A tier-one bank was concerned with fraud levels in partner channels, particularly around synthetic ID fraud. Fraud was only being noticed after the fact through reviews of driver’s license numbers and comparing multiple matches within internal database. The bank was interested in discovering what data could be used automatically at the time of application to flag a portion of this potential fraud.
The client identified 18% of shared customer records as fraudulent by: 1) using name, address, and phone number as keys to append additional data; 2) analyzing and identifying 15 top factors predicting fraud likelihood; and 3) building a model segmenting customers by likelihood of fraud, with 17x differentiation between best and worst customers.
- Segmentation was used to flag additional reviews, saving $475,000 per month at a x10 ROI