Spotlight: Technology: Intelligent fraud detection

From using predictive analytics to identify risks at the point of quote and minimise the potential threat to the book before they are incepted; to using data to raise red flags when a claim is made, and employing a digital workforce to automate the entire process, companies are addressing the challenge of fraud in a variety of ways.

“We recognise that fraud detection is an evolving journey and technology has become key to that. Our aim is to ensure that prevention and detection happens as early in the journey as possible, ideally at the point of quote and to support this we have been using in-house data scientists,” says James Burge, fraud manager, Allianz Insurance.

“We are also looking to use machine learning to build models which will spot suspect claims. However our people are still key to investigations and we’ve only recently strengthened our application and claim frauds to ensure we have a greater view of end-to-end customer journey.

“We are working with the industry to share intelligence where possible and data is key to this. The challenge as always is how quickly fraudsters can change tack, particularly given the current Covid-19 situation, as these organised criminals are not just sitting waiting for the world to go back to normal, they are looking for other ways to commit fraud.”

The insurance industry possesses huge volumes of data that can be used in the fight against fraud, although extracting useful and relevant information to use against fraudsters is a challenge in itself.

Speeding up the process

Intelligent automation is being used by some insurers to speed up the process of intelligence gathering at the point of quote. A digital worker can log into a variety of systems to query data and return it to validate the customer information provided for a quote. Because it’s able to work at machine speeds, it doesn’t slow down the quotation process.

This is particularly useful for retail insurance companies and brokers that sell through aggregator sites, where users may omit key information about previous claims or penalties. It ensures that quotes are accurate and there are fewer mid-term adjustments and issues at claim as a result of faulty data.

“When you’re investigating fraud 90% of what you’re investigating is irrelevant,” says Scott Clayton, head of claims fraud at Zurich Insurance. “What you want is information that’s relevant to the consideration of a claim. We have a provider that gives us specific information in the context of a personal injury claim. It is very focused information and intelligence that it has found on the internet that assists us with either the validation or investigation of a claim.

“We’re a digital world so there’s increasing emphasis on one-touch claims handling and app-based notifications. Therefore, you have to have a lot of this technology behind the scenes.”

Method, techniques and technologies being used by insurers in the fight against fraud include robotic process automation, artificial intelligence and machine learning. However, the industry is still in a relatively early test-and-learn phase of using advanced technology to fight fraud effectively.

RPA and intelligent automation platforms help to maximise the use of the data that is available. Fraud prevention and investigation processes can be time consuming when performed by a person, but take only minutes for a digital worker to complete. Even with other technologies in play, fraud prevention measures may not be efficient in terms of time and resource so it’s important to ensure that all these technical elements are considered as part of a digital fraud strategy.

Michael Müssig, partner at McKinsey, says: “Everyone is using some kind of technology but we don’t see many cases where an insurer has fully integrated digital models to detect and prevent fraud at scale. There also isn’t a one-size-fits-all solution that will work best in the future.

“The industry has moved beyond rules-based models to ML involving network analysis and so on. But challenges remain. Probably the biggest challenge slowing down insurers today is the limited quality and availability of usable data, and the ability to ingest and analyse it.

“ML models are designed to help insurers detect fraud – in the best case before fraud even happens. But it’s not just about the model, it’s also about the type and quality of data you have available, as well as how effectively you connect that and draw insights from it. The input will affect the output that can be produced, which in turn will affect how you can actually use that output to better detect fraud.”

In order to utilise the capabilities of ML models insurers must connect the various sources of data and extract value from unstructured or semi-structured data (for example pdfs and voice recordings) to make it useful for analytics. This process is crucial if insurers are to implement new technologies effectively and achieve scale.

Technologies such as AI-native fraud detection and claims automation solutions, combined with advanced data science to spot suspicious behaviour in the insurance claims process, are integral to this process. Together with a proprietary linking technology they have the ability to resolve, manage and match information to create one consolidated view of the customer and paint the ‘big picture’ of how individual claims may be connected.

“The insurance industry is a massive consumer of data and technology and that holds the key to trying to combat all the frauds and different strands of fraud that materialise over time,” says Martyn Mathews, senior director, personal lines, UK and Ireland at Lexis Nexis Risk Solutions. “The industry works really hard, it strives to share data, which wasn’t always the case, and the use of data has grown exponentially.

“Being able to do innovative, smart data manipulation, aggregation and understanding is what holds the key to being able to validate numerous factors at the point of quote. For motor that means being able to process not just the insured but any named drivers in a response time to enable the quotes to go through an aggregator’s channel.”

“We are looking at cases of false identity, stolen identity and ghost-broking, where individuals are taking out multiple policies in their own or fictitious names. Data sharing often holds the key to combat this.”

Defeating fraud

Ultimately, insurers have to do more than detect fraud, they need the evidence and means to defeat it. The industry is beginning to operationalise data from the connected car as cameras and telematics data become more prevalent, which can be very effective. However, in other lines of business the same data sources are not always available. Together RPA, AI and ML are playing an increasingly important role in the efforts to fight fraud, but it will not fully replace manual detection.

Clayton says: “We still rely on people to detect fraud, their intuition and gut instinct, and what doesn’t look or feel right. That’s always been the most powerful weapon in the fight against fraud. But, where you have some sophisticated technology to help them out that’s where you can start bridging the gaps.”

Sponsored