The Part I Got Wrong
July 2019. I launched the InsurTech Geek Podcast — co-hosted with Rob Galbraith — with a simple premise: stop reading press releases and start talking to practitioners. Seven years and 177 episodes later, I've been wrong about some big things. I've also been right about a few. I think both are worth saying out loud.
Let's start with the misses, because they're more instructive.
Blockchain. I gave it real airtime in the early days. I believed — genuinely believed — that a shared, immutable ledger had obvious applications in a business built on trust between parties who don't trust each other. Reinsurance reconciliation. Parametric triggers. Proof of coverage. It sounded right. By 2021 the conversation had essentially dried up. Not because the technology failed — but because every problem we thought blockchain solved turned out to be solvable better another way. Usually with a well-designed API and a contract.
AI adoption speed. In 2019 and 2020 I was bullish. Too bullish, too early. I thought the technology was there, the use cases were obvious, and the industry would move fast. It didn't. And the reason wasn't the technology — it was everything else. Regulatory complexity, talent scarcity, and the organizational change that comes with rewiring how underwriters and claims adjusters actually work. I underestimated all three. By a lot.
Startup advantage. I thought being born outside the legacy system was a structural moat. No mainframes. No org debt. No sacred cows. Startups could move fast and the incumbents would scramble to keep up. What I learned: speed isn't the moat in insurance. Relationships are the moat. Distribution is the moat. Regulatory standing is the moat. A lot of those startups moved very fast — directly into a wall.
The Part I Got Right
Now for the wins, such as they are.
AI would eventually matter enormously. I was wrong about the timeline. I was right about the direction. By 2025 it was undeniable — not as a marketing claim, but as a measurable force reshaping carrier operations. The companies that started building the capability in 2020 and 2021, even before it was obvious, were three years ahead when it became table stakes.
Data is the real battleground. Not UX. Not workflow automation. Not the interface layer. The companies winning in insurance tech right now are winning because of data quality and data integration. The best model running on bad data is still a bad system. This was true in 2019 and it's more true today.
Agents weren't going anywhere. Every cycle produced a wave of confident predictions about disintermediation. Every cycle was wrong. Independent agents are still here, still a dominant force in commercial distribution, and the channel adapted faster than the disruptors expected. This one I called early and I'll defend it.
The format. Long-form practitioner conversations produce better signal than panels, tweets, or 90-second takes. That was the bet I made in 2019. I built Geek Out: Ten Years of InsurTech, Told by the People Who Built It from those 174 conversations for exactly that reason — the patterns only became visible across time and volume. I think I was right about the format.
The Meta-Lesson
When I look at the pattern across seven years, something stands out. The predictions I got wrong were almost all "X will happen faster than people think." The predictions I got right were almost all "X will happen eventually, because the fundamentals favor it."
Time horizons are where forecasting breaks down in complex industries. Not the direction — the direction is often obvious if you look at incentives and fundamentals. It's the clock that kills you. Insurance is a slow-moving system with a lot of inertia, and that inertia moves in both directions: it delays adoption of good ideas and it also delays the death of bad ones.
"The market can remain irrational longer than you can remain solvent." — commonly attributed to John Maynard Keynes. That's a finance line, but it applies to technology forecasting too. You can be right about the destination and completely wrong about the arrival time.
So what do the next seven years look like? I have views. But I'm holding them more loosely now — and paying a lot more attention to the organizational and regulatory constraints than to what the technology can technically do. Because that's where the predictions actually fall apart.
The technology is usually not the bottleneck. It hasn't been for a while.