The challenge
About one applicant in four came back with no legacy credit score at all. For the others, the score left little room to tell people apart: most of the applicants it scored sat in a narrow band. The bank asked one question: can Monnai measure credit risk where the legacy score sees little?
What Monnai did
- Five fields per applicant went in: name, phone, email, country and IP. About 350 data points came back, from phone, email, digital footprint, network, identity and IP checks.
- A risk score trained on repayment outcomes in Mexico, not a global model applied as is, with the reasons behind every score.
- Run alongside the legacy credit score, not instead of it. Every data point is a fact the bank can check, and write policy on.
What this means for credit and risk teams
- A score where there was none: 98% of the applicants with no legacy credit score got a Monnai risk score, with its reasons.
- More room to tell people apart: Monnai spread the applicants the legacy score bunched together about four times wider.
- Signals the bank can verify: nearly a third of Monnai’s riskiest tenth could not be reached on messaging apps, against almost none of the safest, and their phones and emails were years newer.
- Good applicants the legacy score can’t see: about 15% of the applicants with no legacy credit score were reachable, had a real history and sat in Monnai’s safer half.
- Next: 90 days of the bank’s repayment outcomes, to calibrate the score to its own book and measure default rates.
