The U.S. insurance industry consists of more than 7,000 companies and collects approximately $1 trillion in premiums each year. Fraudsters are eager to get their hands on these massive sums and many of them are succeeding, with the FBI estimating that more than $40 billion in annual non-health insurance payouts are paid for fraudulent claims. The health insurance market is even more vulnerable: Up to $68 billion is lost to health insurance fraud each year.
Fraud doesn’t just hurt insurance companies, however, as the losses are largely passed on to legitimate customers in the form of increased premiums. The FBI calculates that such fraud costs average American families between $400 and $700 annually. It is therefore imperative that insurance companies understand the methods fraudsters use and that they develop the appropriate countermeasures.
The following Deep Dive explores the ways insurance fraudsters — individual scammers as well as service providers and insurance adjusters — leverage false claims and stolen identities to receive payouts, and how advanced tools like artificial intelligence (AI) and machine learning (ML) can stop them.
Types of Insurance Fraud
Most insurance fraud involves putting false claims on applications, acts that comprise up to two-thirds of denied attempts. Bad actors trying to scam automobile insurers may say their cars are worth more than they actually are so they can get larger payouts, for example. Others file false incident reports, claiming insured possessions were stolen when they were not, or even stage elaborate ruses to convince insurers of fake injuries or damaged items.
Some fraudsters file insurance applications and claims under false identities, which is particularly common in the health insurance space, with individuals attempting to receive payment for friends or family members who lack their own policies. These tactics are not necessarily undertaken to cause harm, but they still result in insurance companies driving up premiums to cover their losses. Such efforts are sometimes more nefarious, however, with bad actors stealing identities belonging to Medicare users, ordering costly medical equipment and paying for it with victims’ health insurance policies. The perpetrators then resell the equipment for a profit before victims or insurance companies notice the scam.
Sometimes insurance fraud originates not from scam applications, but from service providers working with insurance companies. A provider might overbill the insurance company for services rendered and pocket the difference, such as an auto body shop claiming that a service costs $200 in parts and labor when it actually only costs $150. Those types of incidents are difficult to verify because the only paper trail is the auto shop’s own receipts, which it could doctor to cover its tracks. Small episodes like these may seem minor, but they can add up over time and cost insurance companies millions of dollars.
Even adjusters working for insurance companies have been known to engage in fraudulent behavior, exploiting their inside knowledge of the industry to overcharge customers on their premiums, “cook the books” or even take bribes. One high-profile incident took place in Florida, where an adjuster working for Citizens Property Insurance was busted for shaking down companies and homeowners — including extorting a service provider to give him a cut of the money — before becoming a state witness in the matter.
Many businesses rely on strictly manual means to identify and stop insurance fraud attempts, but the sheer volume of applications and claims filed every day makes this largely untenable. Companies are thus increasingly turning to automated solutions to inspect these claims en masse.
How AI Can Limit Insurance Fraud
Eighty-three percent of businesses in North America currently use manual insurance claim review processes, but these present many problems when detecting fraud. The first is capacity, as wholly human systems have no way to analyze and approve the quantity of applications firms receive daily: The home insurance industry sees 5 percent of American homeowners filing a claim at least once a year, for example. The second drawback is unreliability, bringing about a rate of false positives as high as 15 percent and preventing legitimate customers from conducting business.
Many insurance providers are therefore using manual review processes as backups to AI- and ML-enhanced tools, which play central roles in identifying and stopping insurance fraud attempts. AI can analyze vast quantities of data in a fraction of the time it takes human analysts. It can then provide more holistic views of trends, while human analysts must separately study each claim. Such systems can highlight those that are abnormally expensive compared to industry averages and flag them for further review.
AI-enhanced programs have become so sophisticated that they can now analyze photographs to determine whether customers’ claims are legitimate. Visual analytics systems can inspect an image of a damaged car, for example, make an educated estimate of the cost to fix it and compare that to the customer’s claim. This can cut down on costs by limiting both customers’ abilities to exaggerate damages to get higher payouts as well as autobody shops’ abilities to overcharge insurance companies.
These systems are not infallible, however, and many companies still employ human analysts to double-check their findings. Such offerings are particularly weak when first launched, as they have relatively small knowledge bases at that point, meaning they could let fraudulent claims through and stop legitimate customers from getting paid. Human teams can coach the AI programs through these early stages by reviewing flagged claims rather than allowing the systems to make decisions about whether to pay customers. Human analysts can then take backseat roles as the systems become smarter over time.
AI-based fraud-fighting systems are advancing quickly, but so are bad actors, who are developing even more convincing ways of filing false claims under false identities. Solutions that tap into such technologies may never be enough to completely stop insurance fraud, but they could reduce the average premium increases each family experiences.