Affirm has unveiled a transformer-based machine learning model for real-time, transaction-level underwriting at U.S. buy-now-pay-later checkouts. Showcased at the company’s Investor Forum after roughly 18 months of development, the model outperformed Affirm’s existing mainline underwriting system in experimental testing and approves some borrowers the previous system would have declined, including people with thin credit files or no FICO scores. Affirm says the approach evaluates current, transaction-specific signals to assess risk and select terms, building on underwriting updates that incorporated linked-bank information, including current balances, by January 2026.