LUMINNOVATE

Insights

Which existing customers will want credit next year?

0.92Real bank data from an open research dataset

Offers before they ask

AUC for next year's first time borrowers, ranked out of time; account age alone gave 0.85

What was observed
3,527 customers' account behaviour in one year; 107 of them (3.0%) took their first loan the next. No offers were ever made.
What a generic approach says
Rank customers by how long they have banked with you: AUC 0.85.
What the engine read
Ranking each customer from their own account behaviour, trained on one year and tested on the next, reached 0.92. A separate risk model ranked who could repay at 0.80, tested on 117 loans with only 16 defaults.
What it meant for the decision
Offers can go first to the customers most likely to want credit and able to repay, with eligibility and risk checked first. The gain over account age is real but modest, and the risk test is small.

AUC, ranked out of time

0.80.91From account behaviour0.92Account age alone0.85
Real bank data from an open research dataset

Figures are reproduced from Luminnovate's research records. Data sources are available on request; methods are proprietary.

Solution
Pricing and offers
Industries
Banks and lenders

Every decision that moves money should be able to show its working.

Book a conversation