United StatesFinance and insurance2026 – ongoingAffirm
Transformer model for real-time checkout underwriting
Affirm reported 3.4% more completed purchases after deploying transformer-based underwriting
Problem
Affirm wanted to approve more eligible purchases while maintaining credit risk controls. Its existing system declined some applicants with limited credit histories.
Approach
Affirm deployed an internally developed transformer model at US checkout. The model learned from the sequence and timing of credit-history events and approved additional applications that the existing system would have declined.
Results
Takeaway
Affirm compared completed purchases with a control group. It did not publish numerical loan-performance results for the additional approvals.
Categories
- Industry
- Finance and insurance
- Business function
- Finance, risk and legal
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Affirm confirmed that the model was live at US checkout and reported 3.4% more completed purchases against a control group. It said the additional loans performed better than a comparable expansion using its previous models.
Full details are in the original sources.