AI helping insurers fight bogus claims fuelled by fraud-as-a-service template farms

Criminal gangs and opportunistic consumers are increasingly using “template farms” to generate fake receipts and invoices for bogus insurance claims. In the light of such booming “fraud-as-a-service” operations, insurers are deploying threat-detection artificial intelligence (AI) solutions to weed out fraudulent claims.

Data from fraud detection company Resistant AI found that between January and May 2025, 6.4% of all documents it tested were fraudulent.

Traditional fraud tools generate a “huge number of false positives that can overwhelm fraud teams”, according to Taskin Ismet, head of travel cost containment and fraud prevention officer for insurance group AXA in Spain. The advantage of AI tools is in “pinpointing fraudulent documents”, which insurance investigators can then “focus their efforts” on, he said during a recent webinar.

In April, insurer Aviva said it had detected 11% more fraudulent claims in 2024 than a year earlier. Aviva said it had stopped more than 12,700 fraudulent claims, worth a combined £127 million over the course of the year.

Aviva also worked with the insurance fraud enforcement department at the City of London Police to secure the first use of both a Serious Crime Prevention Order (SCPO) and the Proceeds of Crime Act (POCA) against an insurance fraudster who submitted 62 fraudulent motor insurance claims.

Criminal gangs and opportunists

Those making fraudulent insurance claims fall into two camps: opportunistic consumers and professional criminals. Generally, the former opts to bump up a genuine insurance claim with a couple of (expensive) false items, while the latter invents an entire claim from scratch.

Historically, a lack of imagination among many opportunistic fraudsters is what often got them caught because insurers have always flagged up expensive “designer” items in insurance claims for further investigation. However, the advent of template farms — run by criminal gangs who offer fraud-as-a-service — now enables such opportunists to purchase a supposedly genuine receipt to submit with their claim.

“If we take baggage claims, where there may be a legitimate loss, the feeling [for some opportunists] is ‘if I’m applying anyway, I may as well throw in a couple of extras, like an iPad, etc’,” Ismet said, adding that Louis Vuitton shoes and handbags are a popular choice. “No one ever seems to shop at [high street store] Zara,” he joked.

Resistant AI product manager Kristyna Schusterova explained that software can now flag up when the same receipt is being submitted across multiple claims.

“We have a dedicated team working on this. Already they have been able to identify over 150 farms offering more than 300,000 different templates, from 180 countries,” she said. “So those are staggering numbers. And so far we have focused only on farms operating in the English language.”

Vit Stachovsky, an anti-money laundering (AML) officer at Czech insurance firm Kooperativa, said he had seen cases where family members had submitted claims using the same medical documents that had been individually modified.

Fraudsters also share details of any vulnerabilities in the claims process. “They know the pulse of the system and they are sharing the experience between them,” he explained.  

Schusterova at Resistant AI believes Gen AI tools could enable the evolution of a notorious form of insurance fraud known as “crash for cash”. In these scams, criminal gangs have traditionally staged car crashes and then submitted claims, not only for the damaged vehicles, but also for injured motorists and passengers.

However, Gen AI can now create fake images of damaged vehicles — thus removing the need for criminals even to stage an accident before submitting a bogus insurance claim.

EU AI Act

Insurers have also raised concerns about having to show the workings of their fraud detection AI in case disclosing these details enables criminals to exploit the software.

In May, Insurance Europe said: “A company using AI for fraud detection purposes should be able to decide not to share information or provide explanations about the model or data it uses to certain audiences, in light of concerns over model manipulation or exploitation.” The trade body was responding to a consultation on AI governance by the European Insurance and Occupational Pensions Authority (EIOPA).

Insurance Europe advocated that EIOPA “leveraged advice” in guidance issued by the UK Information Commissioner’s Office and the Alan Turing Institute on explaining decisions made by AI, rather than requiring full model disclosure.

Earlier this month, EIOPA published an opinion on AI governance and risk management, in which it acknowledged that AI offered insurers significant opportunities for combating fraud. However, it also reminded insurers of the requirement to document situations where the use of AI might lead to unfairness — such as claims being unjustly rejected.