AI’s hidden cost could become operational risk for re/insurers
The growing integration of AI into insurance operations could expose companies to a new form of financial risk if the cost of running AI models rises substantially, according to Owen Williams, group chief information officer at Ascot Group.
Williams was speaking on a panel at ILS Bermuda Convergence, which also included Ben Fox, co-founder & CIO of Members Capital Management and Jamie Rodney, CEO of Reask. It was moderated by Stephen Moss, head of investor analytics at Aeolus Capital Management.
Williams warned that re/insurers risk becoming dependent on the pricing structures of external AI providers as they redesign business processes around the technology.
While AI is already being deployed to improve individual functions such as underwriting and claims handling, Williams argued that more fundamental changes to how re/insurers operate could introduce risks that are not yet receiving sufficient attention.
“One of the things that I have a great concern about is the cost of inference,” he said, referring to the expense of running AI models.
“If you build your entire process around the assumption that the cost of the frontier models you’re going to use, whether OpenAI, Anthropic, or whichever, is going to be X dollars, you build your entire processes on that basis.”
He warned that companies could find themselves exposed to substantial increases once those processes become embedded.
“If those vendors come back to you a year or two years from now and say it’s a 40% or a 2x increase, it would be very, very hard for you to absorb that because it has big implications for the organisation.”
Williams pointed to growing interest in open-weight models and other technologies as potential ways to mitigate that exposure: “It is a risk exposure that you have in the institution.”
The issue also emerged in the discussion of whether AI investment is delivering measurable commercial returns.
Rodney said Reask had experienced considerable gains in coding and data integration but acknowledged that translating those efficiencies into broader business performance remained challenging.
Reask has integrated AI-enabled tools across its operations, including customer relationship management, but faster access to information has not necessarily accelerated commercial processes.
“Integrations haven’t accelerated the sales cycle and haven’t accelerated us getting red lines back from legal teams any quicker,” Rodney said.
He noted that the relatively low cost of experimenting with AI currently makes investment easier to justify, even where financial benefits remain uncertain.
“The cost is also still quite small right now. We’re not blowing the bank up by integrating that. If the cost changes, then that’s a worry. Actually, that ROI equation becomes even more negative.”
Williams said the clearest financial benefits currently appeared in operational expenses, particularly through automating document-intensive activities such as submission ingestion and claims processing, rather than demonstrable improvements in underwriting profitability.
“I don’t think as of now, I’m not seeing any true cases where people say, ‘I'm really shifting my overall loss ratio by X points because of AI.’”
For re/insurers investing in AI-enabled operations, the discussion raised a question that extends beyond immediate efficiency gains: whether the economics supporting those gains will remain attractive once the technology becomes indispensable.
Did you get value from this story? Sign up to our free newsletters and get stories like this sent straight to your inbox.
