The questions shops cannot answer about themselves
Ask a shop owner which items sold the most last month and most can tell you. Ask which items made the most money after what they paid the supplier, and the room goes quiet.
It is not carelessness. That question needs sales data and purchase data joined together, and in most shops those live in different places or different heads. So owners go on feel, and feel is reliably right about the top few items and unreliable about everything below them - which is where the slow leaks are.
The same applies to a whole class of questions. Which supplier has quietly raised prices? Which stock has been sitting untouched since March? Which category's margin is drifting down? The data to answer all of these already exists in the shop. There is just no practical way to ask.
What an AI business analyst actually does
Think of it as an analyst who has read every line of your records and is available whenever you have a question. You ask in ordinary language and it answers with a real report - a table, a chart, something you can export or save and rerun next month.
- Revenue and gross margin by item and by category, over any period you choose.
- Stock turnover and how many days of cover you have left on each item.
- Dead and slow-moving stock, with the capital tied up in it stated in shillings.
- Your best and worst movers, ranked on profit rather than on units shifted.
- Price variance where several suppliers sell you the same item.
- Who owes you money and for how long, and where your till and cash position stands.
Advice is the point, reports are the mechanism
A chart tells you what happened. An owner running a busy shop needs the next sentence: so what should I do about it?
That is where a good analyst earns its keep. Not "sales of power tools are down 14%" but "your margin on power tools fell from 22% to 17% after your supplier's price moved in April, and your selling price has not moved since". Not "these items are slow" but "these twelve items have not sold in ninety days and are holding KES 340,000 you could be trading with".
The rule to hold any tool to is that every recommendation must name the numbers it rests on. Advice without figures is just an opinion generated at speed, and you already have enough of those.
The brief matters more than the questions
Owners expect the value to be in asking questions. In practice most of it comes from the answers that arrive without being asked for.
A short scheduled brief - what moved, what is at risk, three things worth doing - lands before the shop gets busy. Nobody has to remember to check anything. Over a few weeks it stops being a notification and becomes a routine, and that routine is what catches slow problems while they are still small.
Alerts are the other half. A brief is time-based; an alert fires the moment something breaks. A fast-mover about to run out, a supplier's price jumping, a till that does not reconcile. The brief builds the habit; the alerts prevent the losses.
What it needs from you first
This is the part that gets skipped in sales conversations, so here it is plainly. An AI analyst is a lens on your data. A lens on blurred data produces a sharper picture of a blur.
If purchases are not recorded, no tool can compute your margin. If stock counts are adjusted informally, no tool can tell you what is really on the shelf. If three people record sales three different ways, no tool can compare months honestly.
The sequence that works is: get the recording right in a system that suits how your shop actually trades, run it properly for a season, then add analysis on top. Businesses that do it in that order get something trustworthy. Businesses that buy the AI first get fast, confident, wrong answers.
Questions to ask before you buy one
Treat this like any other system purchase and the choice gets much easier.
- Where does each number come from - can the tool show you its working?
- What data leaves my premises, exactly? Ask for specifics, not reassurance.
- Can it write to my records, and if so, does it need my approval first?
- Does it reach me where I already am, or does it need me to log in somewhere?
- What happens when it does not know? A tool that says so is safer than one that always has an answer.
- Who fixes it when it is wrong, and how quickly can they reach me?
How Upeosoft approaches this
We build the systems that record the business first - stock, sales, purchases, M-Pesa and eTIMS - because that is the foundation everything else stands on. Then we add the analysis, so the answers rest on records that are actually complete.
Our own AI analyst work is built around a simple constraint: it reasons over a defined set of business measures that a human has checked, rather than inventing its own arithmetic against your database. That is a slower thing to build and a much safer thing to rely on.
If you want to know whether your business is at the point where this pays for itself, talk to us. Sometimes the honest answer is that fixing how your stock is recorded will earn you more this year than any AI will.
