Affirm Bets on Where Borrowers Are Headed

Affirm, the online lending platform, has introduced a new underwriting model to better assess borrowers' creditworthiness based on their spending habits and repayment history. The new model, which utilizes transformer technology, takes into account the timing of multiple purchases and credit events in a borrower's life. This is a significant improvement over traditional models that only consider the number of missed payments and a customer's current credit score.

According to Libor Michalek, Affirm's President, the new model can identify five out of ten applicants that an older model would have rejected, as it can better predict which borrowers are likely to repay their debts on time. This improvement is particularly noticeable for consumers with thin files, who often have limited credit history and may have been denied credit in the past. The new model can distinguish between these borrowers and provide them with access to credit that they would have previously been denied.

Michalek explained that the new model draws on Affirm's 14 years of lending data and uses transformer technology, similar to large language models, to analyze payment history and predict future behavior. The model was in testing for over a year before being launched, ensuring that it could provide accurate predictions in real-time.

The partnership between the transformer's analysis of payment history and a traditional machine-learning model's assessment of credit, merchant, and user information proved to be more effective than the transformer alone.

The new underwriting model is particularly relevant for Affirm's "pay later" option, which allows consumers to match their purchases to their income by breaking down payments into manageable installments. This type of payment structure can help consumers avoid overspending and manage their cash flow more effectively. As a result, the model can better assess the risk associated with multiple payments due around the same time, making it easier for consumers to access credit when they need it most.