Grant Thornton finds AI governance gap

Insurance companies are starting to get what they wanted from AI. Revenue is up at some firms. Costs are down at others. Decision-making is getting faster. But the awkward part is that many carriers still cannot show, in a clean and auditable way, who owns these systems, how decisions get checked, or whether the controls would survive outside scrutiny. That is the gap Grant Thornton is pointing at in its 2026 insurance AI survey — and it is a pretty insurance-specific problem, because this is an industry that sells trust and lives under regulation. ### What did Grant Thornton actually find? The headline numbers are simple. In Grant Thornton’s 2026 AI Impact Survey, 52% of insurance respondents said AI had driven revenue growth, 62% said it improved decision-making insights, and 50% said it reduced costs. But 44% also said governance or compliance challenges had already contributed to AI project failure or underperformance. That is the split-screen story — AI is working, and the controls are not keeping up. (grantthornton.com) ### Why is the 24% figure the one that sticks? Because it turns an abstract governance problem into a stress test. Only 24% of insurance executives were very confident they could pass an independent AI governance review within 90 days. In other words, most carriers are not saying “we should tighten this u(grantthornton.com)oof. (insurancejournal.com) ### What does “governance” mean here? Basically, decision rights, controls, documentation, and workflow discipline. Who is allowed to approve an AI use case? Who signs off on model changes? How do claims, underwriting, legal, compliance, and IT hand work to each other? What happens when a model drifts, produces a bad output, or touches sensitive cus(insurancejournal.com)n answering those questions in a repeatable way. (grantthornton.com) ### Why is insurance a hard place to wing it? Because insurance decisions are operational and regulated at the same time. An AI tool might summarize claims files, flag fraud, support underwriting, or route customer service work. But those outputs can affect pricing, coverage, claims handling, and customer treatm(grantthornton.com)omer disputes, and weaker evidence when regulators or auditors ask how the system works. That is why “provable governance” matters more here than in a lower-stakes back-office setting. (grantthornton.com) ### Is this just an insurance story? Not really. Grant Thornton has been making a broader argument about an “AI proof gap” across industries. Its wider 2026 survey said 78% of executives lack strong confidence they could pass an independent AI governance audit within 90 days. The insurance slice matters b(grantthornton.com)t that being good at insuring risk is not the same thing as governing AI risk inside your own workflows. (grantthornton.com) ### So why does this create consulting demand? Because most companies do not need another generic AI demo. They need operating plumbing. They need model governance, escalation paths, testing routines, audit trails, role clarity, and a way to connect compliance teams with business teams without freezing deployment. That(grantthornton.com)hat lets you use AI in underwriting or claims without creating a governance mess.” This is partly analysis and partly inference from the services Grant Thornton is emphasizing around AI adoption and governance. (grantthornton.com) ### What is the real takeaway? The interesting part is not that insurers are behind on AI. They are not. The interesting part is that adoption is now far enough along that weak governance is becoming visible in missed performance, failed projects, and audit anxiety. AI in insurance is no longer a pilot problem. It is an operating-model problem.