click here for the Warranty Week home page
February 25, 2010

Warranty Fraud Detection:

SAS says send us your claims data and we'll show you the fraud. If upwards of 10% to 15% of claims are in some way suspect, such a system could pay for itself in just a few months. Does it sound too good to be true? GE Appliances was the first customer.

Warranty fraud reduction has always been a hot topic at the Warranty Chain Management Conference. At the very first WCM gathering five years ago in San Francisco, they ran out of chairs and then they ran out of standing room in the subdivision of the ballroom given over to the topic.

This year, the warranty fraud panel on Wednesday afternoon faces some tough competition from two other tracks. One of those features three attorneys taking a deep dive into warranty law. The other features a multi-industry panel of speakers who will elaborate on the benefits of designing products for customer self-repair. So hopefully there will be enough seats for all.

As was the case last year, the warranty fraud panel in Los Angeles features a software vendor (SAS Institute, a sponsor of this newsletter) and an end user (General Electric Co.) talking about how they reduced the number of suspicious claims. But while last year the system was part of a pilot program whose success had pleasantly surprised the participants, this year it forms the core of a commercial product launch.

SAS has extensive experience helping its customers in the banking, credit card and health care industries to use analytics to detect and reduce fraud. And of course it has a wealth of experience helping manufacturers use analytics to detect defects in their product lines. Now it's taking the best of both and launching the new Suspect Claims Detection solution to help manufacturers detect fraud in their warranty claims data.

Software as a Service

However, rather than selling the solution as a software module the customer installs on its own computer systems, SAS plans to sell Suspect Claims Detection as a service. Customers will send their claims data to SAS. Then SAS will perform the analysis, and will provide its customers with a dashboard interface where they can view the results.

David Froning

David Froning, manager for warranty solutions at SAS, said the company is in the midst of expanding its "software as a service" initiative right now. The price in this case will be based on the total cost of the claims being run through the system, he said. "So if somebody wants to run a million dollars worth of claims through the system, there's one price. If they want to run a hundred million dollars through the system, there's a different price," he said.

The assumption, Froning said, is that the cost and the savings will be highly correlated. But rather than charging by the claim, SAS decided to charge by the total dollar volume. That is expected to broaden its appeal to both personal electronics and appliance makers that have lots of small-dollar claims as well as to large construction equipment manufacturers that have relatively few large-dollar claims.

SAS calls its software-as-a-service effort Solutions OnDemand. (It couldn't very well use the acronym SaaS, or worse, SAS SaaS.) Modules have already been announced for marketing campaign management, K-12 school district administration, and drug development, among others.

Mike Newkirk, the industry product marketing manager at SAS, noted that construction is already well under way on a 38,000-square-foot cloud computing facility within the company's headquarter campus in Cary NC. "It will include two 10,000-square-foot server farms," he said, part of which will be used to support the company's Solutions OnDemand product family, including this claims fraud detection module.

Reducing Warranty Cost

"Obviously, suspect claims is an expensive business problem," Froning said, suggesting that as many as 10% to 15% of warranty claims are fraudulent. "Those are the direct costs. Then there are secondary costs such as stolen parts being resold, and additional security and insurance."

"Recently, with the economy going through its changes, we've seen that there is much more of a focus on fraud," Froning added. Manufacturers are obsessed with reducing costs, and the promise of reducing claims costs by 10% to 15% without denying any legitimate claims is irresistible.

At the same time, repair service providers are looking for ways to increase their revenue, or more precisely, to replace some of their reduced revenue with spurious claims. "Manufacturers are telling us that service providers are much more likely to try to get things through the system right now -- to get cash any way they can," he said.

"We're also seeing the fraudsters are getting more sophisticated," Froning added. Some of their activities now verge on organized crime, with a handful of culprits seemingly setting up shop for no other purpose than to manufacture bogus claims. "It's definitely evolving, and people are getting better at working around systems that have historically been used to find this kind of fraud," he said. "And then to make it all worse, with downsizing and cost-cutting, manufacturers are reducing the size of their audit staffs. So people are forced to try to do more with less."

Less Is More

SAS has now packaged its analytics in a way that will hopefully allow manufacturers to spot more warranty fraud using less people than before. Automation of the fraud detection process allows companies to do more with less -- precisely in keeping with the theme of this era. Fewer auditors, with the help of analytics, can cover more ground than was possible before.

"The traditional approach to finding claim fraud is manual adjudication, where you're reading the claims and trying to see if they're good or bad," Froning said. "And although that's very thorough, it's very manually intensive. Often, what we find is that individual auditors don't see the bigger patterns, because this person is looking at a hundred claims, and somebody else is looking at a different hundred claims, and they're not sharing that knowledge across the broader set of claims. So they can often miss things."

