn more profit. - Assuming more leverage is always better. Too much can reduce quality and increase rework. Compare revenue per fee earner with profit per fee earner and utilisation to see how effectively capacity is turned into earnings. URL: https://zansa.co.za/answers/what-is-revenue-per-fee-earner-and-how-is-leverage-calculated ### How do I track budget burn on fixed-fee projects? Budget burn compares hours or cost spent so far with the project budget, alongside the percentage of work complete. If 70 percent of budgeted hours are used but only 40 percent of the work is done, the project is heading for overrun. Formulas in words: - Budget used: hours or cost to date divided by budget. - Estimate at completion: hours to date divided by percentage complete. It matters because fixed-fee overruns directly reduce margin and cannot be billed. Commonly misread: - Using elapsed time instead of work complete as the progress measure. - Estimating percentage complete optimistically. - Checking only at the end instead of weekly. Set alerts at agreed thresholds such as when 80 percent of budget is used, and discuss scope changes with the client early, while they can still be priced. URL: https://zansa.co.za/answers/how-do-i-track-budget-burn-on-fixed-fee-projects ### How do I set a fair utilisation target for different roles in a firm? Set targets by role from available time, minus time for leave, training, internal work and business development. A junior may reasonably target a high share of billable time, while a manager or partner has lower billable time because of management and sales duties. Method: - Start with contracted hours. - Subtract leave, public holidays and training. - Subtract expected non-billable duties by role. - The remainder is the billable capacity. Commonly misread: - One target for everyone, which punishes people with non-billable responsibilities. - Pushing utilisation too high, leaving no time for quality, learning or sales, which can damage future work. Use your own history to set realistic targets, and read utilisation together with realisation. High utilisation with low realisation means time is recorded but not turning into fees. URL: https://zansa.co.za/answers/how-do-i-set-a-fair-utilisation-target-for-different-roles-in-a-firm ### What does Zansa's professional services dashboard report for firms billing by time? Zansa reconciles timesheets, billing and the ledger for professional services firms, then reports utilisation, WIP and realisation by team and client, plus a monthly client-profitability report. It shows where time and fees are lost between work done and cash received. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process is a free intro call, a fixed-fee Data Health Check of one to two weeks, and then a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Support is optional. Dashboards are usually in Metabase or Power BI. Data lives in a cloud account owned by the client, usually Google BigQuery, and the client owns everything built. URL: https://zansa.co.za/answers/what-does-zansa-s-professional-services-dashboard-report-for-firms-billing-by-ti ### Why reconcile timesheets, billing and the ledger in a services firm? Timesheets show time recorded, billing shows what was invoiced, and the ledger shows what was booked and paid. Reconciling them reveals unbilled time, write-offs, mispriced work and revenue that does not match, so utilisation, WIP and realisation figures can be trusted. Without reconciliation, each system tells a different story: timesheets may overstate billable hours, billing may exclude expenses, and the ledger may book revenue in a different period. Typical checks: - Time recorded but never billed. - Invoices without matching time. - Ledger revenue that differs from billing totals. Zansa reconciles timesheets, billing and ledger data for professional services firms as part of its dashboards, so team and client reports rest on figures that agree across systems. URL: https://zansa.co.za/answers/why-reconcile-timesheets-billing-and-the-ledger-in-a-services-firm ## Property ### How do I calculate occupancy rate for rental property? Occupancy rate is the number of occupied units (or occupied area) divided by the total number of units (or total lettable area), multiplied by 100. A building with 90 of 100 units let is 90 percent occupied by unit count. It matters because empty space earns no rent but still carries costs. Commonly misread: - Unit-based and area-based occupancy differ when units vary in size. State which one you report. - Counting signed leases that have not yet started as occupied, or tenants in arrears as fully paying. - Using a month-end snapshot only, which hides vacancy during the month. Time-weighted occupancy shows more. Report both physical occupancy (someone is in the unit) and economic occupancy (rent actually collected compared with potential rent), because they can differ widely. URL: https://zansa.co.za/answers/how-do-i-calculate-occupancy-rate-for-rental-property ### How should a landlord measure and report rental arrears? Arrears is rent and charges that are due but unpaid. Report the total amount, the number of tenants in