Insurance Says It's Leading on AI. Its Own Org Charts Say Otherwise

Insurance has quietly become one of the better test cases for what "AI adoption" actually means once the pilots end. A KPMG International survey released September 29 found that 44% of insurance executives place their own organization in the top quartile for AI transformation. The same survey found that no insurer, across 20 countries and six regions, has fully redesigned sales, distribution, or underwriting around AI. Just 3% have overhauled claims or policy servicing. The gap between how insurers rate themselves and what they've actually rebuilt is the story — and it tracks almost exactly with what a separate Capgemini study of the same industry found back in May, which makes it less a one-off survey result than a pattern holding across an entire sector for the better part of a year.

Confidence outran the rewiring

KPMG's numbers describe a sector that adopted AI as a layer on top of existing work rather than a reason to rebuild it. 71% of insurers say their primary AI use case is content generation or routine task automation — drafting, summarizing, triaging — not the underwriting or claims decisioning that actually determines margin. Only 29% run any process end-to-end through AI agents. And the money reflects the caution: nearly half of insurance AI budgets go to back-office efficiency, versus just 5-10% aimed at new revenue models. 92% of respondents say AI is cutting costs or lifting productivity, but only 25% say it's driving growth through new products or services — a 67-point split between where AI is making insurers cheaper to run and where it's making them more competitive.

That split matters more in insurance than almost any other regulated industry, because the KPMG data also shows insurers increasingly expect AI to govern functions with real downside risk: 72% expect underwriting to run on a hybrid human-AI model with a smaller workforce by 2029. Deploying agentic tools into cost-saving back-office tasks is a low-stakes way to look like an AI leader on a survey. Redesigning underwriting — the function respondents themselves say is headed for AI-driven headcount cuts within three years — is the harder, slower work almost no one has started.

The gap has a price tag, and Capgemini already measured it

KPMG's September survey isn't an isolated data point. Capgemini's Research Institute ran a parallel study of the property & casualty segment specifically in May, built from 344 executive interviews and over 800 employee surveys, and landed on a strikingly similar split: only 10% of P&C insurers qualified as AI "trailblazers" who had moved past pilots into operational redesign. The other 90% remained in exploration or proof-of-concept. The consequence wasn't abstract — Capgemini found trailblazers posted 21% higher revenue growth and 51% greater share-price appreciation over three years than the rest of the field. Both firms also independently converged on the same root cause: data foundations. KPMG puts the share of insurers with data infrastructure strong enough to scale AI beyond pilots at just 11%; Capgemini found only 12% report high data maturity, and 42% of insurers track no AI performance metrics at all. Two consultancies, two different survey populations, five months apart, and the same number shows up twice — that's a stronger signal than either report is on its own.

"The gap between activity and transformation is where the real opportunity and risk now sit," said Dr. Frank Pfaffenzeller, global head of insurance at KPMG International.

The workforce side compounds the problem. KPMG found just 8% of insurers rate their own staff as highly proficient with AI tools despite 54% claiming effective training programs — a gap between training activity and actual capability that mirrors the gap between AI activity and actual transformation at the org level. Capgemini's employee survey found the same disconnect from the ground up: 47% of employees reported their workday was unchanged 18 months after getting AI tool access, and 43% said they were worried about job security even as the tools sat largely unused for anything consequential.

What to watch: insurance is a useful bellwether precisely because it's a regulated, risk-averse industry where the cost of a botched underwriting redesign is existential in a way a botched chatbot isn't — so its caution is rational, not laggard behavior. The number to track from here isn't adoption (already near-universal) but the 29%-running-agents-end-to-end figure and whether it moves before 2029, when 72% of the industry says underwriting goes hybrid regardless of whether the data foundations are ready. If that figure is still in the 20s a year from now, the "AI leader" self-rating from this survey will look less like confidence and more like a sector that mistook deploying tools for redesigning the business.