Axtria Insights

Axtria Insights

CASE STUDY
Axtria Optimized Rules Engine to Improve Fraud Detection Capability for a Leading Us Credit Card Issuer

Situation

A large US credit card issuer was facing high fraud losses due to sub-optimal rule based triggering and inability to capture substantial fraud volume.

On the segments where the cases were being triggered and worked upon the “payment hold” was impacting significant number of “good” customers

Approach

We created a decision tree to improve fraud rates for existing triggers and added new set of rules to improve capture rate.

Data Preparation
  • Considered merchants with less than 6 months tenure
  • Created list of customers with minimum 1 transaction in the last 18 months on a  defaulted merchant
    • Created transactions  sample on merchants with less than 6 months tenure as of default month
Dependent Variable Creation
  • Merged the 2nd dataset with 1st, summarized to submission level & created the following metrics
    • #CMs in submission which were present in 2nd dataset
    • $ amount in transactions of these CMs in the 2nd dataset
  • Identified submissions which met a $value criteria – fraud indicator
  • Used this indicator as a dependent variable
Independent Variables – Bivariate Profiling
  • Carving out ranges in variable to localize high fraud rates
  • Typical Variables considered
    • Historical Bad Debit Amount
    • Current Bad Charged Amount
    • Current Submitted Debit Amount
    • Bad Customer Submission on # defaulted SE
Segmentation
  • Created high-medium-low fraud rate segments using decision tree and variables identified using bivariate profiling
  • Similar analysis done with existing rules to improve detection rates

Result

  • Fraud capture rate improved from 5% of possible fraud detects to 11%
  • “Good” held rate reduced by 43%

AXTRIA OPTIMIZED RULES ENGINE TO IMPROVE FRAUD DETECTION CAPABILITY FOR A LEADING US CREDIT CARD ISSUER