What’s stalling AI in insurance and what to do about it

Most enterprises building with AI agents right now face the same tension. Give teams too much freedom and you get sprawl and dozens of disconnected agents. Each is useful alone, but none of them work together. Lock everything down instead, and you slow the very innovation you set out to unlock.

Insurance has a reputation for moving carefully. For that reason, it might not be a surprise to learn that, in the OutSystems State of AI Development for Insurance report, 35% of respondents say their AI projects haven’t started. The reason is that they’re too time-consuming. That’s the highest rate of any industry we surveyed. Insurance leads the market in AI projects that get scoped and then shelved.

This isn’t normal caution. Insurers are looking at the projected effort and deciding the AI initiative costs too much to attempt. Something about insurance makes that calculation come out worse than it does for banks, manufacturers, or retailers.

Project avoidance can be a rational decision

What’s happening is avoidance, which is often rational. When the setup cost of a project is high and the payoff is uncertain, waiting is a defensible choice. Insurance executives are doing the math that their training rewards. They weigh the effort of standing up a new capability against the odds it pays back, and many projects fail that test before anyone writes a requirement.

The question worth asking is why the math comes out harder here. Insurers and bankers both work under heavy regulation. Both carry decades of legacy systems. Yet insurance posts the highest avoidance rate, and the reason traces back to how data and governance demands stack up in this particular industry.

Why data and governance weigh more in insurance

Let’s start with the data. Most insurers have accumulated multiple generations of policy administration, claims, and rating systems. This was the result of mergers that added platforms faster than anyone retired them. Before an agent can do anything useful, someone has to reconcile data across those systems. That reconciliation becomes the first and largest line item in the project plan.

Now add governance. Insurance decisions have to be explainable. An underwriting model or a claims recommendation may need to satisfy a regulator, an auditor, and an internal actuarial review. So before a project starts, leaders are already counting the cost of documentation, model validation, and compliance sign-off. That governance overhead is priced into the decision to begin.

A third factor gets less attention. The feedback loop in insurance is slow. An agent that misprices risk doesn’t reveal the mistake the next day. It surfaces months later as adverse selection or a reserve shortfall. That delay raises the bar for proof, which means more upfront testing and more time. When the cost of being wrong is high and slow to surface, leaders want more certainty before they commit.

The cost that never appears in the project plan

Put those factors together and a pattern emerges. The thing stopping insurance AI projects has little to do with AI. It’s the setup around it.

Every project carries a hidden setup tax, one that includes harmonizing the data and wiring up the governance so an agent can work inside the business safely. Under the traditional model, teams pay that tax in full, then pay it again on the next project, and again on the one after that. The context never carries forward. Each effort starts from a blank page, and the blank page is expensive.

That’s why insurance feels the pressure more than other industries. Its setup tax is the heaviest, and the traditional model offers no way to spread the cost across projects.

How a context layer and orchestration change the math

Clearing this hurdle might seem too much, especially when caution has been the AI modus operandi. But all it takes is a change in outlook, from trying to do this with multiple, disconnected AI tools and broken context to moving to an agentic systems platform. It can attack the setup tax by making context and governance shared infrastructure rather than per-project work.

Connected context

The Enterprise Context Graph provides a live understanding of how an insurer’s business works, connecting applications, workflows, data, systems, rules, and operational relationships into a unified layer of enterprise context. The data reconciliation an insurer used to repeat for every project becomes something the platform maintains once.

Intelligent orchestration

Agentic Enterprise Orchestration builds on that foundation. It intelligently coordinates agents, apps, workflows, models, and enterprise context across the organization. And it routes work while preserving the operational understanding underneath. Because governance is grounded in that same context layer, auditability and compliance come built in rather than added at the end.

The effect is to lower the cost of starting. When context and governance carry forward, a project that looked too time-consuming last quarter becomes a reasonable thing to attempt this quarter.

How Hollard harmonized its processes and started delivering

Hollard, one of South Africa’s largest insurers, is a working example of the math changing. The company wanted to modernize its back-office operations, and its team needed a faster way to deliver internal-process innovation than traditional development allowed. Working with OutSystems and certified partner Linkit, Hollard built an agile team that turned internal processes into integrated, customer-centric workflows.

Embedding AI into its operations workflow system cut manual effort sharply. Hollard ran more than 300,000 customer data searches, processed 48,000 cases, and saved over $100,000. Brighton Ravele, the insurer’s Senior Manager for Digital Transformation, says that OutSystems “helped us harmonize processes, aligning with our strategic objective of building customer-centric solutions.”

That’s the setup tax coming down in practice, with the harmonization that once stalled projects now carrying them forward.

The projects insurers keep postponing are the ones worth doing

The 35% figure describes an industry making sound decisions under a cost structure that punishes new work. It isn’t a sign that insurance can’t keep up.

Change the cost structure, and the decisions change with it. The underwriting agent, the claims triage tool, and the broker self-service workflow that kept getting postponed were never too hard to build. Lower the setup tax, and the projects insurers have been avoiding become the projects they finally start. You can do that with an agentic systems platform like OutSystems.

Get all the insurance and agentic stats in the state of AI development report 2026.

Interested in learning more about the agentic systems platform? Check out the OutSystems platform page.