Simon Glairy brings a deep understanding of how artificial intelligence and human behavior intersect in the world of risk management. As insurance fraud now accounts for 44% of all reported crime in England and Wales, reaching a staggering cost of £219 billion annually to the economy, his insights into the evolution of these crimes are more critical than ever. We explore the transition from physical staged accidents to sophisticated telephony-based scams, the rise of AI-generated deception, and how the industry is using metadata and large language models to fight back. This discussion highlights the hidden vulnerabilities in call centers and the shifting tactics of organized criminal networks.
The landscape of motor insurance fraud is changing, with a noticeable shift from the dramatic staged collisions we used to see toward more subtle forms of exaggeration. Why are fraudsters moving away from physical accidents, and what does this mean for the industry’s detection strategies?
The reality is that orchestrating a physical collision is high-risk and incredibly messy for the perpetrator. Fraudsters are realizing that they don’t need to risk their lives or attract the attention of the police on the side of the road when they can simply pick up the phone and inflate a legitimate minor incident. We are seeing a massive pivot where more than seven in ten fraudulent cases detected by major insurers are now linked to motor insurance, but the “staged” accident is being replaced by the “exaggerated” claim. In 2025 alone, insurers like Aviva uncovered more than 18,400 suspect claims worth a total of £233 million, which breaks down to about £638,000 in fraud detected every single day. By focusing on repair costs, credit hire fees, and soft-tissue injury claims, criminals can hide behind the noise of rising inflation and cost-of-living pressures. This shift means our detection strategies have to move away from the physical crash site and into the data—specifically, the conversations and documents that support these inflated costs.
You’ve mentioned that the telephony channel is a critical but often overlooked source of risk intelligence. How exactly do criminals use phone calls to scout for vulnerabilities before they even file a claim?
Telephony is the “soft underbelly” of many insurance operations because it feels so human and immediate, making it a prime target for social engineering. Fraudsters aren’t just calling once to lie; they are conducting reconnaissance calls to test the defenses of a contact center, often trying to figure out which agents are more lenient or what specific information is needed to bypass authentication. They might call multiple times to see how agents handle interactive voice response systems or to see what kind of customer data they can extract through “vishing.” These early-stage calls often leave what I call “digital breadcrumbs”—brief test calls, failed IVR journeys, or unlogged numbers—that many companies unfortunately discard as noise. If an insurer isn’t linking these repeat attempts to a single actor, they are missing the signals of a coordinated attack that is happening right under their nose before a single penny has even been claimed.
Large language models and AI are often discussed as tools for the “bad guys,” but how are these technologies actually being used by fraud teams to make sense of thousands of hours of recorded calls?
For years, the sheer volume of call data was simply too unwieldy to parse, and historically, fraud platforms struggled to turn those qualitative conversations into something actionable. Now, LLMs are finally “bringing structure to the unstructured” by allowing us to transcribe, analyze, and categorize the sentiment and intent behind every word spoken in real-time. It isn’t just about converting speech to text; it is about recognizing patterns across hundreds of different contacts that might seem isolated but are actually linked by the same script or tone. We can now use spectrographic analysis to find anomalies in the audio itself, which is vital because we are starting to see organized gangs use synthetic, AI-generated voices to communicate with us. By using technology to highlight stress, adverse sentiment, or even the “robotic” frequency of a deepfake voice, we are turning the fraudsters’ favorite tools back against them.
Ghost broking seems to be a particularly predatory form of fraud that targets unsuspecting consumers. What do the numbers tell us about the scale of this problem, and how are these middlemen managing to stay under the radar?
Ghost broking is a growing menace that is becoming increasingly sophisticated, with the number of these fraudulent policies rising by 7% year-on-year. In 2025, Aviva alone detected over 105,000 fraudulent insurance applications, and a huge chunk of that is driven by these fake middlemen who pose as legitimate brokers. They use compromised identities—often harvested from larger cyberattacks—to build up “ghost” personas that look perfectly real on paper. They might take out a policy, show the victim a genuine-looking document, and then cancel it immediately after pocketing the victim’s “premium.” The victim often doesn’t realize they are driving uninsured until they are pulled over by the police or, worse, involved in an accident. It is a heartless crime that relies on the speed of digital quotes, which is why we are now focusing so heavily on the application stage, using behavioral insights to disrupt these actors before the policy is even issued.
When a claims handler is on the phone with a potential fraudster, what are the specific behavioral or verbal red flags that suggest they are dealing with an organized criminal rather than a stressed customer?
The most telling sign is often a strange mix of sketchiness and defensiveness; a legitimate claimant usually remembers the emotional details of an accident but might be fuzzy on the exact technicalities, whereas a fraudster often has a perfect “script” but falls apart when you ask a question off-track. Criminals will often become aggressive or pushy very quickly, trying to apply pressure for a fast settlement to avoid a deep dive into their story. We also look for metadata signals that the handler might not “hear,” such as a geographical mismatch where the person claims to be in London but the call is originating from an entirely different location or using a spoofed number. If the caller seems to be reading from a script or if their “recollection” of events feels rehearsed rather than remembered, our experienced handlers know to flag it for a secondary review. It is that human-to-human interaction, backed by data, that remains one of our strongest lines of defense because it is very difficult to fake genuine human memory and emotion under pressure.
With the government’s new fraud strategy and the creation of the Online Crime Centre, how important is industry-wide collaboration in closing the gaps that individual insurers can’t see on their own?
Fraud is a borderless crime that doesn’t care about the boundaries between different insurance companies or even different sectors like banking and retail. If a fraudster is successful at one insurer, they will immediately take that same playbook to the next one, which is why real-time data sharing is the only way to truly win this war. No single organization has the full picture, but when we pool our intelligence, we can see the same phone number or the same synthetic voice popping up across the entire industry. The Online Crime Centre is a huge step forward in creating a centralized mechanism for this kind of collaboration, allowing us to bridge the gaps in our intelligence. By sharing patterns and behavioral indicators in real-time, we move from being reactive—catching fraud after the loss—to being proactive and stopping the criminal at the front door.
What is your forecast for the future of telephony-based fraud as AI continues to evolve?
I believe we are entering an “arms race” of synthetic media where the battle will be fought between AI-generated deception and AI-driven detection. We will see fraudsters using increasingly perfect deepfake voices that can mimic a customer’s actual voice and tone, making it nearly impossible for a human ear to tell the difference. However, I am optimistic because our detection tools are becoming just as sophisticated, moving beyond just “listening” to analyzing the invisible metadata and spectrographic signatures that no AI can perfectly replicate yet. The insurers who survive this shift will be the ones who stop viewing a phone call as just a conversation and start viewing it as a rich stream of data that needs to be scrutinized with the same intensity as a financial statement. The future of fraud prevention isn’t just about better locks on the doors; it’s about having the intelligence to know who is knocking before we even answer.
