Affirm has launched a new underwriting model at U.S. checkouts that utilizes transformer technology to analyze consumer payment histories. By incorporating 14 years of lending data, the system distinguishes between borrowers who have recovered from past financial issues and those currently facing increasing payment difficulties.
The new approach combines transformer models with traditional machine learning to evaluate credit, merchant, and user data. This hybrid system aims to improve risk assessment, particularly for consumers who lack a traditional FICO score. Affirm reported that the model improved its ability to rank potential payment defaults by 2.1 times for this specific demographic.
The technology is designed to provide consistent, explainable decisions in under a second during the checkout process. Initially used to re-evaluate previously declined applications, the model now handles full underwriting for new users. Affirm intends to monitor performance through metrics such as approval rates, delinquency levels, and overall customer growth.
As consumers increasingly use Affirm for everyday purchases like groceries and car repairs, the company faces the challenge of managing multiple concurrent payment obligations. The new underwriting model helps set dynamic spending limits for the Affirm Card, allowing the system to adjust based on individual repayment behavior and financial patterns.
