The insurance sector sits in an uncomfortable position relative to AI. The pressure to adopt is significant, from boards, from competitors, from actuarial teams that can see exactly what the technology could do for underwriting precision and claims efficiency. And yet the pace of structured, scalable adoption remains slow. Most insurers have employees using ChatGPT or Copilot in some capacity. Very few have built the organizational infrastructure to turn that individual usage into institutional capability.
The irony is not lost on the sector's leadership. Insurance companies manage risk for a living. They are, by design, expert at assessing probability, building governance frameworks, and operating in regulated environments. These are precisely the capabilities required to adopt AI responsibly at scale. And yet the same risk culture that makes insurers good at their core business has, in many cases, become a brake on AI adoption rather than an enabler of it.
Understanding why, and what the exceptions look like is the starting point for any serious AI adoption in insurance.
The use cases generating measurable returns in insurance are not the most technically sophisticated. They are the ones where the combination of high task volume, document intensity, and process structure makes AI assistance immediately productive.
Claims management is the highest-impact area in most organizations that have deployed seriously. A large European mutual insurer implemented AI-assisted claims triage in its property damage division, routing incoming claims by complexity and pre-populating assessor reports from initial client documentation. The result was a reduction in average handling time of roughly 35 percent for standard claims, with assessors redirecting freed capacity toward complex cases requiring genuine judgment. AI did not replace the assessor. It removed the administrative layer that was consuming the majority of their working time.
Underwriting support is a second area of substantial value, particularly for commercial lines where risk assessment involves synthesizing large volumes of heterogeneous documents, financial statements, technical reports, inspection records, industry data. AI tools that handle document synthesis and flag relevant risk indicators allow underwriters to process more submissions without sacrificing assessment quality. Several large reinsurers have deployed this capability quietly and effectively, without public announcement, which is itself revealing about how the sector manages the optics of AI adoption.
Regulatory compliance and policy documentation represent a third high-value use case that is often underestimated. Insurers operate under dense and evolving regulatory frameworks, Solvency II, IDD, DORA, and their national transpositions, that generate continuous documentation and review obligations. AI-assisted synthesis of regulatory updates, gap analysis against internal policies, and drafting of compliance documentation has compressed work that previously occupied entire compliance teams for weeks into days. This is one of the clearest ROI cases in the sector, and one of the least discussed, despite being a focus area for the best AI consulting firms for enterprise transformation in 2025.
Client-facing applications, personalized policy explanations, claims status communication, renewal support, have produced more mixed results. Where AI has been deployed in direct client interaction without adequate governance, the outcomes have been inconsistent and in some cases reputationally damaging. Where it has been used to assist human advisors rather than replace them, the results have been significantly more positive.
The gap between these pockets of genuine value creation and the broader adoption picture in insurance is explained by three structural patterns that appear repeatedly across the sector.
The first is the pilot-to-scale problem. Insurance organizations are good at running experiments. They are less good at converting successful experiments into standard operating procedures. A claims AI pilot that works well in one regional division requires process standardization, system integration, change management for AI adoption, and cross-functional governance to scale across the organization. Innovation teams that designed the pilot are rarely positioned, or empowered to drive that organizational work. The pilot succeeds. The capability does not scale. The next year, a new pilot is commissioned.
The second pattern is governance paralysis. Unlike the banking sector, where regulatory constraints around AI are already shaping adoption strategies, insurance regulators have been slower to publish clear AI governance frameworks. This ambiguity has not produced bold experimentation, it has produced defensive inaction. Legal and compliance teams, unable to cite a clear regulatory standard, default to blocking AI deployment rather than designing governance that is proportionate to the actual risk. The result is that organizations whose risk culture should be an asset in AI governance have turned it into an obstacle.
The third pattern is the absence of a shared adoption framework. In most insurers, AI usage is happening, but individually, inconsistently, and invisibly. Actuaries have built their own workflows. Claims teams have their own tools. Marketing uses a different platform. No one has mapped these uses against a common framework, measured their impact, or identified the conflicts and synergies between them. Building structured organizational capability requires exactly this mapping work, and most organizations have not done it.
The insurers that have moved beyond experimentation share a recognizable set of organizational conditions. None of them are technically exceptional. All of them are organizationally intentional.
The first condition is senior sponsorship that is operational, not symbolic. AI programs that report to a Chief Innovation Officer with no operational authority consistently fail to scale. Programs that are co-owned by the COO or a business line CEO with accountability tied to operational metrics, not innovation KPIs, consistently do better. The difference is not enthusiasm. It is organizational leverage.
The second condition is a use case prioritization framework that distinguishes between AI as process efficiency and AI as decision support. These are fundamentally different categories with different governance requirements. Automating the formatting of a claims report carries minimal regulatory risk. Assisting in a coverage decision involves different accountability questions entirely. Organizations that conflate these categories create governance frameworks that are either too restrictive for low-risk use cases or insufficiently rigorous for high-stakes ones.
The third condition is workforce integration that goes beyond training. Deploying AI tools without redesigning the workflows they sit within produces partial adoption at best. The claims assessors who were most resistant to AI-assisted triage in one large mutual were not resistant to the technology, they were resistant to a tool that had been added to their existing process without reducing any of the steps they already performed. When the workflow was redesigned so that AI assistance actually removed work rather than adding a new interface to manage, adoption followed quickly.
The insurance industry's relationship with AI is at an inflection point. The organizations that are still treating AI as a series of individual experiments will find themselves structurally behind competitors who have built organizational capability, not in three years, but in eighteen months.
The path forward is not more pilots. It is the organizational work that turns the evidence already accumulated, from claims pilots, underwriting tools, compliance applications into a coherent adoption architecture. That means a use case map with clear prioritization, a governance framework proportionate to actual risk levels, a measurement methodology that translates operational gains into financial terms, and a workforce development program that builds capability rather than just access.
The sector's risk culture is not the problem. Undirected risk culture, without a framework to channel it productively is the problem. The same discipline that makes a good actuary is exactly what good AI governance requires. The question is whether insurance organizations will apply that discipline to their own transformation.
Mendo works with insurers and mutuals to design AI adoption programs built for regulated environments, from use case prioritization to governance architecture and workforce capability, with the rigor the sector demands.