So I have three thoughts about this: The first thought is this is likely mostly survivorship bias. Yes, the claims that get to the humans are the ones that went wrong, and the mode of failure for LLMs is less that they get within 99% all of the time, it's more like 1% of the time they do something insanely dumb. At which point your customers are pissed off. That's kind of obvious, and it sucks for the humans who have to clear up the situation. But it's still a win for the vast majority of cases.

Secondly, does anyone actually know how much of your insurance premium is going on the call centre staff? most people don't claim, those that do mostly have simple claims. The actual cost for insurance companies is actually paying out claims, so even if you automated all of this it likely doesn't change your economics. Historically "New" Insurance companies only really succeed by mis-pricing risk and gaining market share that way - they often then fail when that risk materializes. Lemonade sounds a lot like that.

Finally, they haven't engaged with the tidal wave that's coming. Sure, AI agents handling claims is happening now. Just you wait. In a couple of years time it'll be AI agents making the claims for you and suddenly there'll be bots filing claims with infinite patience and a direct mandate to try and get as big a payout as possible. That is when it's going to get really hairy.