The hype cycle around AI in insurance is cooling, and that’s a good thing. What’s replacing it is something far more valuable: honest conversations about where AI creates real leverage and where it doesn’t. In a recent InsurTech Association webinar, Joe Zuk, Operating Partner at Altamont Capital Partners, sat down with ITA co-founder Josh Hollander to unpack what it really means for an insurance business to be “AI ready” and why most of the industry is still thinking about it wrong.
Zuk brings a rare dual perspective. He’s spent his career straddling the operator and investor line across MGAs, carriers, reinsurance, and platform plays. His message was clear: AI readiness isn’t a vendor decision. It’s an operating model decision.
Takeaway 1: AI Readiness Is an Operating Model Decision, Not a Vendor Decision
The biggest misconception Zuk sees? Companies treating AI as something to bolt onto existing workflows. “LLMs are becoming much more the commodity now,” he said. “What’s going to be really essential here is data.” Legacy insurers that simply layer AI tools on top of decades-old processes hit an “integration ceiling” fast. The real gains come when companies redesign workflows so that AI and humans collaborate by design, not as an afterthought.
He pointed to a UK firm he’d spoken with that morning that built its entire stack (underwriting workbench, policy admin, billing) as AI-native from the ground up. The result: fewer handoffs, shorter cycle times, and human talent freed to focus on exceptions, portfolio strategy, and creative problem-solving that AI can’t yet replicate.
Takeaway 2: Buy Commoditized Capabilities, Build What’s Proprietary
When it comes to build versus buy, Zuk offered a practical decision framework. If it’s commoditized infrastructure like document processing, OCR, or data ingestion, buy it. Vendors are getting very good at this, and there’s no reason to reinvent the wheel. But if it touches your core differentiation (your proprietary data, your underwriting mousetrap, your service edge), that’s where building makes sense.
There’s also a middle path gaining traction: buy the base model layer from a vendor, then build a proprietary application layer on top with your own data, prompts, and evaluation workflows. “I would not go try to build my own LLM,” he said. “But building the proprietary application layer? That’s where the investment should go.”
Takeaway 3: The Talent Gap Isn’t Technical. It’s Translational.
Perhaps the most resonant theme was talent. Zuk used the metaphor of a “Rosetta Stone”: the kind of person who can translate between business context and technology capability. “I probably don’t need PhDs,” he said, “but I do need folks that can translate both the practical business context and the insurance application, and be at least conversant enough in the technology.”
He also issued a warning: companies that have banned AI tools entirely, even something as basic as Copilot, “aren’t going to be long for this world over the next five, seven, ten years.” Meanwhile, InsurTech startups often underestimate how quickly they need to bring in traditional insurance expertise for regulatory navigation, distribution relationships, and product development.
Zuk’s parting thought summed up the session: “AI, in my view, is going to be an augmenter, a capability multiplier, that in and of itself isn’t a strategy. But if you can deploy it in your proprietary sales niche or your proprietary data capture, that’s where AI really creates value.”
The full recording of this session is now available on our site. Whether you’re a carrier executive weighing your AI roadmap, an InsurTech founder looking for investor signals, or a broker exploring how technology reshapes distribution, this conversation is well worth your time.
