What Ken Bunn Told Me About Claims Departments
Ken Bunn spent eighteen years at Nationwide before Builders Mutual asked him to build a claims department that didn't exist yet. Twelve years later he's still building it. He came on the podcast at the end of July, and the conversation kept landing on one idea I haven't stopped turning over since. The AI tools that actually survive inside a claims department are the ones adjusters never have to think about.
That's not a line from a keynote. Rob Galbraith and I have both watched carriers roll out AI in claims a dozen different ways, and Ken has watched more of them die than either of us. Not from a bad model. From a busy Tuesday.
The Tool You Have to Remember Is the Tool You Don't Use
I wrote here in July that the copilot era was ending, that AI was moving from answering questions to doing the intake, triage, enrichment, and recommendation work before a human ever opens the file. Ken's point sharpens that argument in a way I hadn't said out loud yet. It is not enough for the agent to do the work. The adjuster can't have to go looking for it.
Think about how claims software actually gets used, not how it gets demoed. An adjuster at a mid-size carrier is running twenty or thirty open files at once. She is not opening a fourth application to check what the AI found this morning. She has a system she already lives in. If the AI isn't inside that system, sitting exactly where the next decision gets made, it might as well not exist.
What Builders Mutual Actually Built
Builders Mutual is a construction-only Southeast insurer. Ken's path ran through claims leadership, field training, and IT program management at Nationwide before he got handed a blank page and asked to stand up a department from nothing. That combination is probably why his answer wasn't about the smartest model on the market. It was about the fewest clicks.
The tools that lasted at Builders Mutual all shared one trait. They showed up inside the workflow the adjuster was already running instead of asking her to leave it. A summary inside the claim file, not a separate portal. A fraud flag on the screen she was already looking at, not an email digest nobody opens. The AI that wins the argument in the boardroom and the AI that's still running six months into production are not always the same AI. Ken has buried a few of the boardroom kind.
This Is a Design Problem Before It's a Model Problem
Everybody in insurtech wants to argue about which model is smartest. Twelve years of building a claims department taught Ken a cheaper lesson. The model matters less than where you put it.
That's the same idea Terra, the claims platform I co-founded, is built around. An agent does the preparation and hands a licensed adjuster a case that's already worked, inside the file she's already in, not a tool she has to remember exists. I didn't land on that shape because a slide deck told me to. I landed on it because I've spent seven years listening to operators like Ken tell me the same story on a microphone. Software that sits outside the workflow gets ignored, no matter how good it is.
Something to Ask Before You Buy the Next Tool
If you're a CIO or a claims VP shopping for AI this quarter, ask a different question than the one on the RFP. Don't ask how accurate the model is. Ask where it lives. Ask what your adjuster has to open, click, or remember in order to see it work. If the answer is anything other than nowhere, it's already fighting the busy Tuesday. And the busy Tuesday wins more often than any vendor wants to admit.
Ken Bunn built a claims department from nothing and has kept it running for twelve years. He didn't do it by finding the smartest AI on the market. He did it by making sure the good ones disappeared into the work. That's the whole lesson, and I got it for free on a Friday afternoon in July.