Taking the question out of questionable claims

Copyright © 2009 James Taylor. Visit the original article at Taking the question out of questionable claims.Jeff Moore from General Electric and Greg Spraker from SAS (see my review of the SAS Warranty product here) spoke on using analytics to find and eliminate fraud in claims. GE’s appliance division dealt with paper claims prior to […]

Copyright © 2009 James Taylor. Visit the original article at Taking the question out of questionable claims.

Jeff Moore from General Electric and Greg Spraker from SAS (see my review of the SAS Warranty product here) spoke on using analytics to find and eliminate fraud in claims. GE’s appliance division dealt with paper claims prior to 2003 and randomly selecting claims for audit. Between 2003 and 2005 they increased the number of auditors but hard to find the patterns across auditors and, although they had rules-based flags, they had no way to improve the rules systematically. Since 2006 have been applying increasing amounts of predictive analytic technology.

Once they had implemented a claims system (they get 1M or so claims a year) they still found that tended to focus on single claims and so missed the trends and organized fraud. They identified some key areas like:

• Was a repair was actually performed and what repair was actually done?
• What parts were used?
• 6,000 servicers with multiple employees – who is not trustworthy?
• Is the quality data that comes back accurate so that it can be used for quality improvement?
• Parts costs are escalating fast and inflation of prices is an issue
• Products change rapidly and repair types therefore should change too
• Fuel surcharges or extra mileage to cover fuel costs

GE works with SAS using a SaaS model. Claims are sent to SAS before payment and 26 claim-level analytics are calculated for each claim. Claims are flagged for audit with multiple elements compared to averages. They also do 10 servicer-level analytics based on a comparison to historical data previously sent to SAS.  This has saved GE a huge sum – more than $6.5M in a 10 month period. About $5M of this was in rejected claims – much of this from bad claims rather than deliberate or systematic fraud. They also had some servicers where they could not validate a single claim and another $1.5M savings is estimated from claims that these servicers would have been paid had they not been suspended.

Some examples included:

• Finding a servicer who was using new serial numbers on old products and so claiming for products that had not been made in years.
• Another servicer was filing apparently good claims against imaginary consumers. They got fired but then filed a whole bunch of claims that pre-dated the date they were fired. The system detected this spike too.
• Another used to go back and forth between job codes to boost amount they were paid and this had slipped by when only individual claims were investigated.
• The system finds servicers who have a bad employee or small group of employees.
• Excess mileage detected in another case

GE learned that analytics let’s them work smarter not harder – inspect fewer claims but catch more fraud. In one group for instance they went from looking at 75% of claims to 30% while catching more problems. They also learned that patterns are more important than single claims. The analytics allowed them to find holes in the rules they were using to validate claims and, as usual, 80% of losses are driven by 20% of servicers. Overall they got reduced cost, increased savings and speedier turnaround.

Greg wrapped up by giving some examples of continuous improvement – street address validation, house to house distance calculations, feedback false positives, text mining for repeat calls and so on. He also pointed out how important this is to the customer experience. Customers with a bad warranty history are very unhappy but those who have had a single positive warranty experience actually trust companies more. GE is beginning to integrate these warranty analytics with spare parts optimization (treating each truck as a parts storage center), call center forecasting, reserve forecasting and more.

This is a great example not only of using analytics to find patterns of fraud but of feeding that insight back into operational decisions for real decision management.

Link to original post