Insurance fraud cases expected to double with the rise of automation

Fraud detection has always been an important part of claims management. With the inevitable arrival of claims automation, it will become critical.

As insurers consider using AI based claim-handling bots to handle straightforward low-cost claims, a natural concern would be the resulting increased possibilities of fraud. At Shift we take this risk very seriously since we have built an AI model that is working towards automating claims. To examine the potential impacts claim automation could reap on fraud rates, let’s explore what, today, resembles the latter most closely: Straight Through Processing (STP).

What STP tells us about the relationship between fraud and automation

Many P&C insurance companies have started setting up STP, also known as “One-touch” or “Fast-track” processing to streamline a significant portion of their claim volume. Such processes usually consist of handlers paying claims directly without the involvement of loss adjusters and sometimes even without requiring invoices and quotations. Typically, insurers would apply STP logic to claims falling under a certain monetary threshold, where the pay-out would be less expensive than the costs associated with expert analysis. Some concrete examples include baggage claims under $500USD within travel policies and water damage under $1,500USD for homeowners’ insurance.

Despite still requiring a human touch, STP can be viewed as an important first step toward broader claims automation. And the manner in which STP expedites the payment process sheds light on the ways claim automation will heighten temptation for consumers to behave fraudulently.

STP has already given birth to new fraud trends

With the development of STP, here at Shift, we have identified two emerging fraud trends. Firstly, policyholders have been filing multiple similar, and completely fake claims, just under the STP monetary threshold set by their policies. Within the data of one of our partner travel insurers, we have seen a significant spike in policyholders filing claims for approximately $400 in the months following the implementation of STP. Similarly, we discovered a surge of policyholders faking multiple small water damages across several policies, each time providing nearly identical descriptions within the claim statements. Secondly, policyholders with meritorious damages are starting to exaggerate the costs for reimbursement up to the given threshold amount. This has been especially rampant where repair shops are aware of the number and deliberately encourage customers to fabricate the gravity of the damages. A leakage normally quoted at $500 might change to $1,500 following collusion between the two parties. Generally, the repairer will offer his services for free or at a reduced cost if the policyholder agrees to the scheme.

Both groups have figured out that if they keep their claims under $X, compensation will come without any questions.

The allure to commit fraud is compounded by the relatively low risk for legal repercussions. If an insurer decides to investigate one of these claims further, the policyholder simply drops the claim, while the insurer generally will not take legal action for such low financial stakes.

To combat this phenomenon some insurers have decided to set up some basic rules to limit their exposure to risk, such as a maximum of two STP claims per policyholder per year. Unfortunately, this type of rule is very easy for fraudsters to circumvent as they often communicate the method to their friends and family who can then do the same on their own policies. Clearly, insurers cannot afford two fraudulent claims per policyholder per year, but this is a legitimate, albeit bleak forecast for insurers that do not remain vigilant as they turn more and more towards automation.

Through the data, we have observed that when fraudsters understand the rules behind STP, such as the thresholds or the maximum number of claims per year, the fraud scheme usually spreads like wildfire among policyholders eager to make an extra buck. Just four months after an insurance company set up STP for all glass breakage claims under 900€, we saw sudden spikes in the rate of glass breakage claims concentrated in pockets of communities across the insurers’ geographical coverage zone.

Fraud schemes are also more likely to spread during natural catastrophes for which the associated claims are often treated using STP. Policyholders trying to game the system will capitalize on the insurer’s limited resources to send loss adjusters. During a recent flood in France we detected large numbers of policyholders, who lived clearly outside of the impact zone, using their friends’ flooded basement pictures to make fraudulent claims.

What we can expect with the rise of claims automation

With claim automation, the potential for fraud becomes even more dangerous. Based on our research and experience, we expect instances of fraud to at least double depending on the safeguards insurers have put in place as they turn to bots to handle their claims. Thinking he can outsmart the bot, a policyholder might feel more at ease testing several different versions of his claim statement, until he finds the circumstances that warrant coverage.

Insurers should be very cautious as they apply deterministic sets of rules to their claim-handling bots. If the claim handling bot is programmed to process claims individually and does not have the capacity to detect trends, once a fraudster succeeds in cracking the system, he’ll know that he can repeat the process. The fraudster will inevitably share information with his network, encouraging them to claim exact same accident, with similar circumstances and documents, knowing the bot will pay.

For these reasons, it is crucial for a claim-handling bot to integrate a powerful and comprehensive fraud detection solution with the intelligence to compare multiple seemingly unrelated claims to detect unusual similarity in circumstances or invoices, and detect statistically unlikely trends of claims to raise the red flag for human investigation.

Claim automation faces several complex challenges, among which we can cite: reaching a precise estimate of the amount of the claim, maintaining a good customer relationship, understanding the context and the damage with enough precision to propose an appropriate settlement method (payment on estimate, payment on invoice, sending a loss adjuster…) While bots will always have the recourse to redirect claims back to a human handler at the customer’s request, customers are unlikely to voluntarily notify insurers of fraud detection deficiencies, allowing for fraud schemes to run rampant and millions lost before the insurer catches on.

Thus, the biggest challenge facing insurers wishing to automate claims is not producing a bot capable of going through steps as one might expect. Insurers must first and foremost protect themselves from their increased vulnerabilities through integrating an adequately robust combination of fraud detection tools to safeguard the billions at stake.