Banks explore AI-powered early warning system to curb bad loans

The platform analyses payment histories and market trends to flag high-risk accounts. [Courtesy]

Kenyan commercial banks are in talks with US technology firm IronOne Technologies to deploy an artificial intelligence platform designed to predict borrower defaults up to six months in advance, allowing lenders to intervene before loans sour.

The proposed system, dubbed "Smart Delinquency Predictor", would enable banks and microfinance institutions to shift from reactive debt management to proactive risk prevention, industry executives said.

The platform analyses payment histories and market trends to flag high-risk accounts, giving bank managers time to restructure loans, initiate early recovery or redirect resources away from toxic portfolios.

IronOne Technologies, headquartered in Texas with a Nairobi presence and a tech hub in Colombo, specialises in capital markets trading platforms, AI and machine learning solutions, and bespoke mobile applications. It has partnered with Fidem Financial LLC, a US-based credit and consumer lending firm, to install the system in banks.

Unlike traditional credit assessment models that evaluate borrowers at the point of loan approval, the AI solution continuously monitors both internal and external risk factors throughout the loan lifecycle, said Praveen De Silva, a director in charge of Global Business Solutions and Client Success at IronOne Technologies.

"Bankers look at the history of the borrowers before issuing a loan," De Silva told The Standard, adding,  "Our AI solution continuously monitors borrowers' transaction and repayment patterns while analysing external variables such as economic conditions, unemployment trends, and geopolitical developments that may affect their ability to service loans. Each market shock becomes an additional data layer, allowing us to predict emerging risks well before they translate into defaults."

The platform captures personal spending patterns, repayment records with other credit providers, and external factors including inflation, interest rates, exchange rate movements and their expected impact on borrowers' income over six to 12 months. It also factors in global shocks such as oil prices, commodity volatility and geopolitical tensions.

Borrowers receive scores ranging from 0 to 999, with zero representing the highest risk.

The platform integrates with existing core banking systems, requiring no replacement of current infrastructure, and can be configured for all loan types including personal, MSME, agricultural, mortgage, trade finance and digital loans. De Silva said implementation takes four to six months after contract signing.

Banks in the United States, the United Kingdom and Asia have already adopted the model. If successful, Kenya would become the first African country to deploy the platform for non-performing loan management, potentially positioning its banking sector as a continental leader in AI-driven credit risk management.

The push comes as bad loans emerge as a growing concern for Kenyan lenders. Central Bank of Kenya data shows gross non-performing loans jumped 59 per cent in five years to Sh674.4 billion in 2025, from Sh424.1 billion in 2020. The trend has continued into 2026, with NPLs rising to Sh680.3 billion in January and Sh688.4 billion in February.

"The question is not whether a loan portfolio carries risk; it is whether that risk can be identified six months in advance rather than only after losses have already occurred," De Silva said. "What if you were to know exactly which debts will turn risky 6-12 months in advance? You won't just be reacting to defaults, you will be predicting them."

The platform is not meant to disrupt banks' existing IT systems but to function as an additional component running parallel to them.

IronOne already deploys back-office solutions to stockbrokers and investment banks trading on the Nairobi Securities Exchange.

Banks combine large volumes of high-value decision-making, regulated data and long-standing customer relationships. Industry experts say AI providers will become a permanent part of the banking technology landscape, with their models supporting more processes and influencing more decisions. What remains to be decided is how each institution manages that relationship.

The banks that benefit most according to experts will treat AI provider relationships the way they already treat other critical infrastructure decisions: with clear standards, deliberate redundancy and an operating model that assumes providers will change over time. Whichever model produces the answer, accountability for the outcome sits with the bank, experts add.