Quality Data Drives Moves Against Motor Insurance Fraud

Metromile, the pay-per-mile auto insurer, had a problem: when customers had claims, their satisfaction plummeted. Looking into it, execs realized that, while 70% to 80% of claims were easily validated without any additional support, all claimants were subjected to fraud-detection processes, which annoyed customers and delayed the resolution process. “We found the vast majority of customers are being honest but those customers were still being treated like the potentially fraudulent ones,” says Dan Preston, Metromile’s CEO.

The solution was artificial intelligence and machine learning tools that could instantly take data from Metromile’s OBD-installed device and validate the insured’s statement. Dubbed AVA, the solution can make the determination in seconds, while automating the claim adjustment process to get a car into a repair shop within hours. Customers were happy and so was Metromile’s CFO because by focusing on suspicious claims, instead of using the same process on all claims, the claims operation ran at a 60% lower cost.

Analytics, AI and machine learning

As insurance carriers go digital, and collect reams of data they can use to build better models, they have powerful weapons against fraud. A survey by Willis Towers Watson found that 82% planned to use advanced analytics to evaluate potential fraud while only 26% currently used it. The firm found barriers to expanding use of big data, as well such as infrastructure and/or data warehouse constraints (51%); difficulty in accessing or integrating data (41%) and IT or services bottlenecks (33%).

Jason Rodriguez, North America data science lead for insurance consulting and technology at Willis Towers Watson, says that unstructured claim information is one of the most promising big-data endeavors for insurers when it comes to fraud prevention and detection. Structured data in current databases can be limited by what was deemed important at the time they were set up, Rodriguez says. “Within unstructured data, you might have additional information and details around handling claims that you didn’t have the foresight to include in the data table or claims administration system.”

The development of machine learning and artificial intelligence techniques are critical to extracting information while ignoring data “noise,” according to Rodriguez. That, along with the development of machine learning and artificial intelligence techniques to do things like identifying themes within text or retrieving targeted information are enough to provide real value to insurers.

Fraud detection, however, is never at the forefront of insurers’ AI endeavors, according to Thomas Hallauer, research and marketing director of Ptolemus. “AI can make the claims process more transparent and more efficient,” he says. Fraud prevention or detection “is a by-product of dealing properly with claims by setting up a number of alerts linked to anomalies.” However, the software will detect the claims cost or the expected claims cost from data or a photo and be able to flag if a claim is much higher than what it expected. “In any case,” he says, “I do not believe the technology will replace adjusters, it is mainly a way to make the remote adjusters more effective.”

Learning from ecommerce

In addition to using insurers’ data for fraud detection, Lisa Volmar, senior director of product development for TransUnion’s Insurance Solutions, thinks they should take advantage of other data sets and models, for example, those used in the retail, financial services and ecommerce industries.

Volmer says: “They have huge volumes of transactions every day that all the industries, including insurance, can learn from. While the typical interaction between a consumer and an insurance carrier may be a handful of times during a policy period, in the financial industry, with dozens of transactions a week for every consumer, there is a large volume of information and data out there that we are beginning to tap into.”

In order to approve or deny an online credit card application or purchase in seconds, processors pull in a wide variety of data: is the device, IP address or mobile phone number being used known to be associated with that identity? Is the volume of activity normal for this identity?

“We’re getting more granular and looking at more combinations of behavior and velocity,” Volmer says. “It’s not just, ‘Is this Social Security number valid?’ Instead, it’s, ‘This Social Security number is valid but it’s been used at four addresses in the last 90 days.'”

There are different fraud parameters for insurance, she notes. For example, someone may research a number of credit cards but not normally open five in one day. Yet, it’s very common for customers shopping for insurance to go through applications at several different companies.

Volmer thinks insurers can benefit from “hive thinking,” that is, working together in fraud prevention networks. “There’s pressure for carriers to work together to solve these problems.”

Secure collaboration

Insurers also have pressure to keep their data proprietary, not only for customers’ privacy but also for competitive advantage. Blockchain technology is an ideal mechanism for the secure sharing of data and intelligence among insurers, according to Elena Travkina, director of digital transformation for insurance at Altoros, a provider of software development and managed services.

One use case is preventing multiple claims on the same accident. Because insurers don’t share databases, it’s difficult to track cases. By using blockchain’s decentralized, shared ledger, insurers would be able to see when the first claim on the accident was processed. “Blockchain has characteristics that help to make the process transparent and make it totally accountable,” Travkina says.

Another use case would be creating blockchain-based digital fingerprints for vehicles. The chain could include information about the car’s country of origin, ownership, history and claims history, as well as odometer readings as it changed hands. To get this going, Travkina says insurers should work as a broad consortium, although an individual insurer could begin using blockchain internally. One of the biggest barriers to this is regulatory, according to Travkina. “This is the most conservative part of the community. We can understand the advantages but we still don’t have clear regulations,” she says.

New attack surfaces

While use of big data among insurers is increasing, fraud is on the rise. According to the Coalition Against Insurance Fraud, while most insurers have embraced or expanded fraud-fighting technology, 61% of major insurance companies reported that fraudulent claims had increased significantly between 2013 and 2016. In addition, fraud schemes are getting more complex. In fact, says Dan Viza, vice president of global business development and strategy for Trillium Secure, their big data initiatives subject insurers to new forms of fraud. Trillium provides an automotive cyber-security platform. “Any industry going through digital transformation creates a number of different risks, and the auto industry is no different,” Viza says.

UBI devices, embedded modems and internal networks, as well as communications between cars and insurers, carmakers and service providers, are all attack surfaces. Moreover, in addition to the need to keep data from being stolen, insurers must be sure their data is authentic, Viza says.

Whether the insurer is crunching risk models or applying AI to detect fraud, bad data will lead to erroneous results. “You need to know that you’re getting inherently good data,” Viza says. “Making sure there’s no misappropriation of data along way is the key to ensuring the integrity and value of the data used for business decision making. It’s incumbent on insurers to make sure that the data they are using for all those models, services and financial decisions on is safe, secure and authentic data.”

Leave a comment

Your email address will not be published. Required fields are marked *