arrears, and an ageing by days overdue. Arrears as a percentage of rent billed in the period shows the scale relative to the portfolio. Formula in words: arrears balance divided by total amount billed for the period, multiplied by 100. It matters because overdue rent is cash that is already earned but not received, and the risk of non-payment rises with age. Commonly misread: - Netting tenant deposits or credit balances against arrears without showing both. - Mixing rent with utility recoveries and penalties. - Reporting only a total, which hides one large tenant behind many small ones. Track by tenant, property and age bucket, and set alerts when a tenant crosses an agreed overdue threshold. URL: https://zansa.co.za/answers/how-should-a-landlord-measure-and-report-rental-arrears ### What is vacancy rate and how do I measure the cost of vacancy? Vacancy rate is vacant lettable units or area divided by total lettable units or area, multiplied by 100. The cost of vacancy is the rent the vacant space would have earned, plus holding costs you still pay on it. Formula in words: vacant area divided by total area. Vacancy cost equals market rent for the vacant space multiplied by the time vacant, plus costs such as rates, utilities and security. It matters because it turns a percentage into a currency amount that can be compared with letting costs and incentives. Commonly misread: - Treating all vacant units equally when large or poor units cost more. - Ignoring how long each unit has been vacant. - Using asking rent that the market does not actually pay. Track vacancy duration and the reason each unit is empty. URL: https://zansa.co.za/answers/what-is-vacancy-rate-and-how-do-i-measure-the-cost-of-vacancy ### How do I analyse a lease expiry profile? A lease expiry profile shows how much rent or area comes up for lease expiry in each future period, usually by month or year. Group leases by expiry date and total the annual rent and area expiring in each period, as a share of the portfolio. It matters because a large cluster of expiries in one period concentrates the risk of vacancy and rent resets. Useful companions: - Weighted average lease expiry (WALE), which averages remaining lease terms weighted by rent or area. - Break clauses and renewal options, which can bring effective expiry forward. Commonly misread: - Using contractual end dates when tenants are on month-to-month holdover. - Looking at area only, when rent per unit differs. Set alerts well before expiry, since renewal discussions and re-letting take time. URL: https://zansa.co.za/answers/how-do-i-analyse-a-lease-expiry-profile ### How should a property manager track maintenance backlog? Maintenance backlog is the number and value of open maintenance requests or work orders not yet completed. Track the count, the age of each request, priority, and the estimated cost, and report how many are overdue against target resolution times. It matters because unresolved issues affect tenant satisfaction and renewals, and small repairs can become costly ones. Useful measures: - Open requests by age bucket. - Average days to close, by priority. - Repeat requests on the same unit or asset. Commonly misread: - A falling backlog may reflect requests being closed without the work being done, or tenants giving up reporting. - Counting urgent and cosmetic requests equally. Check closed requests against completion evidence, and look at repeat issues, which often point to an asset to replace rather than keep repairing. URL: https://zansa.co.za/answers/how-should-a-property-manager-track-maintenance-backlog ### How is net operating income calculated for a property? Net operating income (NOI) is property revenue minus operating expenses, before financing costs, income tax and depreciation. Revenue is rent and other income collected or due; operating expenses include rates, insurance, maintenance, management and utilities not recovered. Formula in words: effective rental income plus other income, minus operating expenses. It matters because it measures what the property itself earns, independent of how it is financed, so properties can be compared. Commonly misread: - Including mortgage interest or capital repayments in operating expenses. - Leaving out vacancy and bad debt from income. - Excluding major repairs or treating them inconsistently from period to period. Define which items count as operating versus capital and apply the same rule to every property in the portfolio. URL: https://zansa.co.za/answers/how-is-net-operating-income-calculated-for-a-property ### What is the difference between gross yield and net yield on a property? Gross yield is annual rent divided by property value or purchase price. Net yield is annual net operating income divided by the same value. Net yield is lower because it deducts operating costs, vacancy and bad debt that gross yield ignores. Example: rent of 120,000 on a property worth 1,500,000 is an 8 percent gross yield. If NOI is 90,000, net yield is 6 percent. It matters because gross yield can make a property look better than it performs. Commonly misread: - Comparing a gross yield on one property with a net yield on another. - Using