Affirm has launched a new transformer-based underwriting model for U.S. checkouts. The system analyzes the sequence and timing of events in a user's credit history rather than relying solely on traditional credit scores. This approach allows the company to approve applicants with limited credit files or no FICO scores.
The model is built on 14 years of transaction-level data. In initial tests, it increased completed purchases by 3.4% compared to a control group while maintaining comparable risk levels. These loans performed better than those approved under previous machine learning models used by the company.
The company reported significant growth in its fiscal fourth quarter, processing 53 million transactions, a 41% increase year-over-year. Gross merchandise volume reached $14,1 billion, up 36%. Meanwhile, the average order value fell by 4%, and the number of active consumers grew by 21% to 27,8 million.
To maintain performance at checkout, Affirm developed a proprietary algorithm that ensures the transformer model remains fast and provides explainable results. The company continues to monitor the model to ensure accuracy and reliability in its lending decisions.
