When it comes to fraud, the insurance industry has traditionally focused most of its attention on claims, using analytics, predictive modeling and other tools. But with advanced analytics and big data tools now at their disposal, an increasing number of insurers are beginning to tackle underwriting and application fraud as well.
That’s true at CNA, the eighth largest commercial property/casualty insurance company in the U.S. Since 2011, CNA has used analytics and predictive modeling to identify claims fraud and estimates it has generated more than $8.4 million in savings. But next year, the insurer plans to pull policy application data into its predictive model constructs. “Lots of carriers are starting to talk about what the industry can do about underwriting fraud,” says Tim Wolfe, director of CNA’s Special Investigations Unit (SIU).
Underwriting and application fraud, also known as premium fraud, occurs when someone purposely conceals or misrepresents information when obtaining insurance coverage, typically to get a lower premium. One of the most common forms is data misrepresentation; an auto insurance applicant may not list all of the drivers of a car, or a business may underreport its payroll for workers’ compensation insurance.
But over the past few years, better access to more data and background information is helping to enlighten insurers at the time applications are presented, according to Chad Huneycutt, owner and president of The Huneycutt Group, which operates several insurance agencies on the Southeast Coast.
“When a customer submits an application for auto insurance, for example, the underwriter often knows how many people live at the applicant’s address, how old they are and for how long they’ve lived there. Automatic access to this kind of information is helping to prevent fraud,” says Huneycutt, who expects insurers to continue to become more sophisticated in their use of technology such as analytics. “Catching fraudulent applications is good for the industry overall,” he observes. “It saves insurers money, but it also saves consumers money, because we all pay for insurance fraud with higher premiums.”
A Multibillion-dollar Problem
In fact, insurance fraud costs billions of dollars each year, and those costs are passed on to policyholders. The Coalition Against Insurance Fraud estimates that insurance fraud costs Americans at least $80 billion a year, or nearly $950 per family.
Using technology to catch insurance fraud has already gained a great deal of traction, but more so for claims. According to a study released in mid-September by the coalition and sponsored by SAS, 95 percent of respondents said they use anti-fraud technology.
However, only one-third of insurers are using the technology to combat underwriting fraud and rate evasion. The study, “The State of Insurance Fraud Technology,” consisted of an online survey of 42 insurers, representing a significant share of the property/casualty market.
It’s fair to say that the reason the industry has dragged its feet on aggressively tackling underwriting fraud is because underwriting is directly tied to an insurer’s top line. “Underwriting drives new business,” says James Ruotolo, principal for insurance fraud solutions at SAS. “But what we really should be looking at is not whether sales and underwriters are writing new business, but whether they are writing profitable business.”
Hot Potato’ Issue
CNA’s Wolfe agrees. “Candidly, underwriting fraud is a bit more of a political issue. Companies set ambitious sales goals, and want to keep sales agents happy and keep customers happy. So it is a bit of a hot potato,” he says. “For a long time, there was no appetite for collaboration, but that is changing. No one is necessarily talking about sharing proprietary information, but they are looking at where the gaps are and starting to talk about how to fill those gaps and about best practices.”
Another challenge, Ruotolo notes, comes with big data, which is both a blessing and a curse. “If you’ve got volume, variety and velocity, then you’ve got a big data problem,” he says. Unlike claims processing, which takes days or even weeks, underwriting happens much more quickly, with insurers writing many hundreds of policies per day. The advent of online applications has spurred even greater data velocity, and with increased velocity comes an increased volume of data. Meanwhile, the diversity of data sources also is growing, especially unstructured data sources.
According to the Coalition Against Insurance Fraud study, insurers still use internal data most frequently, but other sources include industry fraud watch lists (67 percent of survey respondents), public records (45 percent), unstructured data (38 percent), third-party data aggregators (29 percent), social-media data (14 percent) and data from connected devices such as dongles (5 percent). CNA also is making use of data from the National Insurance Crime Bureau, which issues alerts, as well as data from the ISO ClaimSearch database. “All of this makes our whole program more robust,” Wolfe says.
Until recently, analyzing a vast amount of different data types hasn’t been possible, says Bob Cummings, head of SAP’s Insurance Business Unit, and insurers have only been able to rely on a small number of data sources. But looking only at claims data to identify claims fraud, for example, has made it challenging to find patterns that might uncover fraud in other aspects of the business, such as in underwriting.
The good news is that technology is catching up. With massive parallel processing technologies, “We can take billions of rows of data, load it into the systems and run analytics in memory. Because you aren’t writing anything to a disk, the speed is much faster,” Ruotolo says. “What used to take 180 hours can now run in 12 minutes.” Faster analytics mean insurers get better information more quickly, and processing that used to occur once or twice a month can now happen daily.
In its study, the Coalition Against Insurance Fraud points out that the ability to use data sources that were previously ignored, either because they were too large or changed too often for more-traditional fraud systems to track, is very promising, and high-performance analytics is driving new innovation in fraud detection. According to the survey, the primary anti-fraud technologies in which insurers plan to invest over the next 12 to 24 months include link analysis (45 percent, up from 19 percent) and predictive modeling (38 percent, up from 33 percent). Nearly one-quarter plan to invest in data visualization, which enables insurers to quickly recognize changes in fraud patterns (data visualization was not included in the previous survey).
CNA started investing in fraud detection technologies as far back as 2008. “It has been quite a long journey for us,” Wolfe says. “It certainly didn’t happen overnight.” The company first built its own predictive modeling system to identify suspicious workers’ compensation claims that weren’t being picked up by its adjusters. “We saw a very minimal lift from that,” he acknowledges.
CNA then began a painstaking effort to source full-fledged predictive modeling programs and undertook three separate proof of concepts (POCs) with different vendors. According to Wolfe, such due diligence is necessary because the software is complex and expensive. As a result of these trials, and with the support of upper management as well as people from multiple lines of business and divisions, many of whom had worked directly on the POCs and provided input, CNA selected a SAS solution. By January 2012, the new system was fully deployed.
Fraud Detection: An Ongoing Process
But that wasn’t the end of that. As Wolfe points out, the use of fraud detection technology is an ongoing process. “You have to keep refining the models,” he notes. “We are now going through a model refresh that we hope to have finished by the end of the year.”
“Insurers are realizing how technology lets them analyze large sets of data and use predictive modeling,” Ruotolo says. SAS recommends a hybrid approach that combines multiple technologies including business rules, anomaly detection, supervised predictive modeling and social-network analysis, which involves analyzing networks within an insurer’s own data sets. Using the SAS Fraud Framework for Insurance, “You incorporate all those elements into the fraud score,” he says. The SAS Framework is designed to help insurers detect, prevent and manage fraud across all their lines of business.
SAP, the other major player in this market, offers the SAP HANA platform, a portfolio of technologies designed to help companies analyze large volumes of internal data across their core insurance applications, along with data derived from external data sources. These may include telematics and weather reports, Cummings says, as well as Twitter, Facebook and other social networks for sentiment analysis. The vendor’s SAP Fraud Management for Insurance leverages HANA and combines business rules, predictive modeling methods and net- work analysis with calibration and simulation technologies. This enables an insurer to simulate the number of alerts a particular detection strategy would generate, and then choose to execute the most productive strategy, given the number and type of fraud investigation resources the firm can throw at the problem.
With these and other software tools, insurers now have an array of sophisticated anti-fraud tools at their disposal. By applying them to premium fraud at the point of sale and during the application and underwriting process, insurers will be able to cut their losses and write policies that contribute to their bottom line.
