Blue Motors Group cut repossessions from 3.2% to 2.3% and decision time from days to milliseconds
Blue Motors Group Ltd already had a low loss rate, but every application passed through two to three days of manual underwriting. Underwrite.ai reverse-engineered that human decision process into a custom nonlinear model that agrees with the underwriters 98% of the time. In production the model decisions in 2-3 milliseconds and the repossession rate fell from 3.2% to 2.3% — a 28% relative reduction — while the lender became the fastest-growing lender in Europe.
Key results
3.2% → 2.3%
2-3 days → 2-3 ms
98%
The challenge
Blue Motors Group's underwriting was working on the dimension that mattered most: losses were low, and that was the product of experienced underwriters reading each file carefully. The cost was throughput. Every application took two to three days of manual review, and in auto finance that latency is lost business — an applicant waiting three days for a decision has usually financed the vehicle somewhere else. Growing the book meant either hiring underwriters linearly with volume or relaxing standards, and relaxing standards would have given away the low loss rate that made the business work.
The approach
The goal was not to replace the underwriters' judgment with a generic bureau score but to reproduce it at machine speed. Working from Blue Motors Group's own loan tape and application tape, we built a custom nonlinear model that captured how their underwriters actually weighed applicant and collateral factors together — including the interactions a linear scorecard cannot represent. The model was validated against how those underwriters had decided historical applications and agreed with them 98% of the time, which is what gave the credit team confidence to let it decision live traffic. It was delivered behind an API that dropped into the existing origination flow.
The results
In production the model returns a decision in 2-3 milliseconds, against the two to three days the manual process required. More importantly, credit quality improved rather than held: the repossession rate fell from 3.2% to 2.3%, a 0.9 percentage point drop and a 28% relative reduction. The model captured risk signal the manual process was missing, so the speed gain did not come at the cost of losses. Blue Motors Group went on to become the fastest-growing lender in Europe.
Validation & controls
- 98% agreement with human underwriter decisions on validation against historical applications
- Decision latency reduced from 2-3 days of manual review to 2-3 milliseconds via API
- Repossession rate reduced from 3.2% to 2.3% (28% relative reduction)
- Adverse-action reason codes generated for every declined application
- Disparate-impact analysis available for fair-lending review
Want results like these on your portfolio?
Models are built from your own anonymized loan and application history. No setup fees.
Start for Free