Photo: Tima Miroshnichenko / Pexels
The buy-now-pay-later giant is borrowing the same AI architecture behind ChatGPT to decide who gets approved for a loan, and how fast.
Affirm is bringing the same class of AI that powers large language models into consumer lending. The company unveiled a transformer-based machine learning model designed for real-time, transaction-level underwriting, a move that could reshape how the entire buy-now-pay-later industry thinks about credit risk.
The model, showcased during Affirm’s Investor Forum, represents roughly 18 months of development work. In experimental testing, it outperformed the company’s existing mainline underwriting system, establishing what Affirm considers a new performance baseline for its credit decisioning engine.
What transformers bring to lending #
Transformer architecture is the backbone of models like ChatGPT and Gemini. Its defining feature is something called an “attention mechanism,” which lets the model weigh different pieces of input data against each other to find patterns that simpler models miss. In Affirm’s case, it means understanding which financial signals matter most for a specific borrower at a specific moment.
For Affirm, this translates into better risk assessment and smarter term selection on every individual transaction. Rather than relying on a static credit score that might be months out of date, the system evaluates borrowers in real time, pulling from a much richer set of signals to make its approval decision. As of January 2026, Affirm had already updated its underwriting engine to incorporate real-time financial signals from linked bank accounts, including current bank balances. That update alone improved underwriting flexibility, pushing approval rates higher while keeping delinquency levels under control. The transformer model takes that same philosophy and supercharges it with a fundamentally more powerful architecture.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
A 13-year data advantage #
Founded in 2012, the company has spent over 13 years building a proprietary dataset that now spans more than 50 million underwritten consumers and over $100 billion in originated loans.
Most competitors in the BNPL space outsource chunks of their credit engineering. Affirm keeps its machine learning and credit engineering teams entirely in-house, which gives it tighter control over model development and iteration cycles.
The company has long positioned itself as a data-first lender rather than a traditional credit-score-first lender. Where a conventional underwriter might pull a FICO score and call it a day, Affirm’s system evaluates each transaction individually, considering the merchant, the purchase amount, the borrower’s recent financial behavior, and now, with the transformer model, the complex interdependencies between all of those factors simultaneously.
What this means for the BNPL landscape #
A lender that approves too aggressively eats losses. One that approves too conservatively leaves revenue on the table and cedes market share. The sweet spot, higher approval rates without higher delinquency, is exactly what Affirm claims its updated models deliver.
For investors tracking Affirm (NASDAQ: AFRM), the transformer rollout signals that the company is investing heavily in the infrastructure that drives its core economics. Underwriting accuracy directly impacts unit economics in lending. Better models mean fewer defaults, which means better margins. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our