AI detection and intelligent automation is redefining fraud detection in insurance and changing how insurers think about the tools they use. Andrew Pollard, insurance specialist at SAS UK & Ireland, looks at why the previously cautious insurance industry is now finally embracing change and reshaping how it detects fraud.
The insurance sector is facing a rapidly evolving fraud landscape. From deepfakes to AI-enhanced image manipulation and synthetic identities, the level of sophistication in fraudulent activity is growing – and so too is the financial and reputational risk to insurers.
Despite this, the vast majority of firms are still relying on legacy systems to catch fraud. According to a joint study from SAS and the Association of Certified Fraud Examiners (ACFE), 60% of insurers are still operating with rules-based detection systems. Another 40% use anomaly detection, which refers to the process of identifying data points or patterns that deviate from expected or normal behaviour within a dataset.
Just 20% of insurers are currently leveraging AI or machine learning as part of their core fraud strategies. But this figure is now slowly starting to shift and fraud is proving to be the main reason for driving AI adoption within the insurance sector.
A fast-growing threat
Fraud has become one of the fastest-growing areas of concern for insurers over the past four to five years. Organised crime groups are exploiting vulnerabilities, at policy level and within claims. We’ve seen everything from fake applications to digitally manipulated images and submissions.
With billions paid out in claims annually, reducing fraudulent payouts is a high-stakes priority for insurance companies. It’s also one of the few areas where AI can deliver measurable improvements quickly.
Cutting the cost of fraud directly improves profitability, and if you can do it without adding friction to the customer journey, it’s a win on every front. However, despite this, the insurance industry has been understandably cautious.
AI’s lack of transparency has already led to serious consequences in the sector. In the US, State Farm faced criticism and financial penalties after algorithms in its decision-making model were found to have a racial bias against certain policyholders. The company couldn’t fully explain the bias in its systems, but the case highlights the importance of having sufficient AI governance processes in place.
This kind of reputational damage doesn’t just impact one insurer, it ripples across the whole sector, and insurers in the UK and across Europe have decided to take a more controlled approach. They want greater transparency, auditability and governance.
As a result, there has been a slower rollout of AI across insurance operations, particularly where models interact with customer data or support underwriting and claims decisions. But fraud detection offers a middle ground: a space where AI can be deployed strategically, with clear oversight and immediate benefits.
Introducing Agentic AI technology
Agentic AI is a more autonomous, responsive breed of artificial intelligence that can adapt and evolve based on emerging fraud patterns. Rather than relying solely on fixed rules or static models, agentic AI can analyse complex data signals in real time, learn from new threats and flag unusual behaviours before any threats escalate further.
The fraudsters are already using AI so insurers just need to ensure they have systems that can keep up. For them, it’s about moving from a reactive stance to a proactive one.
SAS has been deploying these tools with major insurers using its latest Viya 4 platform and the results have been really positive. Insurers have seen the impact from day one with models continuously learning and improving, and within a year they have noticed a measurable reduction in fraud losses – which often amounts to millions.
Yalcin Terlemez, IT Division Manager at one of Turkey’s leading insurance companies, Aksigorta said having a platform for fraud management allowed for better data organisation and advanced analytics, which resulted in an increased fraud detection rate of 66%.
Then there’s the speed with which the investigations could be carried out.
“It used to take our investigators six months to expose cases of organised fraud,” Yalcin said. “SAS allows us to do it in 30 seconds. This level of speed and accuracy is something few insurers in the world possess.”
Given the understandable caution from insurers to embrace AI, any discussions about it must address regulation and its responsible use. The recently introduced EU AI Act and other emerging guidelines are pushing insurers to ensure their systems are transparent, explainable and compliant – therefore there is no reason why Agentic AI would mean unchecked automation.
All AI models need oversight and insurers should be governing their own models, asking questions about how they were trained and how often they are learning and being updated. This ensures that they remain valid over time and this kind of governance is what regulators want to see. It also ultimately gives insurers the confidence to act.
A turning point for insurers
While fraud detection is the main driving force for AI adoption in the insurance sector, and an area where it is so far making the fastest and biggest difference, the same technology has much broader uses – from underwriting to customer retention and pricing optimisation. Agentic AI has the potential to help reshape the entire value chain in the sector. Robust governance frameworks, transparent models, strong data foundations will all serve insurers well as they expand AI into other parts of the business.
Agentic AI offers a new playbook, away from static rules or spreadsheets that’s more flexible and fit for purpose in a digital-first world.
Insurers don’t need to take a giant leap of faith into the world of AI but there is evidence now that there’s a better, safer way to tackle fraud. One that protects customers, reduces costs, and builds trust in the long term.
For a sector built on risk management, this might just be the most compelling argument of all.
Be the first to comment