Industry estimates put personal auto insurance fraud at about 10 percent of paid losses and loss adjustment expenses a year. That equates to roughly $13.3 billion in 2011 alone. And the overall trends don’t appear good. According to the National Insurance Crime Bureau (NICB), questionable claims continue to rise each year, with a 56 percent increase between 2008 and 2012.
Fraud can take the form of fabricated claims or staged accidents (“hard” fraud), or it can be opportunistic, as when claimants pad their losses or falsify information to obtain lower premiums (“soft” fraud). Unfortunately, most carriers tend to find fraud when it becomes most obvious, at the time of claim. But what if you could identify potential fraud even before you write the policy – at the time of quote?
Detecting Potential Fraud?
Can carriers better detect the potential for fraud or false information before they bind the risk? The short answer: Yes, according to a series of broad-based studies looking at premiums and losses from more than 12 million policies across more than 110 carriers. The study was commissioned by Verisk Insurance Solutions – Underwriting, and performed by the analytical consultancy Inovatus. The consultancy focuses on delivering solutions addressing underwriting fraud and premium leakage.
Among the study findings:
- Participating carriers identified potential fraud in 7.4 percent of new policies that filed a claim within the first 125 days. The analysis was able to predict 84 percent of those policies (a total of 6.2 percent of new policies).
- In an astonishing 21.5 percent of policies with early (suspicious) claims, the carrier couldn’t positively identify policyholders and/or positively place them at the provided garaging address.
Fraudsters are beginning to use a “credit card” model to secure larger settlements, filing small initial claims to “test the carrier” and then following up with large multi-feature claims.
How did the insurers achieve those gains in fraud identification? The approach identified patterns of behavior in the information submitted through an insurance application when compared with data from multiple independent sources. The result was a multifactor model that scores the accuracy of an insurance application and its potential for fraud or rate evasion by examining nearly a hundred different variables.
How good was the model? In multiple studies, a regression of the model scores against actual loss ratios produced close correlations (an average R-squared of 0.93). An important additional finding of the study was that insurers couldn’t accurately detect fraud potential based on any single behavioral characteristic viewed in isolation. Precise detection comes from bringing together multiple characteristics to create an overall picture of the probability of fraud – and being able to quantify the degree of probability in a way that’s useful for taking effective action during the underwriting process.
That approach differs from a more traditional underwriting approach, where carriers implement rules that produce various “flags” during the underwriting process indicating the need for additional research and follow-up. Too often, those rules-based approaches rely on the judgment and experience of a handful of individuals. Carriers aren’t deriving the significance of any one or combination of rules empirically through data and analytics but from personal experience and judgment. And all too often, traditional rules-based solutions generate too many false positives, thereby undermining their credibility and causing carriers to ignore real instances of fraud or rate evasion.
Why is fraud becoming more of an issue? The nature of buying auto insurance has fundamentally changed. All market participants – consumers, agents, and call center representatives alike – expect convenience, speed, and ease of doing business. That means carriers have only a matter of seconds or minutes to size up the risk and ferret out any inaccuracies or inconsistencies. And technology is accelerating those trends. The majority of consumers now use the Internet when shopping for auto insurance. They may not always complete the purchase online, but they use it at some point during their shopping experience.
Requesting an Online Quote
To gain more insight from fraud trends, the studies included a review of more than a thousand call center phone calls as well as the “keystrokes” of more than 3,000 online sessions for individuals seeking a quote or policy through the Internet. One conspicuous pattern involved consumers who not only completed the quote process but also went through multiple iterations online. In those cases, the first completed quote contained extensive details, including what appeared to be every household member along with their corresponding information.
Once the system returned the initial quote with an estimated premium amount, the consumer almost immediately began to make changes. Youthful drivers might disappear. Consumers changed performance vehicles to nonperformance, or they modified dates of birth for children, making them appear older to avoid the carrier’s “youthful” categorization. In a handful of cases, they even modified gender to reduce the premium associated with youthful male drivers. Once the consumer was able to manipulate the premium to a more acceptable level, most of them abandoned the online session, and within a minute, a call center representative would quote and write a policy with exactly the last iteration of the online quote.
Carriers Can Make a Difference
What can carriers do to combat fraud? Basically, carriers need to step up their game in a big way. They’ve made large investments deploying technology and data to improve the customer and agent experience. But they’re falling behind in the race to identify fraud and rate evasion – a race they can’t afford to lose. Because rules-based processes tend to be very limiting and completely miss the sophisticated patterns that only become clear through the skilled application of data and analytics, carriers need to take a different approach.
The most effective antifraud measures, as shown by these studies, will likely be measures that use large stores of data with sophisticated analytics in real time. Most carriers will need to look outside their organizations to obtain and deploy the best antifraud solutions.