The encyclopedia · Finance & Accounting · Financial decision · 2018–2025
Tala scored 'unbankable' Kenyans from phone data and tested its own rejections
Tala lent to people traditional banks refused, using SMS, mobility and contacts—then ran experiments to prove the loans help.
Tala
the move
Tala is a fintech lender that determines creditworthiness from borrowers' mobile phone behavior—SMS transaction amounts, movement across cell towers and cities, and the size and strength of their social networks—rather than bank records most Kenyans do not have.
Starting in 2018, Tala randomly approved some applications its own scoring model would have rejected, part of a reject-inference experiment. In a study sample of 20,000+ applications from April 2018 to January 2022, about 5,000 borderline borrowers were approved; average loans were $36 for about 28 days at a ~15% fee.
Researchers found the randomly approved borrowers saw higher mobile-money balances, income, employment and social networks than comparable rejected applicants. The study, published in The Accounting Review in 2025, suggests traditional credit cutoffs can be too conservative where small loans unlock informal-economy opportunities.
why it works
- Phone data substitutes for missing credit files
- Random approvals turn lending policy into a controlled experiment
- Small short-term loans fit informal-economy cash flow
- Results can re-calibrate who the algorithm accepts
what transfers
When your model says no, test the rejected margin: randomized approval of borderline cases turns a guess about conservatism into evidence.
what came after
Tala grew to about 10 million borrowers globally, 3.5 million of them in Kenya, where roughly a quarter of adults had taken a digital loan by 2018. The Accounting Review paper added academic validation; the sector still draws caution over high-interest digital lenders that can prey on the poor.
references
- How $40 Loans Lifted Lives in Kenya
- Digital Lending and Financial Well-Being: Through the Lens of Mobile Phone Data
spotted an error? The archive wants to know.