Detect mobile money payment fraud in Nairobi in 2026

The verdict in three sentences

The most common mobile money fraud hacks nothing: it exploits your trust through fake receipts (screenshots), SIM swap and social chargeback. The decisive defence is simple and free: never validate an order on a screenshot, always check the real status via the operator API, which blocks 100 % of fake receipts. Reserve ML scoring for high volumes; for most stores, a few heuristic rules are enough.

The three frauds that drain margins

In Nairobi, a merchant who accepts "proof of payment" over WhatsApp is an easy target. The fraudster sends a screenshot of an M-Pesa or Airtel Money transfer — faked or reversed — and collects the goods before the seller realises the money never arrived.

The striking figure: fake receipts make up 41 % of fraud attempts, and a simple server-side status check neutralises them entirely.

Heuristics vs ML scoring: which to choose

The question is not "rules or machine learning" in the abstract, but "at what volume". Below a few hundred transactions a day, well-set heuristic rules catch almost all fraud at zero cost. Above that, ML scoring on velocity, device and history becomes worthwhile.

SIM swap is detected when the number moves to a new device within a 48-hour window before a sensitive transaction: it is a strong signal of a compromised account. The average cost of an uncontrolled dispute is estimated at 30,000 FCFA (time, goods, fees), which easily justifies a few hours of rule setup.

Mini case study

Ibrahim runs an electronics store in Nairobi and processes 600 mobile money orders a month. He used to validate on screenshots: about 2 % fraud, i.e. 12 lost orders a month. At an average basket of 45,000 FCFA, that was 540,000 FCFA of monthly losses.

By switching to API status verification (blocking the 41 % of fake receipts) and a velocity rule, Ibrahim brings his fraud rate below 0.5 %. Losses fall to around 135,000 FCFA, i.e. over 400,000 FCFA saved each month, for a one-off development fee.

FAQ

Is a payment screenshot valid proof? No, never. 41 % of fraud attempts use fake receipts. Only confirmation via the operator API counts.

Is the fraud rate really higher in Africa? Yes, 2026 order of magnitude: about 1.9 % in African e-commerce versus 0.7 % globally, hence the need for adapted rules.

How do I spot a SIM swap? By watching for a device change tied to the same number in the 48 hours before a transaction. That signal triggers a manual review.

Should I invest in machine learning? Not at first. Below a few hundred transactions a day, heuristic rules are enough. ML becomes worthwhile at high volume.

How much does an uncontrolled dispute cost? On average 30,000 FCFA per dispute (goods, time, fees). A few hours of anti-fraud rules pay off from the first prevented cases.

Let's talk about your project. We set up API status verification and anti-fraud rules matched to your volume. WhatsApp +221 77 596 93 33.