Fraud Prevention

How Allstate Leverages AI To Show Insurance Fraudsters The Door

Fraudulent claims in the U.S. account for more than $80 billion in payments a year, with legitimate customers paying higher premiums to make up for the losses. While digital claim filing has made processes more frictionless than ever, it’s also opening new doors to fraud. In this month’s Digital Fraud Tracker, Greg Firestone, VP of data science for insurance provider Allstate explains how the company is using AI to keep claimants honest and the review process free from adjusters’ biases.

Insurance fraud is not a new phenomenon, but it is a prevalent one.

Fraudulent claims account for more than $80 billion a year in the U.S. and comprise up to 10 percent of all insurance payouts, meaning legitimate customers are often forced to pay higher premiums to make up for the losses.

Insurance providers are taking action to mitigate such problems and keep their operations and customers safe from fraud, however. Approximately 95 percent of such firms leverage anti-fraud technology, Greg Firestone, vice president of data science at Allstate, told PYMNTS in a recent interview. Employing such solutions is not enough, however.

“It’s very hard to measure sometimes, but it’s happening,” Firestone said. “The best prevention is really being aggressive: using AI and data to find [fraud]. Data is your friend in this regard. [Fraud is] a problem that impacts all insurance companies, and we need to focus on it and make sure the fraudsters realize that we’re not easy marks.”

Allstate is no stranger to insurance fraud, which Firestone said has increased as digital means make it easier to file false claims en masse. The company leverages an AI-based solution to monitor and flag suspicious claims, but keeping an eye on future fraud trends still requires a human touch.

Insurance Fraud Goes Digital

Insurance fraud has only grown more prevalent as claim-filing becomes more digital. What was once an elaborate process that required bad actors to create or compile false identities, in-person deceptions and coordinate false information across a range of forms can now be automated and completed online without a single human interaction.

“The ability to buy a policy online allows a fraudster to create a lot more fake policies than they would in the past,” Firestone explained. “[They are using] some of the digital technologies [to] submit basically the same claim [repeatedly] very quickly using similar photos, unless [the receiving company has] robust capabilities to stop that kind of fraud.”

Insurance companies have set up some capabilities to help users process claims quickly and with fewer interactions with adjusters, and they may have been successful in that regard. Processes are more frictionless than ever, but creating such smooth pathways has opened new doors for fraud, enabling bad actors to more easily file false claims or use fake identities. Every level of convenience and seamlessness can be exploited, especially if the recipient is seeing large volumes of submissions.

“We have the ability to file claims using photos, which is great for our honest customers, but it [sometimes makes it] easy for a fraudster to come in and submit a claim that didn’t really happen [by going] so quickly through our process,” Firestone said. “[The problem is] we don’t always catch it.”

Large insurance companies process so many claims daily that it would be  impossible for human analysts to properly inspect each for malfeasance, but the costs of not flagging fraud before it is paid out are also high. Many are thus leveraging sophisticated AI systems to automate this process and reserving their human teams for claims the AI has flagged as suspicious.

How AI Curbs Insurance Fraud

Allstate’s AI-based insurance fraud prevention system analyzes numerous variables in each claim to determine fraud’s likelihood. These clues can sometimes be obvious, Firestone noted, but would likely be lost in the noise without an AI-powered platform.

“One silly but classic example is, if your home burns down, [but] your dog and all your wedding album photos are with you, that’s a sign that maybe you knew your house was about to burn down,” he said.

AI-powered systems also look at the quantity of claims individuals file as well as their timing. One possible fraud indicator would be a customer attempting to file several claims in sequence, making minor adjustments to each in hopes that one will be accepted and paid out.

“Lots of people have their wedding rings stolen every year, but not many people have [them] stolen five or six times,” Firestone explained. “That’s a sign that maybe there’s fraud going on. We’re not talking [about] just once or twice, but [submitting] very high [numbers of] and very similar kinds of claims would be a warning flag for us.”

AI-driven tools can pick out these warning signs among thousands of claims in a fraction of the time it would take human analysts to do so. A human analytics team backs up Allstate’s system, however, as it cannot deny claims, Firestone noted. The solution passes that responsibility on to the analysts.

“Humans definitely are a very valuable part of our process of detecting and fighting fraudulent claims that don’t smell right to them and using their instincts,” he said. “If [their] investigation proves to show nothing, then there is no impact to the consumer. They continue to get paid — it just might take a little bit longer than it would have otherwise.”

Sometimes AI-powered systems detect new fraud methods that neither they nor human teams have seen before. This is where a second human team, Allstate’s Special Investigative Unit (SIU), steps in. This group is dedicated to anticipating future fraud threats and developing solutions before they further target the insurance industry.

Human Intuition Versus Big Data

Reacting to fraud as it arrives is one thing, but anticipating tomorrow’s threats is another — and the only way to truly put a dent in insurance fraud. Fraudsters are constantly innovating, and their new techniques will not be caught by looking at what has been done in the past.

“The model of being reactive is no longer viable,” Firestone said. “We need to think about how we stop fraud on the front end, which allows us to save our customers money and offer the best customer experience.”

Allstate’s SIU is currently in charge of this task, but the sheer number of fraud techniques means an approach relying entirely on their expertise and intuition may not be viable in the future. Fraudsters are increasingly relying on automated means to conduct their schemes, and insurance companies need to fight fire with fire.

“The industry as a whole is becoming more digital, and that is a whole new set of vectors,” Firestone explained. “The ‘spider-sense’ of our claim adjusters will be something that we can rely less and less on. How do we use AI and machine learning [ML] to replace that with data?”

The good news is that AI and ML are improving as fraudsters’ techniques become more sophisticated. Companies like Allstate that continue to invest in these technologies may never completely solve the fraud problem, but incorporating more data use could be enough to reduce insurance fraud losses.


New PYMNTS Report: The CFO’s Guide To Digitizing B2B Payments – August 2020 

The CFO’s Guide To Digitizing B2B Payments, a PYMNTS and Comdata collaboration, examines how companies are updating their AP approaches to protect their cash flows, support their vendors and enable their financial departments to operate remotely.