What Freddie Scarratt Told Me About Where the Money Went
In the second quarter of this year, ninety-five percent of new insurtech funding went into companies built around AI. Freddie Scarratt had that number on the podcast in July. He's the Global Deputy Head of InsurTech at Gallagher Re, so he watches where the money actually flows for a living. Ninety-five percent. Not a majority. Almost all of it.
That statistic isn't really about AI. It's about what happens next, and Freddie's real point was sharper than the funding number. Everybody is building the AI. Almost nobody is insuring it yet.
A Chatbot in British Columbia Just Wrote the First Page
Here's the case Freddie kept coming back to. Air Canada's website chatbot told a passenger he could apply for a bereavement discount after his flight, when the airline's actual policy required the request before travel. The passenger believed the bot, booked the flight, and later filed a claim for the difference.
Air Canada's defense in front of the tribunal was that the chatbot was, in the company's own words, a separate entity responsible for its own answers. The tribunal called that a remarkable submission and ruled against the airline anyway. The award came to about eight hundred Canadian dollars, small enough that most of the trade press covered it as a curiosity. It wasn't a curiosity. It closed a loophole every company with a chatbot was quietly counting on. You cannot point at your own AI and say it isn't you.
The Product Nobody Has Fully Built Yet
Once a chatbot's bad answer is legally the company's bad answer, the question stops being philosophical and becomes actuarial. Who covers that. Freddie's read, and it's the part of the conversation I keep turning over, is that AI liability probably isn't going to show up as some new standalone product with its own name on it. It's going to get folded into cyber and professional indemnity until the three of them behave like one category, because the exposure doesn't respect the old boundaries. A hallucinated policy quote looks like an E and O claim. A leaked training set looks like a cyber claim. A biased underwriting model can be both at once, plus a regulatory problem, depending on which state's insurance department reads about it first.
Carriers still selling these as three separate line items, underwritten by three separate teams, are going to see claims arrive in a shape none of the three teams was built to recognize.
Software Still Doesn't Repeal the Laws of Underwriting
I've made a version of this argument before, about the demo that always ends with the AI making the decision and the business that always ends with a licensed human who has to answer for it. Freddie's episode is the same law showing up on the other side of the transaction. It isn't only a carrier's own AI that creates exposure. Every policyholder is now running some AI that can say something binding on their behalf, whether or not anyone actually approved it to.
Insurance, almost alone among industries, can never fully cap its own downside. You can reinsure it, price it, syndicate it across a dozen markets, and somebody in the chain is still holding the tail risk when the model is confidently wrong at two in the morning and nobody's watching. That didn't change when the chatbots showed up. It just found a new place to live.
What I'd Tell an Underwriter This Quarter
Stop asking whether your book has AI exposure. Every book does now, the same way every book had cyber exposure years before most policyholders would have called it that. Ask instead whether your E and O, your cyber, and any AI-specific endorsement actually talk to each other, or whether a claim touching all three ends up in a coverage dispute before it ends up in a payout.
Ninety-five percent of the money chasing insurtech this year went toward building the thing. Somebody still has to write the policy for when it's wrong. A chatbot in Vancouver just showed the industry what that claim looks like when it finally lands, and it landed for about eight hundred dollars. The next one won't be that cheap.