Analytics vs. fraud

More and more, I'm hearing that a key application of analytics is fraud detection.  Visa's new anti-fraud application is just one example; other examples may be found among cellular service providers, mortgage lenders, and of course insurers.    And basic BI has been essential to auditing for over 30 years, all the way back to when John Cullinane saved his company by rebranding the hard-to-sell report writer Culprit as "EDP Auditor" (this was before IDBMS); that bright idea even won him a big award from some auditor's association.   This idea is even more important  in today's compliance-heavy environment, as I argued on page 11 of my white paper on Third-Generation Analytic Business Processes.

I was reminded of all this at the recent Text Mining Summit , where some of these applications were discussed openly, and even more came up in private hallway discussions.  I did at one point get puzzled about the mortgage lending application, and ask how much good it did to get early warning of whether a mortgage will go bad.  The answer was quite interesting: 

It isn't so much a question of early warning about the need for aggressive recovery of the money.   Rather, for reputational reasons lenders (who, after all, resell their loans) need to discover and admit which loans are bad as soon as possible.  Ideally, of course, they'll do this before they ever resell the loans, thus maintaining high quality in the Collateralized Mortgage Obligation securities they issue, the sale of which is completely necessary for them to even stay in business.