Many insurance leaders I talk to are already well into their AI journey. They often have pilots in claims, experiments in underwriting, a copilot rolled out to a servicing team, and maybe a proof of concept sitting with compliance. On paper, the industry looks like it is moving quickly.
And yet, when the conversation turns to how much the business itself has changed, the answer is often more modest than expected. The pilots work, and the demos are impressive, but very little of that effort has translated into lasting, enterprise-wide value.
I don’t think that’s a technology problem, and I want to explain why that is. As I see it, the issue is rarely a lack of AI capability. More often, the bottleneck exists within the way insurance work is structured.
We are deploying systems that can reason, adapt, and act into operating models still built around disconnected systems, departmental handoffs, and manually assembled context. We’re observing that the silos resulting from legacy analog processes are perpetuating. The problem is, those models are held together by institutional knowledge in people’s heads, information trapped in legacy systems, and handoffs between departments that were never designed to share much of anything.
Until we address that mismatch, continuing to expand AI deployments will not close the gap, but a new operating model for insurance can. Specifically, we need a model where people, automation, agents, and decisions work from a connected foundation of enterprise knowledge. Only then will claims, underwriting, servicing, and compliance stop operating as isolated functions and begin executing as coordinated parts of the same enterprise. For example, there are several nodes of the policyholder journey that predict churn; when that exchange is siloed in servicing, a critical and immediate “Next Best Action” opportunity is lost and churn risk increases.
Why Insurance is Especially Exposed
Behind every policy, claim, endorsement, and servicing request is a chain of decisions, rules, expertise, and judgment. Information moves constantly between underwriting, policy administration, claims, compliance, customer service, and broker networks. At every handoff, information gets lost, re-keyed, or re-researched, and the exceptions are often where the real cost lives.
And on top of that, we have the reality of the conditions most carriers operate under. Core systems are often decades old and were never built to talk to one another, data sits in silos, each with its own version of the truth, and most of what the organization “knows” lives in the experience of a handful of long-tenured employees rather than anywhere a system can reach. And all of it operates under regulatory pressure that makes auditability and defensible decisions non-negotiable.
The result is fragmented execution. In P&C, that shows up as leakage: claims dollars that were paid and shouldn’t have been. Fraud is part of it, but so are duplicate payments, missed deductibles, settlements closed without documentation, and adjusters working off standard procedure. It runs roughly 6% of total claims payments across all P&C lines, about $67 billion a year in the US. There may be no shortage of capabilities, but there is still a shortage of connection. And when you drop AI into a fragmented environment, you’re automating around it, one silo at a time.
The Real Challenge is Knowledge Fragmentation
Here’s a pattern I see often: Claims builds something to speed up claims. Underwriting builds something to speed up underwriting. Servicing builds something to speed up servicing. Each initiative delivers a real, local win. And collectively, they add up to a more efficient version of the same disconnected organization.
Organizations following that pattern are improving individual workflows, sure, but they aren’t redesigning how work moves across the enterprise. The reason for that, often, is that all of these initiatives are drawing on trapped knowledge. Claims, underwriting, servicing, and compliance often have access to different pieces of the story. Valuable context exists across systems, departments, and documentation, but it is not always available at the moment a decision is being made. As a result, each function can end up operating from a partial view of reality, and AI inherits those same limitations. So, each function ends up operating from its own version of reality, and the AI faithfully reproduces those boundaries, but with greater speed.
This is why isolated AI initiatives stall. It isn’t necessarily that the technology underperforms. The challenge is that applying intelligence to fragmented knowledge often produces fragmented results. An agent is only as capable as the context it can reach, and in most carriers, the context stops at the edge of the department that built it.
All of this points to where the leverage really is: The next wave of insurance transformation will not come from making claims, underwriting, and servicing individually smarter. It will come from connecting what each of them knows and orchestrating how those functions work together across the entire lifecycle of the policyholder experience, leading to improved business performance.
What is the Enterprise Knowledge Layer?
We have heat-mapped 9 Insurance Functions and 47 processes within [MC2.1]those functions to identify the potential gains from automation. While necessary, this isn’t sufficient for a blueprint that differentiates. Instead, the connective tissue between these nodes is where the opportunity exists. The[KR3.1] enterprise knowledge layer is the foundation that allows an organization to coordinate work across people, systems, automation, and agents. It captures and connects the business rules, process knowledge, policy knowledge, historical decisions, compliance requirements, and workflow context that normally exist in isolated places across the enterprise. Rather than forcing employees and systems to reconstruct that knowledge repeatedly, it creates a shared, governed foundation that can be used consistently across workflows and functions. As a further example, we talked about churn signals earlier. There are also growth signals that can be captured and agentically activated such as policyholder lifestage indicators (adding a new driver, a sale of an asset (home or business property) that may signal the need for replacement.
When that layer exists, claims, underwriting, servicing, compliance, and producer-management teams stop working from separate versions of reality. The claim adjuster and the underwriter are drawing on the same understanding of the risk, the servicing agent and the compliance team are working from the same rules, and the teams supporting independent agents have the context they need to make appointment, commission, and policyholder-service processes easier to navigate. For carriers that rely on independent agents who represent multiple insurers, that ease of doing business can become a real distribution advantage. Knowledge stops being something each function reconstructs on its own and becomes something the enterprise holds in common.
