Affirm (NASDAQ: AFRM) has deployed a transformer-based machine learning model for real-time consumer credit underwriting in the U.S., according to a company statement.

The model, now live at checkout, analyzes patterns in a consumer's credit history, including the order and timing of credit events across accounts. Affirm said the system can identify changes in credit behavior over time without requiring individually designed measures for each pattern.

In its initial deployment, the model approved applications that Affirm's previous system would have declined, including consumers with limited credit histories and no FICO scores. Compared to a control group, the new model produced 3.4% more completed purchases, and those additional loans performed better than a comparable expansion under Affirm's prior machine learning models.

Affirm also developed a proprietary algorithm intended to maintain real-time processing speed while producing the same level of explainability as traditional machine learning models.

"We've steadily accelerated the amount of data we use to train each generation of our underwriting models," said Libor Michalek, Affirm President. "What's exciting about the transformer model architecture is that we can now find new information within the data we already have."

Michalek added: "The goal isn't to approve every transaction, it's to make the right decision for each one. We don't benefit from extending credit that can't be repaid, which means saying yes to more people only works when we get even better at saying no."

Affirm said it has underwritten each purchase individually since its founding, using in-house machine learning models.