purchase price for one and current valuation for another. - Forgetting that financing costs are outside both measures. State the value basis used and which costs were deducted. URL: https://zansa.co.za/answers/what-is-the-difference-between-gross-yield-and-net-yield-on-a-property ### How do I measure tenant retention and turnover? Tenant retention rate is the number of tenants (or leases) who renewed divided by the number due to expire in the period, multiplied by 100. Turnover is the share of units where the tenant left in the period. They are complementary measures. It matters because each tenant who leaves costs vacancy time, letting fees, cleaning and repairs. Commonly misread: - Measuring against all tenants rather than those whose leases were actually due. - Counting tenants who moved to another unit in your portfolio as lost. - Ignoring reasons for leaving, such as price, condition or service. Record a leaving reason every time and compare it with maintenance response and rent increases. Estimating the total cost per turnover helps weigh retention spending against the cost of re-letting. URL: https://zansa.co.za/answers/how-do-i-measure-tenant-retention-and-turnover ### What is rental reversion and why does it matter? Rental reversion is the percentage change between the old rent on an expiring lease and the new rent on the renewal or re-let of the same space. It is the new rent minus the old rent, divided by the old rent, multiplied by 100. A positive figure means rents are rising on renewal; a negative one means falling. It shows whether current rents are above or below what the market is paying. Commonly misread: - Comparing headline rent without incentives, rent-free periods or fit-out contributions, which change the effective rent. - Averaging a few large moves across a small number of leases. - Mixing renewals with new tenants in different conditions. Track the reversion by property and lease type, and show the number of leases behind each figure so readers can judge its reliability. URL: https://zansa.co.za/answers/what-is-rental-reversion-and-why-does-it-matter ### How do I calculate rent collection rate for a property portfolio? Rent collection rate is rent received in the period divided by rent billed (due) in the same period, multiplied by 100. It shows how much of what should have been paid was actually collected on time. Example: 95,000 collected against 100,000 billed gives 95 percent. It matters because it is the quickest warning sign of tenant distress. Commonly misread: - Counting payments for earlier arrears as collection of this period's rent. Allocate payments to the period they settle. - Including recoveries and penalties in billed amounts without saying so. - Presenting one portfolio-wide figure that hides a single building with serious problems. Track both on-time collection (paid by due date) and collection after follow-up, by tenant and by property. URL: https://zansa.co.za/answers/how-do-i-calculate-rent-collection-rate-for-a-property-portfolio ### What is operating expense ratio for a property? Operating expense ratio is operating expenses divided by effective gross income, multiplied by 100. It shows what share of income is consumed by running the property, before financing and tax. It matters because two properties with similar rent can have very different profitability if one costs more to run. Commonly misread: - Comparing buildings of different types, such as multi-tenant and single-tenant leases where tenants pay their own costs. - Leaving out management fees or including capital spending. - Looking at a single year when major repairs fall unevenly. Break the ratio into categories such as rates, utilities, insurance, maintenance and management to see which line is changing, and track each property against its own history. URL: https://zansa.co.za/answers/what-is-operating-expense-ratio-for-a-property ### What does Zansa's property dashboard show for a landlord or property manager? Zansa's property dashboard connects lease, billing and maintenance systems into a portfolio view of occupancy, arrears, lease expiries and maintenance backlog. It also sends arrears and lease-expiry alerts so action can be taken before problems grow. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The process starts with a free intro call, then a fixed-fee Data Health Check of one to two weeks, and a Foundation Build, typically four to eight weeks, with fixed price and timeline agreed in writing. Optional support follows. Dashboards are usually in Metabase or Power BI. Data is held in a cloud account owned by the client, usually Google BigQuery, and the client owns everything built. URL: https://zansa.co.za/answers/what-does-zansa-s-property-dashboard-show-for-a-landlord-or-property-manager ### How can alerts for arrears and lease expiries help a property business? Alerts push the right items to the right person at the right time, such as a tenant crossing an overdue threshold or a lease nearing expiry, so action starts earlier than waiting for a monthly report. They turn data into a task list. Effective alerts are: - Based on clear rules, for example arrears over a set number of days or