That shared foundation is what makes coordination possible, not just between people, but also between automated workflows and agents. Agents become useful precisely because they finally have a reliable context to reason over. Automation can become safer because it’s operating against governed rules rather than local interpretation. And humans spend less time hunting for information and more time on the judgment that absolutely requires them.
Still, the knowledge layer is not the end goal, but the foundation that allows the enterprise to safely orchestrate people, agents, automation, and decisions at scale. An insurer with a strong enterprise knowledge layer and practical AI is often in a stronger position than one with impressive AI sitting on top of fragmented knowledge.
What Changes When Knowledge Becomes Enterprise-Wide
Let’s examine how different areas are affected when knowledge becomes enterprise-wide:
Claims. Say a complex claim comes in. Today, an adjuster may spend significant time reconstructing history: pulling the policy, finding prior claims, checking coverage, chasing documentation across systems, confirming what compliance requires for this claim type in this jurisdiction.
In an agentic model, an agent assembles that picture from the enterprise knowledge layer: the full claim and policy history, the relevant coverage terms, the applicable rules, the documentation status, all in one place. The adjuster starts from a complete, consistent view instead of building it by hand. The decision stays firmly with the adjuster: how to adjudicate, whether to escalate, what to pay. What changes is that their capacity is spent on judgment rather than research.
Policy servicing. A servicing request that spans billing, coverage, and producer relationships often requires employees to navigate multiple systems just to answer a simple question. With a connected knowledge layer, an agent can assemble the relevant information into a single working context while the representative focuses on the customer. Judgment stays with the employee; the busywork does not.
When organizations connect knowledge across functions, the outcome is not simply faster work. It is a different operating capacity. Decisions move with more context. Employees spend less time assembling information and more time applying expertise. Customers experience fewer delays, fewer handoffs, and more consistent service. Leaders gain greater visibility into how work moves across the enterprise. The value comes from improving the system as a whole rather than accelerating isolated tasks. Often we encounter operational constraints that are limited by their inability to scale manually-intensive processes. For example, one insurer was able to cut is review threshold in half with the same staffing model.
Agentic systems differ from traditional automation because they can reason through context, adapt to new information, and pursue defined goals within the governance constraints the organization sets. That makes them better suited to exception-heavy work that spans systems, knowledge sources, and departments.
This is also why the first question organizations should ask is not which agent to deploy. The more important question is what business outcome they are trying to improve, what knowledge that work depends on, and what operating model will allow people, automation, and agents to act on that knowledge safely and consistently.
Building an Agentic Insurer
If the enterprise knowledge layer is the foundation, the natural question is how you build toward it. In our work at Naviant, this comes down to three things that reinforce one another:
A strategic blueprint. Before any of this touches a system, leaders must decide what they are trying to change. Which business outcomes matter most, like:
- Where operational friction exists today.
- Which workflows create the greatest opportunity for transformation
- How a human and digital workforce will work together to achieve those goals
- Where connected knowledge and agentic execution create real leverage, and where they introduce risk the business isn’t willing to take
- How success will be measured in terms the business cares about, like cycle time, combined ratio, customer experience, reserve adequacy, and compliance posture, rather than technical benchmarks
Agentic systems amplify whatever intent they are given. If the priorities are vague, you simply execute ambiguity faster, so clarity is what keeps the investment pointed at work that genuinely benefits from it.
Enterprise architecture. This is where the knowledge layer gets built. It means creating the orchestration layer that lets work move end-to-end across functions, and the connected knowledge foundation that layer draws on, replacing one-off integrations over time rather than demanding a fully unified backbone on day one. It means data that is accessible, permissions that are deliberate, governance and auditability built in rather than bolted on, and the flexibility to evolve capabilities without rewriting the enterprise every time. For a regulated business, this is the part that makes agentic execution defensible instead of risky. Architecture that fights you will stall the whole effort; architecture aligned to intent lets it scale. Architecture that is reliant upon an exclusive technology will sentence you to technical debt.
Continuous innovation. An agentic insurer is not a project with a finish line. The organizations that get lasting value treat deployment as the beginning: improving workflows as they learn, tightening the knowledge layer as decisions accumulate, adapting the operating model as autonomy earns more trust. That requires change management that prepares people to work alongside agents, clear accountability for when a system recommends versus when a person decides, and feedback loops that turn what the organization learns into better operating logic over time.
Taken together, these are less a technology roadmap than an operating discipline. Strategy defines the intent, architecture makes it executable and connected, and continuous innovation keeps it improving. The knowledge layer is what all three are ultimately in service of.
The Competitive Advantage isn’t AI
Let me end where I started, because I think this is the thing insurance leaders most need to get right.
The future of this industry will not belong to the carriers with the most AI tools. It won’t belong to whoever runs the most pilots or ships the most agents. Those things are becoming commodities, and a fragmented organization with a lot of AI is still a fragmented organization.
The advantage will belong to insurers that can coordinate people, automation, agents, knowledge, and decisions across the whole enterprise. They can take what claims knows, what underwriting knows, what servicing and compliance know, and make it available as one connected foundation the entire organization operates from. The competitive edge is not the AI itself. It is the ability to turn enterprise knowledge into coordinated execution across the business.
That is what an agentic insurer really is, an organization that has redesigned how work gets done so that its people, its knowledge, and its digital workforce finally pull in the same direction. The carriers that understand this will be better positioned than those still automating one silo at a time.
The question I’d leave every insurance leader with is a simple one: is your organization building a smarter version of the silos you already have, or the connected foundation that makes all of them work together? The answer determines which side of that gap you end up on.
