Finli: 40% more approvals, zero increase in defaults

The problem

Sixty per cent of applicants declined at automated decisioning. Not because they were poor credit risks. Because the bureau score could not assess them. A self-employed plumber with steady invoices and a healthy current account, declined for a thin credit file.

The question was not whether to approve more loans. Any lender can do that by loosening criteria. The question was whether alternative data could identify genuinely creditworthy borrowers that bureau scores were missing, without increasing defaults.

The solution

We built a scoring model using open banking data. Applicants who fall below the bureau threshold get a second assessment against their actual transaction history: income stability, spending patterns, savings behaviour, cash-flow resilience. The second-look pathway is invisible to applicants and processes in under ten seconds.

Every decision is fully explainable. The FCA's expectations around algorithmic lending transparency are met by design. No protected characteristics, no postcode proxies, no features that introduce demographic bias.

The result

Loan approvals up forty per cent with no increase in the default rate. The borrowers were creditworthy all along. They just lacked the conventional paper trail to prove it.

Self-employed workers, recent arrivals, people rebuilding after financial difficulty. Now served on fair terms. Many subsequently build enough credit footprint to qualify through conventional channels for their next loan.