Affirm Launches AI-Powered Underwriting Model to Improve Credit Access
- checkouts that uses transformer technology to evaluate consumer payment history sequentially, according to an interview with Affirm President Libor Michalek.
- Older underwriting models used by lenders often struggled with the temporal aspect of financial events, according to Michalek's statements to PYMNTS CEO Karen Webster.
- The new underwriting model demonstrated significant improvements when evaluating consumers with sparse credit histories.
Affirm has deployed a new underwriting model at U.S. checkouts that uses transformer technology to evaluate consumer payment history sequentially, according to an interview with Affirm President Libor Michalek. The technology, which is similar to that used behind large language models, aims to distinguish between consumers with past credit hurdles who have resumed timely payments and those whose financial difficulties are currently escalating.
Transformer Technology Replaces Older Underwriting Methods
Older underwriting models used by lenders often struggled with the temporal aspect of financial events, according to Michalek’s statements to PYMNTS CEO Karen Webster. While legacy systems could count the total number of missed payments or credit events, they lacked the capability to track the precise timing and order in which those events occurred in a customer’s life. To solve this, Affirm integrated its 14 years of lending history into a transformer-based model designed to read payment histories in chronological order. The company spent more than a year testing the technology to ensure it could consistently provide fast, repeatable decisions in under a second while shoppers wait at checkout. The deployed system pairs the transformer model with a traditional machine-learning model, combining payment-sequence insights with established credit, merchant, and user information. According to Affirm, this paired approach outperforms the standalone transformer. The system initially served as a secondary evaluation layer for applicants declined by the legacy system, and it now handles full underwriting responsibilities for new users across approvals and declines.
Impact on Thin Files and Subprime Consumers
The new underwriting model demonstrated significant improvements when evaluating consumers with sparse credit histories. Among individuals who have some activity on their credit reports but lack enough data to generate a FICO score, the model improved Affirm’s capacity to rank who’s likely to miss a first payment by roughly 2.1 times compared to the previous version of its older underwriting framework. Michalek explained that even limited credit files possess a recognizable sequence. Data from credit bureaus, Affirm’s internal records, and customer-selected cash flow information help separate consumers who look identical on paper under older, sparser evaluation methods. Despite the ability to extend credit without a traditional FICO score, Affirm confirmed it will continue reporting customer repayment behaviors to credit bureaus. This dual approach helps consumers build traditional credit profiles while utilizing alternative underwriting data for immediate point-of-sale decisions.
Everyday Spending and Household Working Capital
The deployment of the transformer-based underwriting model coincides with a shift in how consumers utilize buy now, pay later services. Recent earnings reports from Affirm indicate an increase in smaller purchase amounts, pointing to everyday use across grocery shopping, clothing purchases, and car repairs. This frequent usage functions as short-term working capital for households attempting to align expenses with pay periods. However, multiple small purchases with overlapping repayment dates present new risk evaluation challenges, requiring the underwriting model to assess how a new financial obligation fits alongside existing commitments. The Affirm Card reflects this everyday utility by offering users a spending limit paired with structured payment plans and options to select fast payoffs or low interest rates. Underwriting guardrails determine individual spending power, which can expand or remain steady depending on demonstrated repayment behavior over time. Michalek indicated that future metrics regarding approvals, delinquencies, customer growth, and overall usage will determine the success of the new underwriting model in upcoming quarterly financial results.
