Affirm is looking inward at its customer base for the next leg of its lending business.
Last week, the BNPL firm deployed a new transformer-based underwriting model across its U.S. checkout. The model is designed to identify patterns in a consumer’s credit history that conventional models can miss, particularly the timing and sequence of credit events, and has already enabled Affirm to approve some applicants that its previous system would have declined.
Affirm says those additional approvals generated 3.4% more completed purchases than a control group at comparable risk levels, while performing better than a similar expansion under its previous machine-learning models.
The company that started with point-of-sale installment lending has spent several years building adjacent pieces around that core: the Affirm Card, its app and marketplace, and the Affirm Money Account savings product. Consumers today are using Affirm more frequently and for smaller purchases. In fiscal 2026, active consumers reached 27.8 million, up 21%, while transactions per active consumer rose from 5.8 to 7.0.
This makes the quality of the underwriting engine underneath Affirm’s products increasingly important.
Seeing the trajectory beneath the score
A credit score is useful because it compresses a complicated financial history into a single number. But compression can also obscure context. Two consumers may have similar missed-payment histories but very different trajectories: one recovering from a bad year, the other falling further behind. Traditional models can miss those differences, which is where Affirm’s new underwriting model is taking a deeper look.
“The time dimension of multiple purchases, multiple credit events in a customer’s life wasn’t being represented with particularly high fidelity,” Affirm President Libor Michalek said in a recent podcast.
…