lease expiry within a set number of months. - Sent to a named owner. - Limited in number, so they are not ignored. Commonly misread: - Alerts depend on data quality; a missed payment allocation creates a false arrears alert. - Too many alerts lead to alert fatigue. Zansa builds arrears and lease-expiry alerts into its property dashboards, drawing on lease, billing and maintenance systems that are connected into one portfolio view. URL: https://zansa.co.za/answers/how-can-alerts-for-arrears-and-lease-expiries-help-a-property-business ### How do I measure average repair time and maintenance cost per unit? Average repair time is the mean number of days from a maintenance request being logged to it being completed. Maintenance cost per unit is total maintenance spend divided by the number of units (or area) over the same period. It matters because slow repairs reduce tenant satisfaction and cost per unit shows which buildings are expensive to keep up. Commonly misread: - Averages hide long-running jobs. Report the share completed within target time by priority. - Comparing old and new buildings, or those with different services, as if equal. - Mixing planned maintenance with reactive repairs. Split the cost into planned and reactive, and watch repeat jobs on the same asset. High reactive spend on an old asset can show that replacement is cheaper than continued repair. URL: https://zansa.co.za/answers/how-do-i-measure-average-repair-time-and-maintenance-cost-per-unit ## Retail and e-commerce ### How do I calculate gross margin by sales channel for an online and physical retailer? Gross margin by channel is channel revenue minus the cost of goods sold for that channel's sales, divided by channel revenue. Calculate it separately for each channel, such as web store, shop floor and marketplace, and include costs that belong only to that channel, such as marketplace commission, payment fees and packaging. Formula in words: (channel revenue minus product cost minus channel-specific selling costs) divided by channel revenue. It matters because a blended margin hides channels that look busy but earn little once fees, discounts, shipping subsidies and returns are charged to them. Commonly misread: - Using list price instead of the price actually paid after discounts. - Leaving out free-shipping cost or payment fees on online orders. - Using one average product cost for every channel when stock was bought at different prices. Decide up front which costs sit above the margin line, and apply that rule to every channel. URL: https://zansa.co.za/answers/how-do-i-calculate-gross-margin-by-sales-channel-for-an-online-and-physical-reta ### What is stock cover and how do I calculate it for my shop? Stock cover is how long current stock will last at the current rate of sales. Divide units on hand by average units sold per day, week or month. If you hold 600 units and sell 100 a week, stock cover is six weeks. Formula in words: stock on hand divided by average sales rate over a chosen period. It matters because it shows both overstock, which ties up cash, and understock, which causes lost sales. Compare it to supplier lead time: if cover is shorter than the time a reorder takes to arrive, you will run out. Commonly misread: - Using a sales average that includes a one-off promotion or a stock-out period. - Calculating it only at total level, which hides products with ten weeks of cover next to products with none. - Ignoring stock already on order. Calculate it per product or category, using a recent, representative sales rate. URL: https://zansa.co.za/answers/what-is-stock-cover-and-how-do-i-calculate-it-for-my-shop ### What is sell-through rate and how is it calculated in retail? Sell-through rate is the percentage of the stock you received during a period that you actually sold in that period. Divide units sold by units received (or by opening stock plus receipts), then multiply by 100. Selling 70 of 100 units gives 70 percent. Formula in words: units sold divided by units available to sell, over the same period, expressed as a percentage. It matters most for seasonal and fashion lines, where unsold stock must be marked down. A low figure early in a season is an early warning, and a very high figure may mean you under-bought. Commonly misread: - Mixing the denominator definition between products, so figures cannot be compared. - Judging by the figure alone without the time elapsed since the stock arrived. - Treating a high sell-through as always good, when it may mean lost sales from stock-outs. Pick one denominator definition and keep it fixed. URL: https://zansa.co.za/answers/what-is-sell-through-rate-and-how-is-it-calculated-in-retail ### How is e-commerce conversion rate calculated and why is it often misread? E-commerce conversion rate is the number of orders divided by the number of sessions (or visitors) in the same period, multiplied by 100. If 2,000 sessions produce 50 orders, the conversion rate is 2.5 percent. It shows how effectively traffic turns into purchases. It matters because it separates two problems: not enough traffic and traffic that does not buy. Commonly misread: - Comparing sessions-based and visitors-based