Guest blog from Tim Graham, Founder and CEO at Marrow
AI gathering momentum
18 months ago, I was trying to build Corgi. Or at least something quite like it, focusing on the UK micro-SME market. In trying to work out how to scale that business, I spent some time exploring if and how we could bring insurance products directly to the people asking about them in AI chatbots. Back then, the idea that small businesses would run a lot of their business ops in AI was imaginable, but maybe not mainstream. Given where things are today, it’s funny to think there was even any doubt – with a billion monthly active users on ChatGPT alone, consumers and businesses alike are adopting AI at scale, and particularly so for for high-impact, high-complexity products like insurance.
Back then, MCP was a just-launched framework, and the tools really didn’t exist to do anything too interesting for the mass-market – but what was plain as day was that out of the box, LLMs and agents would only get you so far. It was just too hard to bring products into AI models you didn’t directly control without exposing yourself to all manner of risk. I’ve always worked in insurtech, and this isn’t my first time launching insurance products – so I was very cognisant of that and the need for any solution here to be robust in a way that went far beyond what was being established for general e-commerce protocols, etc. Asking an AI for recommendations about what washing machine you should buy is great, and it’ll do a stand-up job of that. Asking it what insurance product is right for you is a very different beast – and one fraught with a lot of customer and regulatory risk.
At the time, all the headlines and funding announcements in fintech were about people building the agent itself – focusing on internal deployments and productivity gains. In my mind, there was a very conspicuous gap for the compliance-grade infrastructure that brought that efficiency and utility to the customer directly – with their own assistant of choice. Against that backdrop, the Marrow penny dropped – and we started building the rails that’d open up agentic commerce in the regulated world.
Market tailwinds
On 10 February this year, Money Group, MoneySuperMarket’s parent, fell 13% in a day to its lowest level in thirteen years, on the news that a US aggregator had put car insurance comparison inside ChatGPT. This wasn’t a UK product or a UK insurer, and yet the UK market tanked anyway – because everyone could see the shape of what was coming: the entry point to insurance moving from a search box to a conversation, with no safe way for a regulated product to join the dots.
Fast forward to more recent history, and in July the FCA published the Mills Review – its first proper look at what AI does to retail financial services. The recommendation that caught our eye most was (unsurprisingly) the one on agentic finance, where the regulator says the consumer-chosen agent model is the most consumer-favourable version of this future, and also the one that “could depend most heavily on the enabling infrastructure: data access, interoperability, identity, mandates, liability and audit”. Without those foundations, it warns, independent AI agents may remain partial, fragmented or unsafe. When that report landed, we gobbled it up – not least because we’d been making the same case since the start of 2026.
Meet Marrow
Marrow is exactly that infrastructure, for insurance. Customers map their existing pricing and policy systems to our schema once, in config, with next-to-no change required their side. On the other side of the coin, AI surfaces get to talk to one protocol rather than integrating fifteen insurers one at a time. In between, we do everything you need to to make sure that interaction’s happened in the right way – and that this can be demonstrated after the fact. In practice, this means juggling the AI assistant’s intelligence, which consumers want to benefit from, with the control required to make sure it’s happened safely. That’s technically not at all an easy problem to solve, and it’s a (massively) fast-moving set of goal posts, too. From a personal point of view, it’s the most challenging and bleeding-edge tech I’ve ever built. It’s a cliche, but these are hitherto unsolved problems – and we’ve now got the pleasure of great investors backing us in cracking that at scale.
What next?
The other thing that I have to say, to give an industry which is so often maligned for moving slow some credit, is that the market is increasingly moving from initial exploration to serious conversations about what operating as an insurance player in an agent-first world means. Some are taking that even further – actively putting the groundwork in place to win in an agent-led future, which shows really conviction and delivery muscle that is great to see. For us, that speaks to a set of insurers and brokers who are actually very willing to innovate, subject to very appropriate guardrails, and an acknowledgement that this isn’t going to be a situation where you come out on top by running a “wait and see” strategy.
If in reading this, you think any of what I’ve written above feels abstract or sci-fi, have a think about what you’d do if: something happened to your car and you weren’t sure if your insurance would cover it; you were buying a house and your mortgage lender said you needed life insurance; you got a quote that you didn’t think was a fair price. And these sorts of questions are really just the starting point – if someone’s willing to turn to their preferred AI assistant to help them understand insurance, quote it, and ultimately buy it – why would they not also expect it to handle policy changes, renewals, cancellations, even claims?
For a good chunk of people, we see that direction of travel being inexorable – and we’re laying the track that lets it happen safely.
If you want to take some no-regret first steps forward on agent-readiness, get in touch – we’d love to talk.
A note from Insurtech Gateway
Thank you for sharing your story, Tim. We are proud to support Marrow as it builds an AI distribution layer for insurance. At Insurtech Gateway, we are always excited to meet founders with strong insurance backgrounds, like Tim and his co-founder Adam. Having experienced the industry’s challenges first-hand, they are well placed to build products centred on customer needs. If you have identified a problem and know you could build a solution, we have the experience and tools to help you launch and scale faster. Check out our incubator and fund or get in touch.”