Kerry Williams, AEGIS London Head of Claims 

Claims has long been an area where technology can improve efficiency and accelerate processes. As AI develops across the insurance value chain, it creates a particular opportunity in claims: not only to streamline processes, but to make better use of the insight claims teams hold and strengthen how we deliver for policyholders.

I am genuinely excited by the benefits AI can bring to claims. But the more I consider its application to claims, the more we need to recognise the risks, particularly if we allow its capabilities to run ahead of our ability to challenge and control them.

Turning claims data into insight

One of the most exciting applications of AI in claims is the opportunity to unlock the wealth of information already held within claims departments, turning fragmented data into meaningful insight that can support the wider business. Done properly, that could provide underwriters with much richer insight into what is happening across their portfolios and ultimately support better risk decisions.

The challenge is getting the information into a form that allows us to do it. Claims data is often fragmented, particularly across the Lloyd's market, and information may exist across different systems, documents and individual files.

An experienced adjuster working on a particular book develops a feel for what is moving claims and which behaviours are influencing outcomes. At scale, however, it becomes much harder for any individual to develop that same picture across an entire portfolio.

This is where AI becomes particularly interesting. If we can structure and ingest claims information properly, the technology could potentially take the analysis to much greater levels of granularity, whether that means understanding what is happening in particular jurisdictions, which factors are influencing outcomes or where patterns and behaviours are beginning to change. Used well, that insight could help claims teams and the wider business get ahead of emerging trends rather than simply responding once they are established.

That could be immensely valuable, but there are important foundations to get right before investing in increasingly sophisticated technology. Insurers need to ensure the information AI will interrogate is properly structured and of sufficient quality, while also being clear about ownership, access and where that data is going when it is used by AI tools.

AI should assist, not decide

My enthusiasm for AI is also tempered with a healthy dose of caution. 

I think of AI as an assistant. I am very comfortable with technology gathering information, organising it and presenting it in a more digestible form. If AI can take a ten-page document, structure the information clearly and help an adjuster understand where to focus their attention, that is a valuable use of the technology. I am far less comfortable with it making the claims decision.

Claims are not simply exercises in processing information. They can involve complex policy interpretation, conflicting legal arguments and, importantly, real people experiencing extremely difficult circumstances. If you or a member of your family were seriously injured in an accident, would you want an AI agent determining the outcome of the claim, or would you expect a person to be engaged in that decision?

Throughout the process, there needs to be a claims professional capable of understanding the policy, interrogating the information and applying judgement to the individual circumstances.

Guardrails must keep pace with adoption

There is also a related issue the market needs to consider carefully: regulation. AI is advancing at considerable speed, but we do not know what the litigation and regulatory environment surrounding its use will ultimately look like.

That creates a potentially significant exposure if adoption moves too quickly into decision-making. A claims decision supported or made by AI today may be scrutinised several years from now, when expectations around governance, disclosure and accountability could look very different. We are already seeing AI records being requested in investigations, so insurers need to consider whether they could defend the role technology played in reaching a particular outcome.

That does not mean the market should be frightened of AI or wait for every regulatory question to be resolved. However, it does mean we need appropriate guardrails around where automation ends and human responsibility begins.

Expertise remains the most important safeguard

Perhaps the most important guardrail is expertise, because keeping a person involved only provides meaningful protection if they have the knowledge and experience to challenge what the technology is telling them.

When I was training, you had to be able to read and interpret the policy. If a lawyer presented an interpretation that did not make sense, you needed the knowledge and confidence to return to the wording, interrogate it and form your own judgement. We cannot afford to lose that capability as AI becomes more sophisticated.

If the next generation of adjusters becomes accustomed to AI providing both the analysis and the answer, the concern is that we gradually erode the expertise people need to recognise when the technology has reached the wrong conclusion.

This is why I do not see AI reducing the importance of claims talent, if anything it makes developing that expertise more important. The technology can process more information than any individual ever could, but the person using it still needs to understand what they are looking at, recognise when something is wrong and have the confidence to go back to basics.

There is a great deal that AI can do for claims, from making fragmented data more useful to identifying portfolio trends and removing time-consuming administrative work, and we should take advantage of those capabilities. But we should also be clear about where the boundary lies.

AI can gather the information, organise it and help us interrogate it at a scale, but the expertise to challenge what it produces, the judgement applied to the individual claim, and responsibility for the ultimate decision should remain with the claims professional.

Ends

This article was published in Insurance Day on 19 October 2026.