Work · Kaleidofin
Dada Biashara: a GenAI financial coach on WhatsApp
A WhatsApp coach in English, Swahili and Sheng that turns M-Pesa activity into a bankable credit file. Coaching lifted approvals by a third: for every three users approved on the first check, one more got there after coaching.
Context
Market traders, salon owners and food kiosk operators in Kenya run on M-Pesa and cash. Lenders can’t see their income, so they stay outside formal credit or borrow on poor terms. Most will never download a lending app.
The hard call
Coach first, lend second. The easy product is a credit check that says yes or no. I chose to treat a decline as the start of a plan: Dada tells her what is holding her back and coaches her toward approval. That costs more effort per user and delays revenue from people who don’t qualify today. It also turns a declined applicant into a future borrower who trusts the lender, and that is where the extra third of approvals came from.
What I did
- Put the product on WhatsApp instead of an app: nothing to install, works on any phone.
- Supported English, Swahili and Sheng from day one.
- Used four consented data sources: M-Pesa statements, bank statements, credit bureau data and her own answers.
- Built an M-Pesa categorisation framework around operating and capital inflows and outflows.
- Designed a credit framework combining an ML probability-of-default model, a cash-flow rule overlay and a tier-calibrated decision matrix.
- Built to Kenya’s Data Protection Act 2019 and the Credit Reference Bureau Regulations 2020.
- Ran the pilot through a WhatsApp business provider rather than a direct Meta integration, to reach users faster.
My role
I led product strategy and the build: the PRD, the credit framework, the M-Pesa categorisation, the regulatory work and the architecture choices.
Results
| Measure | Result |
|---|---|
| Approvals | For every three users approved on the first check, one more approved after coaching |
| Cash flow | Improved for a good share of users |
| For the lender | A credit-readiness assessment and coaching plan built on richer data than bureau data alone |
What I take from it
A decline is a product moment. If you can tell someone exactly what to change, you keep the customer and earn the next loan.
Next case: From custom models to a multi-product lending platform