Rory Yates is Chief Strategy Officer at EIS, a global core technology platform provider for the insurance sector. In this piece, Rory takes a look at how fraud is evolving and puts the case that although AI can be used to commit fraud, it can also be one of the best fraud detection tools available to insurance brands.
Digital transformation has led to ever-more sophisticated fraud methods like synthetic identities, deep fakes, and scaled attacks. These are also escalating rapidly due to changes in economic activities and general consumer behavior, including opportunistic fraud. Traditional fraud detection systems face challenges in scalability, data balance, require significant manual input, and are struggling to adapt to new fraud patterns.
Generative AI can theoretically revolutionize fraud detection with its adaptive learning abilities, capability to handle large data sets, improved anomaly detection, and reduction in false positives. Combining Generative AI with Machine Learning, enhances adaptability, fraud pattern analysis, and contextual understanding in fraud detection.
More generally, though, fraud detection, even that generated by AI, requires most of the same principles – access to high-quality external and internal data, collaboration across the industry including fraud bureaus, and a seat at the top table of insurance to get the priority and investment it needs.
The Looming Threat
Speaking to industry experts on this topic regularly, it’s almost certainly the case that insurers are laggards to the protagonists. We detect about a third of fraud in insurance, so it’s easy to argue they’re always ahead. Despite getting more of the attention it needs, there’s some real concerns about how to keep up in this next wave, and for good reason.
George Santayana once said that “Those who cannot learn from history are doomed to repeat it”. This history lesson then, if nothing else, shows that where there’s insurance, there’ll be fraud! In 1862, Mr Calvert attempted to defraud underwriters at Lloyd’s when he insured an oil cargo ship called the Poseidon for £12,000. He alleged that pirates had boarded and set fire to the ship, but investigations found that Mr Calvert forged documents and that the Poseidon never existed.
Fraudsters, as I now know to call them, are always looking for weak spots, and unlike our mental image of them, a lot are highly professionalised, organised, and now digitised. They’re intelligent and make sure that they are clothed from detection. They use different fraud methods, different insurers (sometimes simultaneously), and fake identities, just to make sure that they don’t get caught or to obscure the other profiles. You might capture the “faked” individual, but the real person continues their vocational pursuit anonymously to insurers.
And they have always found a way to leverage new technologies that are not yet fraud-proof, and with more coming online, they’ll make use of those as well. Fraudsters move fast, and most companies can only follow their trends.
A good friend and industry lead in the Fraud Avoidance market, Matt Gilham of Whitelk Consulting, made the point recently at an Insurance Innovators event when commenting on some of the presentations. For the fraudster, he said, the cost and barriers of getting to these new technologies are vastly lower than it is for the insurer, who is having to act on this at scale and within a much wider organization. This raises the question, how can you write a business case for something that you can’t necessarily quantify and keep up with them?
Where Do We Go from Here?
The truth is that business casing fraud prevention ahead of quantifiable evidence of something is hard. It’s a bizarre paradox. We’re all very confident across the industry that we are receiving GenAI-based fake images in claims and so forth. The issue is, if we can’t detect them, how can we quantify this and move forward? The short answer is, we can’t!
The reality is that the same luxuries we place on pioneering technologies in customer experience and digitisation across insurers should include and be matched in efforts to meet ongoing consumer fairness, responsible development, and the prevention of risks. This therefore should firmly include fraud.
Yet many still tell me that while they’ve made huge progress in the last ten years, in particular, to embed fraud transformation into the heart of insurance, and raise awareness that this operates full lifecycle, many still feel that on this topic they’re constrained, undertrained, under-prepared, and not in the same R&D cycles as other aspects of their businesses.
So, fraud needs to be front and centre of AI discussions and efforts. There’s no doubt it’s over-hyped at this stage, but there’s also no doubt it changes everything going forward. Those in the highest and most advanced learning cycles are therefore the ones most likely to realise the most value and avoid the most risk.
The work of Stop Scams is a good example of how this is happening at an industry level. In 2023, they organsied a summit meeting hosted at the Bank of England by the Govenor Andrew Bailey, bringing together senior policy makers and representatives of industry. Their ambition is to move beyond the headlines and hyperbole and look at the real-world implications of fraud and AI.
PwC produced a report with Stop Scams UK looking at the impact of AI on fraud and scams and among a range of conclusions, technology companies involved in the research were clear that AI is undoubtedly being used to generate fake content and create malicious user profiles. Across participants, it was suggested that it’s only a matter of time before fraudsters adopt it at scale. While there were a range of views on the immediacy, the closest to the development of AI tended to talk about threats in timescales of weeks and months.
The time for investment in learning is now. The time for exploration and experimentation is now. The adoption of advanced detection capabilities is essential now to any insurance ecosystem. Machines learn rapidly from patterning with machine learning techniques. Adapting to changes quickly is vital. There’s a need to look at wider methods around claims fraud such as dashcams, car telemetry, mobile technologies, and so forth, all providing ways to validate claims and general data at various stages of the insurance lifecycle as well.
It’s also clear that ambitious insurers will pursue more advanced fraud strategies, and that this will be vital to viability and fairness.
Be the first to comment