Predicting Sales With AI: A Crystal Ball for SMEs?

Every January, Njoroge orders too little cement. Every December, he orders too many school bags. His hardware and general shop in Nakuru has been running for eleven years, and he still guesses his stock levels based on a feeling in his stomach and whatever his neighbour is doing. When a sales rep told him AI could "predict the future of his business", he leaned in so far he nearly fell off his stool. So, is AI sales forecasting for SMEs a genuine crystal ball, or just another shiny promise?

What AI forecasting actually does

Forecasting tools look at your past sales and find patterns: which products sell when, how sales change with the seasons, what happens around paydays and holidays. They then use those patterns to estimate what is likely to happen next. More advanced tools can also consider outside factors, like weather or promotions.

This is less magic and more mathematics. A good forecast does not tell you exactly what will happen. It tells you what is likely, with a range, so you can make better decisions about stock, staffing and cash.

Kenyan patterns your data already knows

Even without fancy tools, Kenyan businesses have strong, predictable rhythms that any decent forecast should pick up:

• The January squeeze, when school fees swallow household budgets

• Month-end salary spikes, followed by the quiet mid-month stretch

• Rainy seasons, which change what farmers, builders and transport businesses need

• Festive seasons and the December exodus upcountry

• Back-to-school rushes for uniforms, books and shoes

• Election periods, when many businesses notice spending slows down

If you have kept sales records for a couple of years, these patterns are already sitting in your data, waiting to be noticed.

The catch: AI sales forecasting for SMEs needs clean data

Here is the honest truth. A forecasting tool is only as good as the information you give it. If half your sales are recorded in a notebook, some on M-Pesa, some in a till, and the rest live in the shop attendant's memory, no AI on earth can make sense of it.

Before you think about AI, ask yourself:

• Do I record every sale, including cash and mobile money?

• Can I see sales by product, not just total daily takings?

• Do I have at least a year or two of records in one place?

• Are returns, discounts and stock-outs recorded properly?

If the answer is mostly no, your first project is not AI. It is getting a proper sales and payments system in place. A platform such as PesaVista, which brings your collections and invoices together in one place, gives you the clean transaction history that any future forecast will need.

Start simpler than you think

You do not need to jump straight to a sophisticated AI platform. Many SMEs get most of the benefit from simple steps:

• Export your sales to a spreadsheet and chart them month by month

• Compare this year with last year for your top ten products

• Ask an AI assistant to help you spot patterns or explain a trend, using anonymised figures

• Set minimum stock levels for your best sellers based on past demand

• Review your forecast monthly and adjust as you learn

Once this habit is in place, a more advanced tool will have something real to work with, and you will know enough to judge whether its predictions make sense.

Where forecasts go wrong

Forecasts struggle with surprises: a sudden fuel price change, a new competitor across the road, a viral TikTok that makes one product disappear overnight. They also assume the future looks a bit like the past, which is not always true. Treat any prediction as advice, not instructions. Your experience still matters, and the stomach feeling Njoroge trusts is sometimes picking up on things no spreadsheet can see.

What happened to Njoroge

Njoroge started small. His daughter moved two years of sales from the notebook into a spreadsheet, and they found that cement sales reliably jumped in the dry months while school bags peaked in late December and early January, not November as he had assumed. This year, he ordered differently. He has fewer empty shelves, less dead stock and, for the first time, a little extra cash in February.

Your next step: gather one year of sales, by product, in one place. If you want help building a system that captures that data automatically, talk to Philsoft. The crystal ball is optional; the receipts are not.