Your business already collects data. Analysing it is the part that pays.

Every business in Nairobi is already running a data operation, whether it admits it or not. The till prints a record of every sale. The M-PESA statement logs every payment, timestamped to the second. WhatsApp holds every customer question you have ever been asked. The delivery book knows which routes run late. The problem is almost never a shortage of data. The problem is that nobody is looking at it.

The 90 percent you never look at

Researchers have a name for this: dark data. IBM defines it as the information a company collects in the normal course of business and then never uses for anything, and the estimates are brutal. Somewhere between 80 and 90 percent of the data a typical organisation holds is dark. IDC estimates that 90 percent of unstructured data, the messages, documents, photos and free-text notes that make up most of what a business generates, is never analysed at all. Gartner-adjacent studies put the share of data most companies actually analyse at around one percent.

Those numbers come from studies of large enterprises, but the pattern is worse in a small business, not better, because nothing is instrumented on purpose. The sales history sits inside a till nobody exports. The payment record sits in a statement PDF. Customer complaints sit in a personal phone. Each system is a silo, and no silo talks to another, so the questions that matter most stay unanswered: which product actually made money last month, which customers came back, which day of the week deserves more stock.

What analysis changes, in numbers

The case for fixing this is not sentimental. Industry studies year after year find the same shape: analytics projects routinely pay for themselves within the first year, with returns in the low hundreds of percent over three. Firms that run on evidence rather than recollection report revenue gains in the 15 to 20 percent range from better pricing, stock and marketing decisions, and meaningful reductions in operating cost from cutting what the numbers show is not working. Organisations that make decisions from data are also simply faster, several times faster by most measures, because a question that used to trigger a week of arguing becomes a thirty-second glance at a dashboard.

That speed is the part we see most clearly in our own engagements. The first dashboard rarely tells a client something nobody suspected. It confirms one suspicion, kills another, and ends three standing arguments. That alone changes how the next quarter is planned.

The Kenyan version of this problem

Kenya is an unusually good place to be a small business with data ambitions, because the payment layer is already digital. With mobile money penetration at 98 percent, the average Kenyan SME has a machine-readable record of nearly every shilling in and out, which is a starting position European corner shops would envy. The raw material exists. What is missing is the pipeline: getting the till, the statements, the spreadsheet and the WhatsApp orders into one place where they can be compared.

That is also why the excuses travel badly here. A business that takes M-PESA does not need a data collection project. It needs a data consolidation one, and consolidation is cheaper, faster and much less risky.

What a data analysis service actually does

The phrase sounds grander than the work. A good engagement is not "big data", it does not start with hiring a data scientist, and it should not start with buying software. Ours follow the same four steps every time.

• Questions before tools. We start by writing down the five decisions the owner makes every week and what information each one needs. If a number does not serve a recurring decision, we do not build it.
• One source of truth. Exports from the till, the bank, M-PESA and the order book get cleaned and joined so that "sales" means one thing everywhere. Most of the value is created in this unglamorous step.
• Dashboards people actually open. A handful of live views, one per decision, readable on a phone in the matatu. We deliberately ship fewer charts than anyone expects, because ten charts get read and forty get ignored.
• A decision loop. A monthly session where the numbers are read, one change is made, and the effect is checked the following month. Analysis without a loop is decoration.

We wrote before about the self-hosted analytics stack we ship by default for website data. The same philosophy applies to business data: own the pipeline, keep it simple, and measure only what you will act on.

Where to start

You do not need a strategy document. You need honest answers to three questions. What decision do you repeat every week? Where does the data for that decision currently live? And how long would it take you, right now, to say which product made you the most money last month? If the last answer is more than a minute, the gap between you and the answer is the project, and it is usually a small one.

Start with one decision, one consolidated dataset and one dashboard, and let the results argue for the rest. If you want help building that first pipeline, our data analytics team does exactly this, and the first conversation about what your numbers could tell you costs nothing.