The solution is to automate the process so that a single system is "reading" all the claims, setting rules for items such as what parts go with which labor codes and looking for unusual patterns in the data. The problem, Froning said, is that the perpetrators can quickly figure out what the fraud detection system is looking for, and they can modify their tactics accordingly.

"As I talk to some of our customers, they say one of the most common questions they get is 'Why did this claim get rejected?'" he said. "And they know that if they explicitly tell them exactly why a claim got rejected, the answer is going to be written down in a notebook, and they'll make sure that doesn't happen again. They'll figure out another way to work around that business rule."

Second Line of Defense

Most of the warranty claims processing systems already have a rules-based engine built into them. And that's just fine with SAS, Froning said. For instance, General Electric is first running its claims through another company's warranty claims processing system, which applies a set of rules against the claims, looking for anomalies. And then SAS runs those claims through its Suspect Claims Detection module, which looks for less obvious patterns in the data (as well as in the text).

Froning said that asking service providers to return replaced parts, or to send inspectors out into the field to look at repairs while they're in process are additional fraud prevention tools. The problem is those methods are very expensive and time consuming. The same goes for field audits.

Analytics, he suggests, can help a manufacturer to create a priority list of which repair facilities deserve the most immediate attention. Where is there a high probability of fraud? Or, if one prefers to give repair shops the benefit of a doubt, where are there "training issues" that are resulting in inadvertent mistakes and unnecessary charges? In other words, if an auditor can perform only a hundred field audits per year, the analytics can help them pick the hundred best prospects.

Froning also suggests that manufacturers can use a kind of scoring system that learns from past claims and audits to find similar patterns in new claims. If it was fraud before, chances are good that it's also fraud this time, even if it's not the same company or people involved. But sometimes it is the same people, moving from one company to another, or changing company names. That requires some degree of "social network analysis," which looks deeper into factors such as company ownership and employment details.

Suggested Best Practices

What seems to work best is a system where multiple sets of rules and analytics are looking for unusual patterns and suspicious links in the claims data. The claims processing system should take a first pass at the data, and then the analytics system should take another look. In any event, all this analyzing and scoring and investigating should be done before the suspect claims are paid, because it's easier than trying to get the money back.

Froning suggested a list of seven best practices that manufacturers should follow to reduce warranty fraud:

  • Utilize automated, rules-based adjudication,
  • Use multiple text and data based analytic models,
  • Look for anomalies across claims, service providers, and networks of service providers,
  • Score claims before they are paid,
  • Score claims in real time,
  • Rank claims and service providers based on likelihood of fraud and indicate type of fraud suspected, and
  • Focus field auditors on the right claims and service providers.

"It�s really about having a comprehensive fraud detection process and a solution that supports it," he said. And at the WCM Conference on Wednesday afternoon (3:45-5:30pm), Froning will team up with Jeff Moore from General Electric to jointly detail how they implemented such a system for GE Home & Business Solutions.

Swift Payback

The payback was both large and swift. A year ago, at the WCM Conference in Orlando, Moore said the system had already paid for itself four times over. In its first year of operation, the SAS Suspect Claims Detection saved GE over $5 million, GE estimated.

That came as something of a pleasant surprise. "We knew there would be quite a big number, but we were completely shocked about some of the stuff we found," he told Warranty Week at the time.

When the system first went in, GE had a little test for SAS. GE had recently been tipped off about three service providers committing warranty fraud. But it didn't tell SAS which ones. Instead, GE handed over its warranty data to SAS, and asked SAS to find them. SAS found all three.

Moore and Froning are expected to provide an update on the second and the beginning of the third year of the pilot project next week, including the estimated savings it has produced for GE. Froning declined to pre-announce the latest figures, but said that they have continued.

In fact, Froning said GE's auditors have now noticed that the number of suspect claims has begun to decrease even without their intervention. It's as if the mere existence of the fraud detection system was deterring the service providers from trying to push the envelope.

"GE saw this," Froning said. "They saw some of their service providers were now taking an extra step to validate and verify claims before they were submitted. We were talking with one of the major service providers that does work for GE, and they are interested in purchasing this solution, to run their claims through it before they submit them. And if there are any questionable ones, they'll validate them before they get sent in."

Product Demonstrations

SAS plans to publicly launch the Suspect Claims Detection module next week at WCM, and to demonstrate it at its booth alongside the SAS Warranty Analysis module. Froning noted that while the two products are sold separately, they work quite well together. Both aim to lower warranty costs -- one by reducing defects and the other by reducing fraud.

