Why technology isn't a real barrier to AI

The insurance carriers getting the most out of generative AI right now tend to have something in common, and it has little to do with the technology itself. They spent years doing the less glamorous work like cloud migrations, retiring legacy systems, and consolidating data, before AI was even a line item. That investment, often justified purely on operational grounds, is what's allowing them to plug in AI tools and get results they can actually trust. Carriers still stuck in pilot mode or experiencing inconsistent accuracy are, in many cases, trying to skip that step entirely.

A recent AM Best survey found nearly 60% of carriers and MGAs expect AI to significantly change their business models within three years, but data readiness, security, and legacy integration remain their biggest barriers to deployment. 

At the end of the day, there is no substitute for having a clean, properly structured environment to support AI. Carriers can invest in the most sophisticated tools on the market, but if those tools pull from a patchwork of disconnected systems, the output will be unreliable. The AutoRek Insurance Operations report found that insurers manage an average of 17 separate data sources feeding their premium processes alone, and two-thirds handle more than 10. That kind of fragmentation doesn't get fixed by adding another tool on top. It takes real investment in the underlying infrastructure, and that work has to come first.

Picking the right tool for the right problem

Getting the infrastructure right is only half the equation. Even with the right foundation in place, choosing the right technology is trickier than it looks. Every major platform provider is pitching AI, and they all come with a vendor-centric view of what a carrier should buy. Put yourself in the CIO seat: On any given week, you've got multiple vendors making their case at the same time, each one framing their product as the answer.

It gets more complicated inside a single ecosystem. One cloud provider might offer several AI products at different maturity levels, and the distinctions between them aren't always obvious — sometimes even the vendors' own experts go back and forth on which one fits a given use case. 

An agentic AI tool that works well for call center knowledge retrieval may be entirely wrong for claims adjudication. Something that handles structured data efficiently may struggle with unstructured documents. The more productive path is to start with a clearly defined business problem, evaluate tools specifically for that problem, and accept that different use cases may require different solutions.

The people side matters more than most carriers expect

Most IT teams at mid-size carriers are already running at capacity on day-to-day operations, and AI work keeps getting fit into the margins. Trying to fit AI initiatives into existing workloads doesn't produce meaningful progress. These projects need dedicated people with protected time and clear ownership. That's easier said than done, especially at mid-size carriers where there isn't the luxury of standing up a separate AI team. But without that focus, the work stalls.

There's a human element to this, too. PwC's 2025 Global Workforce Hopes and Fears Survey found that nearly one-third of entry-level employees are worried about how AI will affect their careers. That apprehension is understandable. But carriers that have gotten through their first deployments tend to find that the same staff who were wary early on often become the most engaged once they see the technology working. 

When an AI tool takes over the tedious lookup tasks that used to eat up significant time on every call or claim, and people can spend more of their day on work that actually requires their judgment, the attitude changes. The buy-in has to be earned through results, which is why starting with the deployment most likely to produce a clear, measurable win makes more sense than swinging for the fences on the first attempt.

Building lasting competence, not permanent dependency

Vendor partners are valuable for getting early AI work off the ground. They bring expertise and resources that most carriers don't have in-house yet, and the right partner can meaningfully accelerate the learning curve. But carriers that hand over all AI work to outside partners without building their own knowledge base set themselves up for a dependency that only gets harder to unwind.

The better approach is to treat every vendor engagement as a chance to build internal capability. Each project should leave the carrier's own team a little more capable — better at evaluating tools, more familiar with how the technology actually works, and more confident taking on the next project with less outside help. That competence compounds over time. Structured AI training programs across business functions help everyone understand faster and give more people across the organization the fluency to identify where AI can add value in their own workflows. A team that has been through two or three deployments understands its own data, systems, and operational constraints better than any outside partner can. That institutional knowledge is what makes the fourth and fifth projects faster, cheaper, and better targeted.

Moving the fastest or spending the most does not guarantee real progress with AI. Those who have gained the most were candid about where their infrastructure stood, did the foundational work, picked specific problems to solve, and built momentum one deployment at a time. Do a few things well, learn from each one, and build from there. That foundation is what everything else runs on.