Current approach to AI ‘not generating a lot of value’
The following article was originally published on Insurance Day on 11 March, 2026.
‘I’m a big cynic of tech people coming to insurance, having a great idea…and burning through those investment dollars and not actually solving a problem,’ Martin Henley says
The insurance sector is approaching artificial intelligence in an inefficient and sometimes misguided way, says the chief executive of insurtech mea Platform.
Martin Henley, who founded mea in 2021, says insurers and brokers are spending vast sums trying to optimise routine processes which cannot provide much of a competitive advantage over their rivals.
“That’s probably not generating a lot of value for the industry or the consumers”, while raising their insurance costs, Henley says.
Insurance firms “probably shouldn’t be competing here, which is why we try to bring to market something that works”. Indeed, mea’s tagline is: “It just works”.
Henley was previously chief information officer at Axa between 2010 and 2013, and XL Catlin between 2015 and 2019. He looked to veteran underwriters and insurance operations leaders to staff mea — “So people have really lived and breathed insurance,” he says. Former XL Group chief executive Mike McGavick is the firm’s chairman.
Agentic Offering
Mea’s core offering is a “domain-specific language model” (DSLM) that mea has built on insurance-related data, along with an “insurance specific knowledge graph”.
The DSLM is built on a large database of insurance-specific information and prompts, Henley explains. Mea spent years “building huge lists, by line of business, of fields or questions and answers that they want to be able to underwrite”. By training the DSLM on this model, mea’s software can be adopted by insurers with a minimum of modification.
Meanwhile, the knowledge graph is a “modular platform” that works with both mea’s AI and public LLMs, which allows insurers to organise and connect data in ways that are relevant to managing insurance policies.
The company works with a wide array of insurance-sector clients, from major insurers and leading brokers to delegated underwriters. Mea’s website lists several large insurance concerns as customers, including Ardonagh, Axis, Markel, Lloyd’s and Scor.
Because mea’s DSLM is designed for insurance, clients can start using it quickly. Frequently, Henley says, customers have come to mea saying they had spent many months trying to get a more generalised LLM to work. Mea’s model “gets clients to accurate answers an order of magnitude faster”, which ultimately saves them money. The insurtech firm claims its products can reduce combined ratios by between 0.5% and 3% and expand underwriting capacity by an average of 40%.
Each client can customise its DSLM as needed, so these evolve with the client while feeding back data and experiences to the parent model at mea. From last October mea has also offered agents to handle routine operations activities for underwriting, broking, claims and finance professionals.
Mea’s agentic AI model is in full commercial use. However, although AI in general is widely used in the insurance sector, Henley says most insurers have not deployed agentic AI beyond the prototype stage.
Agentic AI implies an AI system that can act, to at least some degree, autonomously. Henley stresses that mea’s clients can tailor the amount of autonomy they give their mea AI systems.
Some insurance functions and decisions can be fully automated, especially in simpler retail lines, Henley says. Wholesale and specialty insurers may make somewhat less use of autonomous AI, but even there, Henley suggests 30%-70% of their activity today could be replaced with a machine, allowing greater emphasis on other business development activities.
Although insurers and brokers are unlikely to allow agentic AI to make big decisions on its own, Henley does expect AI may provide human operators with valuable advice.
Henley backs the approach of UK regulators towards artificial intelligence, a “principles-based” approach which offers broad guidelines and eschews a single artificial intelligence law, as in the EU. Henley says regulators should focus less on the internal operations of AI systems and more on the consequences of AI use.
Capital Infusion
In February, mea announced its first infusion of external capital, a $50m minority stake from SEP (formerly Scottish Equity Partners). At the time, SEP managing partner Angus Conroy described mea as “an excellent fit with our strategy of backing IP-rich technology companies that solve complex problems for the world’s largest organisations. In a dynamic market, mea stands out for what is live, proven, and scaled today,” Conroy added.
In a report published earlier this month, Gallagher Re tallied slightly more than $5bn in insurtech funding over 2025, a slight increase over 2023 and 2024 figures. Gallagher Re said nearly two-thirds of this funding went to firms focused on AI, although this reflected the tendency of many firms to rebrand as AI-focused enterprises, the firm said.
Henley believes many tech firms and tech-focused investors and funds have made ill-advised bets on insurtechs. “I’m a fairly big cynic of tech people coming to insurance and having a great idea and getting lots of investment dollars, burning through those investment dollars and not actually solving a problem,” he says.
Prior to SEP’s arrival, Henley says the company was “mainly funded by me convincing an extended number of people to work for free for an extended period of time”.
Henley says mea will use the capital “to continue the innovation at an even faster pace”. Mea will also expand sales operations and invest in marketing, to which mea has so far devoted little attention. “We feel we’ve got some quite compelling different views on [marketing in] the industry”,Henley says.
He also laments the inefficient ways insurers invest in technology, pointing to the “huge [amount of] wasted money”, which produces little of use to anyone aside from consultants and software providers.
Read the originally published version in Insurance Day: https://www.insuranceday.com/ID1155617/Current-approach-to-AI-not-generating-a-lot-of-value