rates as if they were the same. - Looking only at the site-wide figure, when mobile, desktop, source and landing page can differ widely. - Ignoring that a campaign bringing cheap, low-intent traffic lowers the rate without harming profit. - Comparing against someone else's published rate; products, prices and traffic mix make that unreliable. Track your own rate by device and traffic source over time, and pair it with average order value and margin. URL: https://zansa.co.za/answers/how-is-e-commerce-conversion-rate-calculated-and-why-is-it-often-misread ### How do I calculate return rate and what does it tell a retailer? Return rate is the number of items (or orders) returned divided by the number of items (or orders) sold in the same period, multiplied by 100. It shows how much sold revenue is reversed and how much handling cost your returns create. Measure by units and also by value, since expensive items may return differently from cheap ones. It matters because returns carry hidden cost: shipping both ways, handling, restocking, and stock that cannot be resold at full price. Commonly misread: - Dividing returns by sales in the same month, when returns arrive weeks after the sale. Use a cohort of orders instead. - Reporting one blended figure, hiding products, sizes or suppliers with high returns. - Ignoring the return reason, which is where the fix usually lies. Segment by product, channel and reason code to find what to change. URL: https://zansa.co.za/answers/how-do-i-calculate-return-rate-and-what-does-it-tell-a-retailer ### How do I reconcile Shopify orders, POS sales and the accounting ledger? Reconcile by matching three things for each period: orders recorded in Shopify or the POS, payments received from the payment provider or bank, and the sales and fee entries in the ledger. Differences are investigated until each order, payout and journal ties to the others. Typical differences to expect: - Payment provider payouts are net of fees and arrive days after the sale. - Refunds and chargebacks fall in a different period from the original sale. - Gift cards, discounts and shipping are booked on different lines. - Timezone cut-offs move sales across month-end. A repeatable approach is to give every order a common key, such as the order number, and compare totals per day and per payout. Zansa builds this order-to-payment-to-ledger reconciliation by connecting Shopify, POS and Xero or Sage, so exceptions are listed rather than hunted for by hand. URL: https://zansa.co.za/answers/how-do-i-reconcile-shopify-orders-pos-sales-and-the-accounting-ledger ### What is customer lifetime value and how do I calculate it simply? Customer lifetime value (CLV) is the total gross profit you expect from a customer over the whole relationship. A simple version multiplies average order value by gross margin percentage, by orders per year, by expected years as a customer. Example with currency units: an average order of 50, a 40 percent margin, three orders per year and two years gives 50 x 0.40 x 3 x 2 = 120. It matters because it sets how much you can sensibly spend to acquire a customer. Commonly misread: - Using revenue instead of margin, which overstates value. - Using averages across all customers when a few repeat buyers drive most of the value. - Projecting years of life from only a few months of data. Where possible, calculate it from actual historical cohorts rather than assumptions, and label clearly whether it is historical or predicted. URL: https://zansa.co.za/answers/what-is-customer-lifetime-value-and-how-do-i-calculate-it-simply ### What is GMROI and how does it help a retailer judge inventory? GMROI, gross margin return on inventory investment, is gross margin divided by average inventory cost over a period. It shows how many currency units of gross profit each unit of money tied up in stock earns, helping compare products or categories. Formula in words: gross margin in currency units divided by average inventory at cost. A result above 1 means the stock earned more gross margin than it cost to hold, before other expenses. A fast-selling low-margin item can beat a slow high-margin one. Commonly misread: - Using retail value for inventory in the denominator rather than cost. - Using year-end stock instead of an average, which distorts seasonal businesses. - Treating it as profit, when overheads have not yet been deducted. Use it to rank categories and decide where to invest more purchasing budget. URL: https://zansa.co.za/answers/what-is-gmroi-and-how-does-it-help-a-retailer-judge-inventory ### How do I measure average order value and avoid being misled by it? Average order value (AOV) is total revenue divided by the number of orders in the period. If 100 orders bring in 5,000 currency units, AOV is 50. It tells you how much a typical basket is worth. It matters because raising AOV through bundles or free-shipping thresholds can lift profit without more traffic. Commonly misread: - A rising AOV can come from a few very large orders. Check the median as well. - Revenue-based AOV ignores discounts, returns and margin. A higher AOV with deeper discounts may earn less. - Mixing wholesale and retail