Earlier this week, Insurance Insider hosted a webinar titled The AI Readiness Gap: Why Carriers Need a New Document Strategy, in partnership with Vertesia.
Moderator Meg Green led a candid discussion with four industry experts: Keith Schlosser, insurance industry advisor at Vertesia; Jenn Kyung, SVP of the P&C Business Performance Office at Nationwide; Shashi Reddy, Chief Digital, AI, and Technology Officer at PEMCO Insurance; and David Proestos, President and CEO of One Fathom Above.
What followed was one of the most honest conversations I've heard about why AI is underdelivering at so many carriers, and what it actually takes to fix it. The short answer? The problem is rarely the AI model itself. It almost always traces back to documents.
Insurance is a document-heavy business. Policies, claims, submissions, endorsements. Every core workflow is built on documents. But most carriers store those documents in systems that were designed decades ago, and those systems were never built to support the kind of intelligent, agentic AI that the industry is now trying to deploy.
Keith Schlosser put it plainly early in the discussion: "Documents are the lifeblood of our industry... those documents tend to sit in a repository that we would refer to as a legacy repository. I think of it as that drawer in the kitchen where everything goes into the drawer, and when you need something out of it, it takes you an hour to try and find it, and maybe you don't even find it."
That image (the kitchen drawer) resonated with every practitioner on the call. And it points to something deeper: the underlying issue is not that carriers lack ambition or talent. It's that the foundation their AI is being built on isn't ready.
Traditional enterprise content management (ECM) platforms like FileNet, Documentum, and OpenText were built to centralize and store documents. They handle metadata classification, version control, lifecycle governance, and access permissions reasonably well, but understanding a document is a different problem entirely.
David Proestos described the core challenge: "You've got to extract it from that content system. You've got to read it. Now you've got duplicate data... it doesn't follow the rules of compliance."
The moment a carrier pulls content out of a legacy ECM system to feed it into an AI system, they create a copy of that data, one that sits outside the governance controls of the original system. That creates real compliance exposure, especially in a regulated industry like insurance.
Keith Schlosser offered a useful analogy for what it feels like to add AI capabilities on top of a legacy document management platform: "It's kind of like buying a car and taking the engine out and putting in a different manufacturer's engine. It might work. It's going to be costly, and you're going to have challenges, but it might get you to the grocery store and back, but not all the time."
Modern, AI-native document platforms are different. They're built from the ground up to integrate with agentic AI workflows, which means they can extract the right data in a compliant way without creating the duplication problem that legacy systems produce.
Shashi Reddy pointed to another structural problem: fragmentation. "Most of our knowledge today is distributed across both structured and unstructured systems, with different degrees of accuracy." Getting AI to reason well across that landscape requires not just better tools, but a deliberate effort to consolidate and improve data quality before AI is introduced.
Shashi Reddy offered what may be the clearest definition of AI readiness to come out of the discussion: "It'd be like buying a Ferrari before you built the road." His point was that readiness is not primarily a technology question. It requires a clear strategy, trusted data in both structured and unstructured form, a scalable AI platform with good integration, and a thorough understanding of end-to-end business workflows before AI is introduced.
"AI itself isn't a strategy," Reddy added. "The real opportunity is to use AI to make work simpler, make better decisions, and really make our customers' and employee experience more human."
The panel was consistent on one point: don't try to solve every problem at once. David Proestos put it directly: "Don't try to do everything. AI is a fantastic tool, but you also have to be cognizant that it accomplishes certain things."
Shashi Reddy framed the same idea in operational terms: "To deploy AI at scale, we need to do it in thin slices, incremental value addition."
Jenn Kyung added that the starting point should always be the business outcome, not the technology: "Start with the end in mind. Think about the business process or the outcome that you're looking to achieve, and then the use case that will support that, and it will lead you to the answer."
Jenn Kyung shared a framework for monitoring AI programs in flight. The key, she said, is setting clear milestones upfront and not only financial ones. "What you would see are things like enhanced responsiveness, turnaround times, quality measures, capacity of resources." These early indicators tell a team whether the program is on track before the ROI timeline arrives.
As the session closed, each panelist offered their single biggest takeaway.
David Proestos addressed the majority of the audience who haven't started yet: "Don't be afraid of success... don't be afraid of change. At the same time, be smart about it."
Shashi Reddy returned to purpose: "We are at a rare moment where technology can fundamentally change how we work. But AI itself isn't a strategy."
Jenn Kyung reinforced the importance of beginning with a clear destination: "Start with the end in mind."
And Keith Schlosser closed with a call to action: "There's been a rush to implement AI. And what we've learned is that the underpinnings of AI are often somewhat challenging. And now is an opportunity to take a look at those underpinnings and accelerate your growth."
The message from this group of experienced practitioners was clear and consistent. The carriers that will win with AI are not the ones racing to deploy the most models. They're the ones taking the time to get their documents (and their data) in order first.
You can watch the full webinar recording here or on our YouTube channel.