Create a single scorecard for each consumer ranking those most likely to pay higher
The Business Challenge
For this large credit bureau, Fractal considered two key factors to help its customers:
- The client wanted to build a tool that helped its customers effectively allocate resources to recover delinquent loans.
- The client recognized the need for a robust solution that could work over a broad range of industries and products.
Fractal decided to deploy a two-tiered, multi-segmented approach for model development. This approach:
- Included a large list of defaulters from a number of collection agencies and debt buyers.
- Used a CHAID (Chi-Squared Automatic Interaction Detector) decision tree and logistical regression techniques to segment the defaulters via their propensity to pay and separately, by the likely amount recovered, or yield, for each customer.
- Developed segment-specific models that were then combined to create a single score for each consumer, providing higher rankings for those most likely and able to pay more than those less likely to pay.
Fractal Analytics utilized a CHAID decision tree and logistical regression techniques to demonstrate that up to an estimated additional 6% in delinquent balances could be recovered with the same resources.
- This decision tree identified five segments for colleciton models and four more for yield models.
- The high segment yield accounted for 80% of payers in the top deciles, having a total recovery amount of 83%.
Fractal’s scorecard helped its client achieve a 6% higher volume of recovery balances with a similar quantity of payers.
Fractal’s high yield segment demonstrated that up to an estimated additional 6% in delinquent balances could be recovered with the same resources
A robust single scorecard was deployed for this cross-industry application.
Varied techniques explored to suit the dynamics of the portfolio yielded a better fit and greater stability, including Tweedie,a generalized linear model, Tobit, Linear and Logistic