"It's really a complementary product," Froning said. "But it uses a lot of the same data. So if a customer has one, it makes the implementation much easier for the other one." In other words, if the claims data has been cleaned up and formatted for early warning analysis, it's already in good shape for fraud detection.

There's also another benefit of using both together, he added. The fraud detection system, by eliminating the invalid claims, is actually making the early warning analysis more accurate.

"By getting those bad claims out of the system, you're really cleaning up some of the noise," Froning said. "So it helps you focus more on true product problems in warranty analysis, instead of having a lot of noise from suspicious activity."

SAS Warranty Analysis Enhancements

SAS is also planning to exhibit the latest enhancements it has made to SAS Warranty Analysis at WCM. In December, SAS announced Release 4.2 of the product, the seventh major update it has issued since 2002.

"We've tried to make the early warning more useable and more valuable to our customers," Froning said, by allowing users to specify exactly what they need to monitor. One of the new features, called My Emerging Issues, is a personalized list that an engineer can use, inputting specific part numbers and labor operations of interest to them.

"Now, instead of looking through the whole list of emerging issues that have come up, it's going to surface just those that are relevant to me," he said.

Froning said the email alerts have also now become more specific. Instead of merely alerting the recipient that there are items that require their attention, the alerts now detail exactly what those items are.

SAS has also simplified the "drill down" procedures, eliminating the need to retype the metrics into the system when moving from the alert to the problem definition area. Now, the user can simply click on the alert and begin to probe the issue. Users can also drill down on groups of issues and analyze them together or separately.

Teradata Integration

"What we've done for this release, based on our broader partnership with Teradata, is to allow our warranty analysis product to run directly on top of Teradata," Froning added. "And we've integrated their early warning system so you can view the results from within our application."

Kamau Njenga, Teradata

Kamau Njenga, program director at Teradata, said that partnering with SAS allows the companies to combine the best of both their worlds. SAS has the analytics horsepower and the warranty-specific experience, and Teradata is able to scale that functionality to huge data warehouses.

The integrated module, which Teradata has made part of its the Warranty Analysis Advantage Program, will be on view next week at Teradata's exhibit booth at the WCM Conference, Njenga said. It is the latest in a series of joint announcements the companies are calling their Advantage Program, which was first announced in October 2007 and was expanded last October.

Another module in the Advantage Program, called the SAS Credit Risk Advantage for Teradata, is aimed at the financial services industry. It places SAS 9.2 analytics atop a Teradata 13 database, which has a theoretical capacity of 50 petabytes (50 million gigabytes). But it also can be scaled down to a smaller size -- what Teradata calls its Advantage Express configuration -- aimed at mid-sized companies or single departments within larger companies.

Njenga said select customers have already been approached about implementing the joint solution for warranty. He said it should appeal to manufacturers in numerous industries, including automotive, industrial machinery, consumer electronics, and appliances.

Customer Benefits

So far, SAS Warranty Analysis has attracted 23 customers around the world, including Briggs & Stratton, Hewlett-Packard, W.L. Gore & Associates, American Honda Motor Co., Shanghai General Motors, Sub-Zero and Wolf Appliance, and Maytag and Whirlpool. Several have agreed to share their success stories, allowing SAS to publish cost saving statistics that have become more precise over time.

"The value that our customers generally see is on average a four month improvement in detection of new issues," Froning said. "Now they can perform that problem-solving process four months earlier than they could before. And that generally translates into an 18% reduction in warranty cost. So by knowing about it sooner, you can get the problem fixed sooner, and you've produced fewer defective products, which means fewer claims and lower costs."

Froning said his main role is to talk with SWA customers about their experiences with the product, and then to channel their feedback to the R&D and programming folks at SAS, so they can work it into new releases of the product. And that exposure has helped him to sharpen the cost-benefit data he gets from customers. "In the past we've stated ranges," Froning said, "15% to 20%, from 10 to 20%, things like that."

Shanghai General Motors, a joint venture founded in 1997, has stated publicly that it saw a 34% reduction in warranty cost per vehicle. Other customers that have been in business longer are seeing 10% to 12% improvements. "But the average turns out to be 18%, so I've decided to go with that more solid number."

Froning notes that there is also a secondary effect of fewer defects that can be seen in higher customer satisfaction levels and more customer loyalty. "And when a recall is necessary, we're providing the tools so you know exactly what combination of factors is driving the failure rates. So you can recall maybe 10,000 units instead of 500,000 units."





AMT Warranty Corp.
Fulcrum Analytics
Warranty Chain Management Conference
Sign up for a free subscription to Warranty Week:
     subscribe     change of address     unsubscribe

Related Articles From Warranty Week:





fraud detection sas claims data suspect system ge customer year product customers analytics manufacturers service module analysis froning providers teradata