This morning, a mama mboga needed Sh5,000 to restock. She tapped her phone, and 90 seconds later the money hit her M-Pesa. No logbook, no payslip, no guarantor. That same morning, a credit committee sat down to decide whether to release Sh2 million of stock to a distributor they'd supplied for three years.
They opened his audited accounts from 18 months ago, a bank reference, and a trade reference he'd chosen himself. Then they argued for an hour.
Look at those two decisions. The app lending 5,000 read 100 live signals about how that woman handles money. The committee vetting Sh2 million; read a story the customer curated about a version of himself that expired a year and a half ago. The small loan was the smarter one. That gap is where credit is going, and most corporates are on the wrong side of it.
A sharper question is winning. Not what you own, but what you do. The answer lives in behaviour.
The digital lenders built everything on that shift. Kenya now has more than 250 licensed digital credit providers, an industry that didn't exist 15 years ago.
By mid-2025, their loan book had passed the entire portfolio of the country's microfinance banks. Around eight million Kenyans, one in six, now lean on their phones for credit. They won by reading the present while banks audited the past.
Here's the part corporates must sit with. A digital lender studies how money moves through your M-Pesa and how faithfully you cleared the last loan.
A credit committee studies a document the customer prepared about a period already gone.
So, the first correction is obvious. Read behaviour, not just financials. But stop there, and you've only caught up to 2022, because reading behaviour isn't the same as understanding it.
The digital lenders read correlational behaviour. They notice that people who top up airtime on a steady rhythm tend to repay, so they lend to them. What they don't know is why.
Maybe the rhythm reflects discipline. Maybe it reflects a salaried job, and the day the job goes, the signal and the repayment vanish together.
Correlation tells you what patterns tend to pay. It can't tell you which pattern causes payment and which is just keeping it company. That difference is invisible in good times and brutal in bad ones.
When one shock hits the whole market at once, correlation-based models fail in a herd. Loan books that had never been wrong got everything wrong during the pandemic.
The future of scoring is causal. It asks why you pay, not just what you do. A model built on causes is harder to game, because you can't fake a reason the way you mimic a pattern, and it holds up when the ground shifts.
There's a quiet clue about where causal knowledge comes from. The auditor reads the entrance. The digital lender reads the present. Recovery reads the exit, the only place you learn what truly caused a solid account to collapse. That knowledge is the one dataset correlation can't buy.
A warning belongs here. The same data that frees people can punish them, through brutal rates and apps that smile while lending and bare their teeth while collecting.
The winners won't be the fastest readers of data. They'll be the most disciplined. Collateral told you what a customer had. Behaviour tells you what they do. Only causality tells you what they'll do next, which is the only thing a credit decision was ever about.
-The writer is a CPA and is the founder and CEO of Marathon Debt Recovery Ltd
