Agentic AI: The biggest shift in insurance since digital transformation

For the past few years, the AI conversation in insurance has revolved around one word: automation. Insurers have been drawn in by the promise of faster claims processing, smarter chatbots and more efficient workflows.

But those gains are only part of the opportunity. The next step is to use AI to help grow the business.

Insurers are beginning to use agentic AI to to do exactly that. The technology reasons, plans and takes action to achieve a goal, working alongside people rather than waiting for instructions. It’s the difference between software that follows a script and software that understands the objective.

This doesn’t mean it can simply take over complex insurance processes, however. Yes, an AI agent can complete a task, but there are levels to that, and it should never take over human expertise. In my view, the true value of AI lies in removing the administrative burden that gets in the way of expert judgment. When insurers get that right, the payoffs will be significant.

• ALSO READ: Why insurers’ AI ambitions are exposing the cloud execution gap

Why insurance is ready for agentic AI

A decade ago, the insurance industry was often seen as a technology laggard. Because it’s a highly regulated sector, change naturally happened more cautiously.

The picture looks very different now. Insurers are sitting on one of the richest collections of data of any industry — millions of policies, claims files, broker communications and inspection reports, and decades of underwriting decisions.

Until recently, much of that information was locked down in PDFs, forms and other unstructured documents that were difficult to use at scale. But all that data has rapidly turned into a powerful advantage, as large language models (LLMs) have finally become good at understanding unstructured information.

Insurers have also spent years strengthening their underlying data foundations, which means they are now in the ideal position to move beyond AI initiatives that focus only on efficiency gains to look at where AI can drive profitable growth.

AI pursues goals, but human experts have the final say

Traditional AI is very good at analyzing information and automating predefined tasks, but it’s still fundamentally reactive. Agentic AI behaves differently.

Imagine an underwriter preparing a submission for a complex commercial property policy. Instead of downloading reports, reading lengthy engineering documents, checking historical claims and emailing colleagues for missing information, the underwriter gives the AI a goal: Prepare a complete risk assessment.

The AI agent gathers data from multiple systems, identifies gaps, retrieves supporting documents, highlights inconsistencies, applies the organization’s underwriting guidelines and presents a recommendation, along with the evidence behind every conclusion.

Underwriting is, in fact, one of the strongest opportunities for agentic AI because it’s one of the core functions of an insurance company. I’ve spoken with carriers whose underwriters simply don’t have the capacity to review every policy submission they receive. There are opportunities they never get to because there just aren’t enough hours in the day.

That changes the economics of underwriting. Giving an experienced underwriter time back is a productivity gain, but it also creates capacity to evaluate submissions that might otherwise go untouched, respond to brokers faster and spend more time on the risks where human expertise creates value. Agentic AI therefore becomes more than a cost-reduction tool: it can increase the productive capacity of the underwriting organization and help the carrier pursue more profitable business.

Define the work boundaries for AI and humans

When people hear the word “autonomous”, they often assume AI is going to start making insurance decisions on its own. That’s not the future I see.

Some activities are obvious candidates for AI. No human should be spending hours searching through 200-page PDF documents for a handful of data points. AI should be doing that work.

But other decisions clearly belong with people. I won’t have AI decide whether to pay a $15 million claim, and I won’t have AI independently determine whether someone qualifies for life-insurance coverage. Those decisions involve legal, ethical and regulatory considerations that require human accountability.

Insurers therefore need to determine where AI stops and human judgment begins. AI should handle repetitive work, surface relevant insights and provide recommendations. Humans should remain responsible for judgment, oversight and accountability. Think of it as a spectrum of autonomy.

This way, agentic AI makes the expertise of seasoned underwriters and claims professionals even more valuable.

Think bigger than just piecemeal gains

One of the biggest reasons I see AI initiatives stall has surprisingly little to do with AI itself. Too many organizations are trying to automate pieces of existing processes instead of asking whether those processes still make sense.

I was recently asked: What would an AI-native insurance company really look like? That’s a very different question from asking how to add an AI chatbot into an existing workflow.

Whereas the latter approach delivers incremental improvements, the former asks whether claims, underwriting, policy servicing or distribution should be redesigned from the ground up. That takes more work up-front because it brings together people, processes and technology - but the potential upside is also much greater. Instead of making a process 10% or 20% more efficient, you may discover you can redesign it entirely and improve it by 80%.

At this point, we should spend less time asking what AI can do and more time asking what humans shouldn’t be spending their time doing in the first place.

Access to AI isn’t your competitive advantage

Competitive advantage doesn’t come simply from having more data or access to the same LLMs as everyone else. It comes from turning your proprietary data — along with decades of expert decisions, underwriting and claims outcomes, business rules and operational processes — into intelligence that AI can use again and again.

That institutional knowledge gives AI the context to understand not just insurance in general but also how your organization evaluates risk, makes decisions and creates value.

And the more your employees work with AI, the richer that knowledge becomes. Each interaction can capture decisions, actions and outcomes in a more structured form, creating a learning loop that helps AI support the next decision more effectively. Over time, you build an intelligence advantage that access to the underlying models alone cannot replicate.

When you combine that with thoughtful governance and leadership, you create a competitive edge that new entrants won’t have. Because the value doesn’t lie in the AI model itself but in what it can learn from your business, leading to a future where insurance is intelligently collaborative, with people and AI each doing what they do best.

This is why I believe agentic AI is the next frontier in transforming insurance for the better.

• ALSO READ: Why mainframe modernization in insurance is a strategic advantage

What to do next

Read more about NTT DATA AI for Insurance to see how you can use agentic AI to do business faster, more fairly and more profitably.