AI for West African business — beyond the hype.
The conference-circuit version of AI sells you a future. The version that earns its keep in Lagos and Accra today is more boring and more useful. Here's where the line actually is.
Every AI pitch deck in the region looks roughly the same in 2026. Chatbot on the homepage, "intelligent" dashboard somewhere, a vague promise about anomaly detection. The deck doesn't lie — those things exist — but it doesn't tell you which of them move the P&L and which are theatre. Having built against the real work for two years, here is the honest split.
Where AI pays for itself today.
Reconciliation and matching. If your business touches more than 500 payments a week, AI-assisted matching of bank statements, POS batches and customer ledgers is a clean win. We see 60–80% of routine reconciliation go from a three-person morning to a supervised review by lunch. It's not glamorous. It works.
Document understanding at the edge. Invoices, waybills, bank drafts, NIN slips, CAC documents. OCR plus an LLM that actually knows Nigerian document conventions is a different product from a generic extraction tool. Our customers are routinely removing 2–3 data-entry roles per 1,000 documents per day.
Field-rep decision support. "This customer hasn't paid in 28 days, stock at mama Ada is below reorder, here's the script." That kind of next-best-action on a low-bandwidth phone turns out to be where AI is most immediately felt by a rep in the field.
Anomaly flagging in ops. Detecting that a fleet vehicle is drifting off its ordinary fuel-efficiency curve, or that a branch is posting negative variances in a suspicious pattern. Boring, reliable, measurable.
Where AI still doesn't deliver (honestly).
Fully autonomous customer service in Nigerian English and Pidgin. It's close. It's not there. A hybrid model — AI handles 70%, the last 30% routes to a human with the AI's summary — is the working pattern in 2026. Anyone promising full automation is underestimating the linguistic range of real customers.
Generalist "business analyst" agents. The demo where an AI answers every question about your business from a chat box is a hard problem on clean data, and most businesses do not have clean data. Start with three or four scoped dashboards that each do one thing well. Then widen.
Forecasting in genuinely noisy markets. Short-horizon predictions (7–30 days) on high-volume items work. Long-horizon anything against the backdrop of an FX-volatile, policy-sensitive economy is closer to astrology than science. Use it for narrative, not for commitments.
The deciding question.
When a team asks us whether to adopt AI for a use case, we ask one question back: how would you measure the result in 90 days without talking about the tool? If the answer is "it'll make us more efficient" — soft. If the answer is "reconciliation staff hours fall from 180 to 40 per month" — sharp. Build the second kind.
If you want a half-hour conversation about what's worth doing with AI in your specific operation — and what isn't — book a slot. We'll tell you when we think it's not yet ready, too. That's the honest version of the pitch.