Outcomes

From Expense Ratio to Loss Ratio: Where Next for AI in Specialty Insurance?

LMF’s latest Underwriting Special Roundtable explored how the Market can move from faster processes towards better underwriting, claims and portfolio decisions, assisted by investment in and deployment of Ai technologies.

AI is already helping insurance firms reduce manual work and improve the flow of information. But how far has that investment translated into better underwriting performance, and what needs to happen before it can?

That question brought London Market practitioners together at The Wagtail on 6 October for LMF’s Community Roundtable Breakfast, supported by Portage AI. Facilitated by Roger Oldham, Founder & CEO of LMF, the session featured John Hele, Executive Chair of Portage AI, alongside live polling, audience questions and an open discussion of the Market’s progress.

The conversation explored a growing ambition: to use AI not only to lower the cost of doing business, but also to improve the decisions that shape claims outcomes and underwriting profitability.

Efficiency gains are emerging, but progress varies

The opening polling and discussion highlighted organisations at different stages of adoption, with many still experimenting or developing specific applications.

Practitioners described benefits in areas such as extracting information from documents, reducing rekeying and moving data into connected systems. These improvements can give underwriting teams more capacity and reduce the administrative work surrounding a submission.

However, the discussion also drew an important distinction between processing a risk more efficiently and understanding whether it is a good risk to write. The first is delivering practical benefits in parts of the Market. The second remains a developing opportunity.

Participants also raised the challenge of measuring results. In a changing insurance market, separating the contribution of AI from wider movements in pricing and business performance is not straightforward. Demonstrating an impact on the loss ratio will require careful evaluation over time.

Making historical data usable

A central theme of John’s presentation was that the Market already holds valuable information across its underwriting, pricing, claims and delegated authority records. Much of it remains difficult to bring together and analyse.

Information arrives in spreadsheets, PDFs, contracts and other documents, with different organisations using different descriptions for similar concepts. Historical systems may also contain only part of the detail available in the original source material.

John argued that some of the most useful insights could sit within the information that has previously been too difficult or expensive to extract.

His proposed starting point was a common data language: a consistent way of identifying and mapping information from different sources. Bringing that information into a usable structure could help firms connect the risks they accepted, the prices they charged and the claims that followed.

The emphasis was on creating data that people can trust, trace and use repeatedly.

Choosing the right tool for the task

John encouraged firms to be specific about what they need an AI application to do, distinguishing between tools used to interpret information and those used for repeatable extraction and processing.

In the approach he outlined, language models can help interpret differing terminology and explore patterns, while supervised machine learning can support structured extraction. The wider point was that model selection, validation, and controls should reflect the task being undertaken.

An audience example involving inconsistent results when extracting claims information reinforced the importance of checking outputs against source documents.

For underwriting and claims teams, a convincing answer alone is insufficient. They need confidence in the information behind it, visibility of exceptions, and a clear process for human review.

From data extraction to underwriting insight

Once historical information is usable, the opportunity becomes broader than just administrative efficiency.

John discussed the potential to identify relationships across policies and claims, test assumptions against actual experience, and give practitioners a richer view of their own portfolios. Audience questions also explored how better data could support product development and reveal opportunities for new or more tailored cover.

Experienced underwriters and claims professionals have an important role in this process. Their knowledge can help frame useful questions, challenge unexpected findings, and determine whether a pattern has commercial significance.

The session also acknowledged the maturity gap. John said he had seen organisations working on parts of this journey but had not yet encountered a client advanced across the full approach he described. The opportunity was presented as a direction for development, with substantial work still to do.

Trust also depends on resilience and people

The roundtable considered what boards, Lloyd’s and regulators would need to see before relying on AI-extracted information for material insurance decisions.

Participants highlighted traceability, auditability and the ability to understand how an output was reached. They also raised operational resilience: what happens when a system fails, and whether firms retain the expertise and processes needed to continue operating.

Adoption brought a further set of considerations. Practitioners discussed the value of involving colleagues early, demonstrating useful results and helping people understand how a tool supports their work.

Benefits may include greater capacity, less repetitive administration, and a more satisfying or enriched working environment. Those improvements can help build confidence and engagement, alongside the financial measures used to assess an investment.

Start with a clear business problem

A practical recommendation running through the session was to begin with a defined use case or line of business, deliver it properly and build from there.

That means understanding the problem, preparing the relevant information, involving the people who will use the output and agreeing how success will be measured. Integration with existing systems and ownership within the business also matters if a pilot is to evolve into part of everyday operations.

The discussion continued around the table

The morning concluded with an interactive roundtable discussion, giving practitioners the opportunity to move beyond the presentation and compare their own experiences of AI adoption across underwriting, claims, data and operations.

The discussion centred on three questions: after several years of AI investment, where are firms seeing the greatest benefit — in the expense ratio, the loss ratio, or neither yet; if data were clean and usable today, where could it create the most value; and what would boards, Lloyd’s and regulators need to see before relying on AI-extracted data to support material insurance decisions.

Across the tables, data quality and readiness emerged as a recurring theme. Participants recognised that organisations may possess significant volumes of valuable information, but its usefulness depends on how accurately it can be extracted, structured, understood and connected. The ability to understand existing processes and data flows was therefore seen as an important foundation for more sophisticated AI applications.

Trust was another major talking point. Practitioners discussed the importance of understanding the limitations of AI, including the potential for hallucinations or inconsistent outputs, and ensuring appropriate governance keeps pace with technological development. Clear parameters, validation processes and human oversight were considered essential where outputs begin to influence more significant insurance decisions.

There was also discussion around accountability. For AI to become more deeply embedded within underwriting and claims processes, practitioners need confidence not only in the answer being produced, but also in the information behind it and the process used to reach it. The ability to trace outputs back to underlying information and challenge them where necessary remains an important part of building that confidence.

The role of domain expertise featured strongly. Rather than seeing AI as a replacement for experienced insurance professionals, the discussion highlighted the importance of combining technology with the judgement of underwriters, claims specialists and other practitioners. Those professionals remain critical in asking the right questions, identifying when an output does not look right and determining whether an insight has genuine commercial relevance.

Participants also explored the potential for AI to move further upstream in the insurance value chain. While many current applications focus on administrative efficiency, the greater opportunity may come when trusted data and technology begin helping firms make stronger decisions around risk selection, pricing, portfolio management, claims and the development of solutions for clients.

For that transition to happen, confidence will need to grow across organisations. Better understanding of how the technology works, clearer governance and regulation, stronger data foundations and practical evidence of successful outcomes were all discussed as factors that could encourage greater adoption.

The roundtable ultimately reinforced that improving the loss ratio through AI will not depend on technology alone. It will require a combination of accurate data, appropriate governance, human expertise and a willingness to challenge and validate what the technology produces.

The breakfast left the Market with a useful challenge: to connect AI investment more closely with the decisions that determine insurance performance. Efficiency remains valuable, and the next stage will depend on turning trusted information into insight that practitioners can act on.

Our thanks to John Hele and the Portage AI team, a valued LMF Corporate Member, for supporting the session, and to all the practitioners who contributed their questions, experience and perspectives.

Continue the conversation with LMF. Explore our upcoming events and join fellow London Market professionals to learn, share insight and collaborate.

If you’re not already part of this interactive community, get in touch to find out how you can join the LMF conversation shaping the future of the London Market.