Spotlight: Shifting fraud detection to the point of application

As AI and deepfake technology enable ever more convincing scams, insurers are being forced to invest in real-time identity verification, behavioural analytics and document forensics.

Insurers we spoke to are now taking a more preventative approach. Markerstudy’s head of fraud, Clare Lunn, warns: “Allowing fraud into your business creates a high level of risk, increases bad debt and raises operational costs to investigate, and will have a real negative impact on reputation.”

Josh Barnsdale, chief technology officer at One Call, says: “We spend a lot of money implementing fraud controls at application stage because it’s much easier to stop someone getting cover in the first place than subsequently having to prove a claim is fraudulent. In motor insurance, even if you manage to prove it’s fraudulent, you are still exposed to third-party liability.”

One Call uses device ID checks to help identify fraudsters. “If we can identify a device that’s been used multiple times for multiple applications, it’s a strong indicator of a possible problem even though the identity and customer information may be genuine.”

We spend a lot of money implementing fraud controls at application stage because it’s much easier to stop someone getting cover in the first place than subsequently having to prove a claim is fraudulent.

Josh Barnsdale, One Call

Katie Davies, Ageas’ director of underwriting services and fraud, claims that when their ID theft model went live, they quickly saw a 72% reduction in fraud coming in. “The more data fed in, the more accurate the model becomes. If we look at the ABI fraud calculations, it saves us about £1m a year in fraudulent spend.”

These machine-learning models placed at point-of-quote and application stage are beginning to have a real impact. “Markerstudy now has various machine-learning models to prevent fraudsters taking out policies, which has reduced the number of staged-accident claims down the line,” Lunn says.

“However, some fraud will always get through however good one’s controls are, and so insurers still need defences across the entire lifecycle of the policy,” she adds.

“Where fraudsters have evolved their tactics to avoid insurers’ point-of-quote controls, we have needed to bolster our defences to changes made to the policy at mid-term. Each day, we rank each policy and claim for fraud risk, ensuring we focus on the most pressing risks across the customer cycle,” says Stuart Stevens, head of counter fraud intelligence at Direct Line.

Embracing AI

As fraudsters continue to develop their tactics, using synthetic identities and spoofing-as-a-service platforms, they can commit substantial crime and financial harm at volume and scale. “The shift from opportunistic to industrialised fraud means insurers must adapt quickly, using smarter tech to detect deception across digital channels,” says Mike Haley, chief executive of Cifas.

He explains how AI is transforming fraud detection from reactive to predictive. Insurers now use multimodal AI to analyse text, images, voice and behavioural data in real time. This enables early detection of anomalies, deepfake manipulation and synthetic identities.

The more data fed in, the more accurate the model becomes. If we look at the ABI fraud calculations, it saves us about £1million a year in fraudulent spend.

Katie Davies, Ageas

AI models are being trained on historical fraud patterns to flag suspicious claims. The key is combining AI with human oversight to ensure accuracy and fairness.

Some insurers have set up in-house specialist teams to develop their own tools. For example, One Call’s development team recently put all their fraud data into a machine-learning model to identify trends and behaviours. This provides the fraud team with an additional flag on whether something is highly or least likely to be fraud. A fraud team member would then review those files to decide whether its fraud or not.

Barnsdale argues it’s not just staying ahead of the fraudsters, but also of one’s peers. “We do it because we want to add value to the insurers we work with, and if those insurers see value, as in better performance, we can write more business.”

More tools!

Offering huge potential in early fraud detection is establishing customers’ digital identities by harnessing global shared intelligence from millions of daily consumer interactions. For example, LexisNexis Risk Solutions ThreatMetrix solution looks holistically at connections between devices, locations, email addresses and threat intelligence to detect anomalies that might indicate fraud right from a first application.

“Digital intelligence can help identify normal behaviour. For example: is that customer using the same details as when logging into their banking app?  Insurers already have a lot of physical data, but we are contextually layering over what insurers’ fraud prevention kits already have with something that financial services have,” Paul Brockway, LexisNexis Risk Solutions’ director of insurance and investments, explains.

“It’s also about customer experience: if you have better trust in a particular transaction, you can be more lenient on pricing.”

But more could still be done at the ID stage, he says. “Comparison sites have huge amounts of data that they could easily share. However, it would have to be done universally as they would be limiting their sales going through.”

Meanwhile, voice and image analytics are being widely adopted. Image analysis software helps identify ‘shallow fakes’ where fraudsters doctor or even fabricate images or documents. Lunn says Markerstudy is currently developing an image validation tool that would help identify an image’s exact location.

Biometric checks – where facial recognition is matched to a document – could also create significant operational efficiencies. LexisNexis Risk Solutions is currently assessing the value and process of this type of solution, with customer expectations in mind.  

Insurers already have a lot of physical data, but we are contextually layering over what insurers’ fraud prevention kits already have with something that financial services have.

Paul Brockway, LexisNexis Risk Solutions

Tight budgets

However, with budgets tight, insurers must make the most of their investments.

“Fraud controls need to be real time to keep up, which can be more challenging to build and add more costs to the development of the automation process. But it’s about getting the balance right of handling genuine customer requests as quickly as possible whilst identifying and preventing fraud in order that premiums don’t keep going up,” says Markerstudy’s Lunn.

“Therefore, we need to be creative and look at ways to ensure fraud controls are effective, timely and not creating too many false positives that impact our resources. It would be easy to throw the kitchen sink at every decision point, but it wouldn’t make much commercial sense and could negatively impact genuine customers. It’s more about deploying the right tool at the right time of the journey to leverage best outcome.”

Sponsored content