Product Manager who turnsambiguity into shipped product.

Sat with 14 users in Yaba and Ikeja and watched them send real money on their own phones and their own networks — no lab, no prototype. Eleven of them had a personal rule for what to do when the spinner hung.

Pulled six months of transaction logs with the data analyst and found the silent-failure window: transfers stuck between 40 and 90 seconds were where every double-send lived.

Wrote the spec around one idea — never show a spinner without a promise. Every pending state got a plain-English status, an expected settlement time, and a locked resend button that unlocked only once the ledger confirmed a genuine failure.

Shipped it behind a flag to 10% of users, held it for three weeks, and read the reversal numbers before we let it out wide.

Spent four Saturdays at Balogun and Mile 12 markets with the credit team, mapping what traders actually kept: restock cadence, supplier WhatsApp receipts, daily till notes, association membership.

Turned that into a signal spec with the credit analysts — twelve alternative inputs, each scored on how hard it was to fake and how hard it was to collect.

Designed the intake around field agents rather than the borrower’s phone, because half of our pilot cohort shared a device. Agents captured evidence offline; the app synced when it found signal.

Ran a 200-shop pilot in two markets with a hard stop at ₦150k exposure per shop so we could be wrong cheaply.

Rode along on eleven delivery runs across the mainland. Timed every interaction with the app and logged where riders stopped and why.

Found the real cost: the app demanded a photo, a signature, a cash-collected amount and a status change as four separate blocking steps, in an order that didn’t match how a handover actually happens.

Reordered the flow to match reality — one screen, glanceable, with cash confirmation defaulted from the order and the photo optional for prepaid drops. Nine taps became three.

Wrote the rollout comms in pidgin and English and recorded a 40-second walkthrough, because a changelog nobody reads is a change nobody makes.

Instrumented the activation funnel properly — six events where there had been one — and let it run for two weeks before touching anything.

Cut the welcome video entirely and replaced it with a text-and-image intro that rendered under 80KB, then made the first lesson downloadable on Wi-Fi for offline play.

Ran the change as a clean A/B rather than a big-bang launch, so the founder had a number to raise on rather than a story.