Case Studies

We led loss mitigation initiatives at a major global credit card issuer by developing a dynamic, transaction-based scoring algorithm. The goal was to reduce credit losses incurred by new card customers. The approach  complemented the traditional credit bureau based approach and took into account unstructured data such as transaction-level information, merchant categorization, velocity of transactions, etc. The solution was able to identify segments with default propensity that was 40x that of the average.

Approach

• Client was a major global bank wanting to reduce charge-offs among new customers.
• Sized the impact of different kinds of losses; Selected low-volume but high-impact defaults for modeling purposes.
• Created exhaustive list of available and derived attributes that would serve as predictors of early churn.
• Used decision tree modeling to develop predictions; Horse-raced models that could be run at different points in customer life-cycle.

Analysis

Results

• Identified predictors of early default, using transaction level data.
• Highest risk segment was 40 times as likely to charge-off as the average