LoanPerformance, which sells retail credit analytics to mortgage lenders, says text mining can boost the accuracy of default models. If a customer uses any of these words when talking to a call center rep, the software flags the customer as more likely to default:
Rental
Renter
Rent
Roommate
Hazard
Debris
Fraud
Death
Marital Problem, Divorce
Mold
Sad
Jail
Motorcycle
If I were a sad divorced guy riding a motorcycle to a moldy jail, I’d probably default too. But I sure wouldn’t talk to a customer rep about it.
Comments on this entry are closed.