# Affirm's new underwriting sees credit history clearly

> Affirm's transformer-based model approves borrowers traditional credit scores would reject, including those with thin files and no FICO score.
- Title: Affirm's new underwriting sees credit history clearly · Pymnts
- Summary: Affirm's transformer-based model approves borrowers traditional credit scores would reject, including those with thin files and no FICO score. By analyzing…
- Keywords: affirm, buy now pay later, underwriting, machine learning, credit scoring, transformer model, lending, fintech, checkout, credit history, business, finance
- Source: Pymnts — https://pymnts.com/bnpl/2026/exclusive-affirm-rebuilds-underwriting-to-approve-borrowers-a-credit-score-cant-see
- Author: PYMNTS
- Published: 2026-09-17T11:00:24+00:00
- Read time: 1 min
- Topics: affirm, buy now pay later, underwriting, machine learning, credit scoring, transformer model, lending, fintech, checkout, credit history, business, finance
## Affirm launches transformer underwriting model
Affirm's new real-time underwriting model is now live at U.S. checkouts. It learns from patterns in credit history rather than relying on a single credit score.
## Built on 14 years of transaction data
The model was trained on 14 years of Affirm's transaction-level lending data. It identifies patterns within and across credit accounts, including how they change over time.
## Approves borrowers previous system declined
The model approved eligible applicants Affirm's existing system would have rejected, including consumers with thin credit files and no FICO scores at all.

> Seeing a credit history more clearly means we can responsibly say yes to more people.
## Results beat previous expansion efforts
The new approvals generated 3.4% more completed purchases than a control group, at comparable risk levels. Loans performed better than similar expansions under Affirm's previous models.
## Saying yes requires getting better at saying no
Affirm built a proprietary algorithm that explains the model's decisions while keeping it fast enough for checkout. The goal is right decisions per transaction, not approving everyone.

> The goal isn't to approve every transaction, it's to make the right decision for each one.
## Customers using Affirm for everyday purchases
Affirm processed 53 million transactions in fiscal Q4, up 41% year over year, while average order value fell 4%. Transactions per active consumer rose 20% to 7.0 over 12 months.
## Model widens access to everyday financing
More frequent, smaller transactions show consumers using Affirm as a regular payment method. A model that approves more eligible borrowers expands who can access this option.
## Key takeaway

By analyzing patterns in credit data over time rather than compressing history into a single score, Affirm approves more eligible borrowers while maintaining risk discipline.