orders in one figure. - Comparing across periods with different promotions. Track AOV alongside gross margin per order and return rate, and segment by channel and customer type. URL: https://zansa.co.za/answers/how-do-i-measure-average-order-value-and-avoid-being-misled-by-it ### How do I calculate customer acquisition cost and payback period for an online store? Customer acquisition cost (CAC) is sales and marketing spend in a period divided by the number of new customers won in that period. Payback period is CAC divided by the gross profit a customer generates per month, giving the months needed to recover the spend. Example: 3,000 currency units spent and 100 new customers gives a CAC of 30. If each new customer yields 10 gross profit a month, payback is three months. It matters because growth that costs more than a customer returns drains cash. Commonly misread: - Counting only advertising spend and leaving out agency fees, tools and staff time. - Counting all orders, including repeat customers, as new customers. - Using revenue instead of gross profit for payback. Calculate it per channel, since blended CAC hides expensive sources. URL: https://zansa.co.za/answers/how-do-i-calculate-customer-acquisition-cost-and-payback-period-for-an-online-st ### What is inventory turnover and how should a small retailer interpret it? Inventory turnover is cost of goods sold divided by average inventory at cost for a period. It shows how many times you sell through and replace your stock. Divide 365 by the result to get days of inventory held. Example: cost of goods sold of 600,000 and average stock of 100,000 gives a turnover of six, or about 61 days of stock. It matters because stock is cash waiting on a shelf. Commonly misread: - Comparing across product types: groceries and furniture naturally turn at very different speeds. - Using a single closing stock figure instead of an average. - Treating higher as always better, as very high turnover can mean frequent stock-outs. Compare your own figure by category and over time rather than against a generic target. URL: https://zansa.co.za/answers/what-is-inventory-turnover-and-how-should-a-small-retailer-interpret-it ### What is repeat purchase rate and how do I measure customer retention in retail? Repeat purchase rate is the percentage of customers who bought in a period and have bought more than once. Divide customers with two or more orders by total customers in the same group. It indicates whether customers come back after the first sale. It matters because repeat customers usually cost less to serve than new ones, so retention drives long-term profit. Commonly misread: - Measuring over too short a window for products that are naturally bought once a year. - Mixing old and new customers, so improvement in new cohorts is hidden. Group customers by first-purchase month. - Counting guest checkouts as different people when the same email or phone buys twice. Cohort tables, showing each month's new customers and how many return later, give a much clearer picture than one overall percentage. URL: https://zansa.co.za/answers/what-is-repeat-purchase-rate-and-how-do-i-measure-customer-retention-in-retail ### How do I measure stock shrinkage and stock variance in a retail business? Shrinkage is the value of stock lost between what your records say you should have and what a physical count finds. Calculate book stock minus counted stock, and express it as a percentage of sales or of stock value. Causes include theft, damage, supplier short deliveries, admin errors and mis-scanned items, so the figure alone does not tell you which one is at work. Commonly misread: - Comparing counts taken at different times of year or by different methods. - Netting overages against shortages, so errors cancel out and hide problems. - Assuming all variance is theft when receiving or transfer errors are common. Count more often for high-value or high-variance lines, record adjustments with reason codes, and review variance by location, product and period. URL: https://zansa.co.za/answers/how-do-i-measure-stock-shrinkage-and-stock-variance-in-a-retail-business ### What does a retail and e-commerce analytics dashboard from Zansa include? Zansa builds retail and e-commerce dashboards that connect Shopify, point-of-sale and Xero or Sage data. The core feature is order-to-payment-to-ledger reconciliation, alongside margin, stock and customer views, so the numbers in operations and in accounts agree. Zansa is a data analytics and data science team for small and mid-sized companies anywhere in the world, based in South Africa and working remotely. The usual process is a free intro call, a fixed-fee Data Health Check of one to two weeks, then a Foundation Build typically taking four to eight weeks, with fixed price and timeline agreed in writing. Dashboards are usually in Metabase or Power BI, on data held in a cloud account the client owns, usually Google BigQuery. Guidance pricing is Health Checks from R25,000 and builds typically R80,000 to R250,000. Optional support starts from R12,000 a month. URL: https://zansa.co.za/answers/what-does-a-retail-and-e-commerce-analytics-dashboard-from-zansa-include