Artificial intelligence is becoming commonplace in insurance. It can draft an email, summarise a meeting and save a few minutes here and there.

Useful, certainly. But insurance has bigger problems to solve.

Claims are increasingly complex. Files contain growing volumes of documents, data and digital evidence. Experienced people are difficult to replace, and valuable knowledge can leave with them. Fraud is also becoming more sophisticated: insurers detected £1.34 billion of fraudulent claims in 2025, up 14% in value on the previous year.

AI is contributing to that threat as well as offering part of the solution. The technology can help insurers find connections and inconsistencies across huge amounts of information, but it can also help fraudsters create increasingly convincing documents, images and identities.

The challenge is therefore not simply doing the same work faster. It is applying specialist knowledge more consistently, spotting risks earlier and helping people make better decisions.

Start with the problem

At HF, we handle hundreds of thousands of claims with our insurer clients. Much of that work still involves people reading substantial volumes of evidence, identifying what matters and applying experience built up over many years.

That expertise is invaluable, but difficult to scale.

It often sits in the heads of experienced lawyers, investigators and claims handlers. When those people are busy, their knowledge is not always available at the moment it is needed. When they move roles or leave the industry, some of it can disappear altogether.

At the same time, newer colleagues are expected to develop judgement while dealing with increasingly complicated claims. Training helps, but there is a limit to how much knowledge can be captured in a manual or transferred in a classroom.

Insurance is not short of information. The harder problem is getting the right knowledge to the right person at the right point in a claim.

That was the problem behind ELI, HF’s Expert Legal Intelligence platform.

ELI is not a general chatbot or a claims-branded version of a large language model. It contains multiple specialist skills designed around specific legal and insurance tasks.

These include document analysis, timelines, inconsistency checks, liability, damages, fraud review, financial disclosure, intelligence, injury recognition, plausibility and costs. Several skills can work together to provide a joined-up view of a claim rather than dealing with one document or question in isolation.

Each skill starts with a real operational problem. HF’s lawyers, investigators and claims specialists help define the evidence to consider, the questions to address and what a useful output should contain. Our technical teams then turn that knowledge into a structured AI process.

That is an important difference.

A general AI assistant waits for the user to ask the right question. In a specialist claims environment, the person may not know what to ask, particularly if they are less experienced. Even experienced people can approach the same problem differently.

With purpose-built AI, the expertise sits within the design of the task. The user should not need a qualification in prompt engineering before starting their actual job.

Making expertise easier to access

This approach can help address one of the industry’s most difficult workforce challenges.

The objective is not to replace experienced professionals. Their judgement, curiosity and understanding of context remain essential. Instead, it is about capturing elements of their expertise and making that knowledge available more consistently across a wider team.

For a newer handler, that could mean receiving a structured view of the evidence and the issues requiring attention. For an experienced professional, it could mean avoiding hours of manual document review and spending more time on strategy, negotiation or the genuinely difficult parts of a claim.

It can also improve consistency. People will not always reach identical conclusions, nor should they. Claims work frequently involves judgement. But technology can help ensure that the important evidence has been considered, the relevant questions have been asked and the reasoning is clearly recorded.

The value lies not in allowing AI to make the decision, but in helping the professional reach that decision with better information.

From experiments to live claims

ELI has moved well beyond proof of concept. Its specialist skills have now been used more than 100,000 times across HF and our insurer clients, with usage continuing to grow rapidly.

One skill alone has assessed nearly 20,000 credit-hire payment packs worth around £170 million, reducing an initial review from about an hour to roughly three minutes.

It is a powerful example, but only one.

The wider platform is being used across different departments, disciplines and insurer clients. Some skills help users understand large files and construct clear timelines. Others test evidence, identify inconsistencies or support specialist reviews in areas such as fraud, liability, damages and financial disclosure.

This breadth matters because claims rarely fit neatly into one category. A single case may involve medical evidence, financial records, questions of causation, conflicting accounts and potential fraud indicators. The opportunity is not simply to automate one task. It is to bring different forms of specialist support together across the claims journey.

That is when AI begins to move beyond isolated demonstrations and into everyday operations.

The test is not whether the technology looks impressive on a stage. Most AI does when it is given the perfect document and a carefully rehearsed question. Real claims are less considerate. The key fact may be buried in hundreds of pages, contradicted elsewhere and written in handwriting that appears to pre-date the invention of the pen.

The real test is whether AI can operate repeatedly on messy, complicated claims and produce something a professional can check, understand and use.

Keep the human in charge

AI can produce an answer that is clear, confident and wrong. Confidence has never guaranteed accuracy, in technology or elsewhere.

That risk is particularly important in claims, where an output may influence the treatment of a customer, the direction of an investigation or the value of a settlement.

ELI therefore supports decisions rather than making them. Its outputs require professional review, client information is not used to train the underlying models, and each skill has a defined purpose and boundary.

We have also worked with the University of Salford to test areas including privacy, security, transparency, explainability, bias and fairness. Governance was not added after development; it helped shape the platform from the beginning.

That does not mean the technology will always be perfect. No serious AI strategy should make that claim. It means its limitations must be understood, its outputs must be capable of challenge and responsibility must remain with the person making the decision.

Organisations should also measure whether AI is solving the original problem. Usage alone is not enough. The meaningful questions are whether it improves consistency, releases capacity, helps identify risk, supports better decisions and produces better outcomes.

General AI assistants will become part of everyday working life, just as email, search engines and smartphones did before them. They will make many routine tasks easier.

But the bigger opportunity for insurance lies beyond the chatbot. It lies in combining AI with an organisation’s own knowledge, evidence and processes to solve problems that a general assistant was never designed to address.

That is where AI stops being a novelty and starts becoming genuinely useful.

Learn More About HighFive and ELI

Visit HighFive or contact David Scott, Partner & Head of Innovation and Client Solutions, at david.scott@h-f.co.uk, to discuss how innovation and expert insight can help your organisation tackle real-world claims challenges.