Affirm launches AI model to boost U.S. credit approvals

On 18 September, Affirm announced the launch of a new transformer-based machine learning model for real-time credit underwriting at U.S. checkouts. The system analyzes the sequence and timing of events across a borrower's credit history. It draws on 14 years of proprietary transaction and repayment data to deliver fast and explainable credit decisions.

During initial tests, the model approved applications that Affirm's prior system rejected, including borrowers with limited credit histories and no FICO scores. These incremental approvals generated 3.4% more completed purchases than the control group. The resulting loans also delivered stronger early credit performance than comparable expansions under the previous system.

The higher completion rate could lift loan originations, merchant activity and revenue, though Affirm did not provide a specific dollar forecast. The company competes in digital payments and lending against firms such as Upstart Holdings, which connects borrowers to over 100 banks, and PayPal Holdings, which offers checkout financing ranging from $49 to $10,000.

Shares of Affirm have gained 58.2% over the past six months, outperforming the industry gain of 20.2%. The stock trades at a forward price-to-sales multiple of 4.1, compared to an industry average of 4.2. Zacks consensus estimates project earnings of $1.87 per share for 2026, followed by 53.3% earnings growth the next year.