Agentic AI removes ‘barrier’ for fraud investigators across different insurance lines

When artificial intelligence (AI) is spoken about in the fraud sector, it is often in relation to how it is being exploited by fraudsters to manipulate insurers and customers.

But, as the insurance industry seeks to keep pace with this evolving threat, joining the AI revolution also offers opportunities for counter-fraud teams to beat these tech-savvy criminals at their own game.

According to data analytics and technology firm EXL’s 2025 Enterprise AI Study, 36% of UK insurers had progressed AI in their top use cases, while 52% had deployed generative AI in industry-specific applications beyond the pilot stage.

With adoption rising, agentic AI has opened up new possibilities for fraud investigation that “stretches out to different types of insurance” previously constrained by the volume and complexity of their data.

This view was shared by George Robbins, head of UK, Ireland and Nordics at Shift Technology, during the recent Insurance Times webinar entitled Fraud or Claim? Making Agentic AI work for the investigator, hosted in association with Shift Technology on 15 October 2026.

Robbins noted that current fraud investigators had to navigate numerous internal and external data sources and alert systems, leading to what he calls “swivel chairing”.

He said that for many years fraud investigators have had to swap between “multiple systems, multiple sets of data, multiple workflow processes not really integrated together”.

Agentic AI changes this by removing much of the heavy lifting involved in acquiring and summarising data, he explained, while also providing the “possibility of executing actions” that would typically not be economically viable or “high value” enough to pursue.

With agentic AI systems able to action “onerous tasks” at the “click of a button”, Robbins believes this makes parts of the counter-fraud process accessible in ways they were not before.

“A lot of commercial insurance, for example, is heavily characterised by large quantities of documents and that large quantity of unstructured data has traditionally prevented systems working well in that environment,” he continued.

“But nowadays that barrier has gone away. Not only can we work in property and casualty, but we can work in commercial, we can work in health – we can integrate sets of data that weren’t possible before.”

Maintaining accountability

While agentic AI adoption removes barriers for fraud investigators, the market has become increasingly aware of the need for guardrails when deploying autonomous technology.

Nick Rawlings, director of transformation at Covéa Insurance, said that while the insurer delegates tasks to AI it “can never delegate accountability” as that firmly sits with a human.

He explained that the insurer has considered which tasks machines and humans are respectively better suited to in order to maintain a “balance between the two”.

While AI is particularly useful for processing and synthesising large amounts of data, spotting patterns and filling in gaps, he said that human judgement remained essential.

This is especially the case when interacting with people because, as Rawlings put it, “it’s easier to lie to a machine than it is to lie to a human being”.

“There’s a line between letting AI run off and do a whole process versus every two minutes it is coming back to you to validate what it wants to do next because that creates a lot of friction in the process.

“It’s a fine balance and everyone’s trying to work out where those lines are at the moment.”

Echoing this view, Robbins said that governance “should never allow” the insurance industry to “abdicate responsibility” because its role is founded on “looking after the few in their time of need”.

He believes investigators must remain in the loop, warning that if firms reduce headcount and remove expertise, there may be nobody left to “challenge its decisions”.

He continued: “In a regulated industry, that is always going to be absolutely essential.

“We have to understand how the technology is working, what decisions it’s giving us, and we have to be able to challenge that and understand where it may be going outside the guardrails that are actually correct for our industry”.

For Robbins, this requires current fraud investigators to be upskilled in using AI tools and building the corporate knowledge repository that AI systems can help provide for the next generation”.

“The industry is becoming a much more knowledge-based environment [and] a much less manual environment,” he explained.

“The quality of your career will be decided by your ability to make great decisions.”

‘Quality of decision-making’

The biggest value from using agentic AI in counter-fraud lies in “the quality of decision-making”, Robbins noted.

He explained: “It’s in the quality of decision-making because, for example, fraud is a throughput problem.

“It’s a combination of how much you detect, how accurately you detect it and how fast you can process it,” he continued.

“When you change the dynamic on how much time somebody needs to spend on an investigation or the fidelity with which they can find things, what you’re actually changing is the throughput of the system – the output benefit of prevented fraud.”

While agentic AI can improve access to data and information, Rawlings said that Covéa’s business case centres on “cost avoidance” which ultimately “blows the efficiency savings out of the water”.

He added: “There are efficiencies, but you wouldn’t write a business case for this kind of process based on efficiency savings. It’s not the principal driver of value in the business.”