Affirm Boosts Loan Approvals with Transformer-Based Underwriting Model
Event summary
- Affirm launched a transformer-based machine learning model for real-time underwriting, live at U.S. checkout as of September 17, 2026.
- The model approved 3.4% more eligible applications than the previous system, particularly for consumers with limited credit histories.
- Approved loans performed better than comparable expansions under prior models.
- The model identifies patterns within and across credit accounts, improving signal extraction from existing data.
The big picture
Affirm’s new transformer-based model represents a strategic shift in leveraging existing data for more precise underwriting. This aligns with broader industry trends toward AI-driven credit decisions, particularly for consumers with thin credit files. The model’s ability to identify temporal patterns in credit histories could set a new standard for responsible lending in fintech.
What we're watching
- Model Performance
- Whether Affirm can sustain the 3.4% increase in approvals while maintaining loan performance metrics.
- Competitive Edge
- How this advancement positions Affirm against competitors in real-time underwriting.
- Regulatory Scrutiny
- The pace at which regulators may examine Affirm’s use of transformer models in credit decisions.
Related topics