Why Most AI Pilots in African Businesses Stall Before They Start

Ask most African businesses if they’re interested in AI, and the answer is yes, often enthusiastically. Ask how many pilots that enthusiasm actually turned into something running in production six months later, and the number drops sharply. The AI itself is rarely the reason for the gap. In nearly every stalled pilot we’ve been asked to look at, the real obstacle was sitting one layer down: the data the AI was supposed to work from.

A typical mid-sized business here runs its finances through one system, its stock through another, and keeps a set of manually maintained spreadsheets that reconcile the two because the two systems were never built to talk to each other. Each of those sources is internally consistent but was never designed with the others in mind. An AI analysis layer, however capable, cannot produce a trustworthy answer from data that disagrees with itself, and it especially cannot flag that disagreement usefully unless someone has already mapped what “agreement” is supposed to look like across those sources.

The unglamorous work that actually unlocks value

This is why the most valuable early step in an AI engagement is rarely the AI model at all. It’s a careful reconciliation pass across whatever exports the business already has, done once, properly, so that the true picture, real receivables net of internal transfers, real stock value across every warehouse, real margin by product line, is established and agreed before anything gets automated on top of it. In one engagement, that reconciliation step alone surfaced that internal balances between related entities were inflating the reported debtors book by a meaningful margin, and that two existing reports disagreed with each other by several million dollars on stock value. Neither of those findings required AI. Both had been sitting there, unexamined, because nobody had been positioned to check the sources against each other until someone was asked to build something on top of them.

Once that reconciliation exists, the AI layer’s job becomes much more tractable: keep the sources aligned going forward, flag the moment they drift apart again, and free the finance or operations team from redoing that reconciliation by hand every month. That’s a realistic, valuable first deliverable. It’s also a much less exciting pitch than “AI dashboard,” which is exactly why it tends to get skipped by teams moving fast, and exactly why skipping it is the most common reason pilots stall.

The practical implication for any business considering a first AI project: resist the temptation to start with the most visible, most impressive-sounding use case. Start by asking whether the data underneath it can currently be trusted, and if the honest answer is no, start there instead. It’s less exciting in month one, and it’s the difference between a pilot that quietly becomes permanent infrastructure and one that becomes a slide in a deck nobody looks at again.