Kenya's banks now watch every transaction. The loans are still going bad.

Kenya's banks are now watching their borrowers every day — every mobile-money transfer, every electricity payment, every grocery purchase. The Central Bank of Kenya's own data says the loans are still going bad.

In the first quarter of 2026, the stock of non-performing loans in the Kenyan banking sector rose by Sh21 billion, from Sh674.4 billion in December to Sh695.4 billion by the end of March. The default ratio climbed to 15.6 percent from 15.4. This happened while interest rates were falling.

The lenders tell a different story. They say machine learning, natural language processing, and large language models now let them monitor borrowers continuously after a loan is disbursed — scoring mobile money flows, utility payments, and merchant activity to predict financial distress before a payment is missed. The pitch is that early intervention will protect both bank and borrower.

So whose intervention is it, exactly?

"Increases in NPLs were noted in the personal and household, trade, agriculture and manufacturing sectors," Central Bank Governor Kamau Thugge said. Those are the same informal workers and small businesses whose digital footprints the new models are designed to read.

The asymmetry is the story. When a Kenyan bank now flags a borrower before they miss a payment, the bank sees a risk to manage. The borrower sees a score that can move without explanation, a credit line that can be cut off, a phone call that can come at any hour. Continuous algorithmic monitoring moves the line of intervention earlier — but earlier for whom?

If the AI can flag a borrower before they miss a payment, will that change what happens to the borrowers the Governor named? The lenders' predictive claim and the regulator's headline number are sitting in the same document, telling a more complicated story.