US consumer lender · 6-week proof of concept · 394 applications · client-reported
Direct fraud losses plus investigation, recovery and compliance cost on every case.
Nearly half of online applications in the sample were fraudulent.
Fraud was found after approval or funding, once money had moved.
Stopping it had to happen without blocking genuine customers.
A 0 to 1 risk score with reasons, from 650+ behavioural signals.
Low risk fast-tracked with sampled audit; medium to a reviewer; high to compliance.
Nobody is auto-declined. Every decision is logged.
Drift is monitored, and a retrained model is promoted only if it beats the last one.
Loan application system
Case management for reviewers
Fraud rate on online applications with AI scoring.
Human review closed the remaining gap.
No one is declined by the model alone.
Effort goes to the applications that need it.
Industry research (LexisNexis True Cost of Fraud, 2025) puts the total cost to US lenders at $5.38 for every $1 of fraud lost. Per $1M of direct fraud loss at the 48% baseline: Illustrative. The multiplier is a US figure; we validate it for the UK market.
Across 394 online applications in a 6-week proof of concept, AI scoring brought fraud from 48% to 3%, and human review brought it to 0%.
Client-reported results from a 6-week proof of concept on 394 applications.
How we label resultsTell us about it. The first discovery call is free.