Insurer AI adoption has moved decisively from isolated experimentation toward formal strategic integration, but governance maturity, not adoption speed alone, will increasingly determine which insurers actually benefit, according to the latest S&P Global Ratings survey.
The survey covered 121 rated re/insurance entities globally, representing roughly 38% of the total assets S&P rates in the sector.
"The survey results show that AI is increasingly used to improve customer experience, underwriting and risk management, and claims processing, with operational benefits appearing earlier and more visibly than financial gains," said S&P Global Ratings analyst Andreas Lindberg.
The numbers behind that shift are substantial.
Insurers surveyed expect 6% to 7% efficiency gains and 4% to 5% revenue improvements from AI by 2028, and the share expecting AI-related cost savings exceeding 3% is projected to rise sharply, from 16% in 2025 to 80% by 2028.
S&P also found insurers plan to more than double the share of technology budgets allocated to AI over the next three years, concentrated on workflow automation, workforce productivity, customer solutions and risk management. Meanwhile, governance infrastructure is already widespread: nearly all insurers surveyed have established or are developing formal AI governance frameworks, and almost two-thirds maintain AI model inventories.
S&P was explicit that this isn't simply a best-practices observation.
"We view this focus on governance as a signal of industry maturity and a recognition that governance weaknesses could inhibit effective AI scaling, potentially resulting in model inaccuracies, regulatory breaches, and costly remediation efforts that could negatively impact credit quality," the report said.
While S&P noted no AI-related rating actions have occurred in the sector to date, it noted that "variations in AI readiness, governance maturity, and data capabilities may increasingly influence competitive advantage, risk exposure, and ultimately creditworthiness" going forward.
Ross Sinclair, CEO of insurtech EIP, argued the real test of this governance question plays out most concretely in claims handling specifically.
"While there are myriad opportunities to automate the insurance workflow, the challenge for the sector is making sure that enthusiasm doesn't outpace the need for strong controls that are required to implement and use AI safely," Sinclair said. "Within claims management in particular, there is a big difference between using AI to gather information, spot anomalies or process straightforward queries, as opposed to asking it to decide whether a claim should or shouldn't be paid."
Sinclair expects that as AI agents become more autonomous, insurance-specific tools will become increasingly important.
That distinction isn't abstract for Sinclair's own business.
EIP, founded in the UK in 2004, built a claims rules engine in 2013 that it says has driven automated decisioning for more than a decade, and layered a voice-led AI agent on top of it this year specifically to handle information-gathering and customer interaction, while keeping actual claims-payment decisions governed by insurer-configured rules rather than a probabilistic model.
Sinclair has also been a vocal critic of vendors making broader automation claims, writing publicly last December that "to suggest that a software system, however advanced, can make binding claims decisions or interpret complex, evolving regulation without human oversight is more than optimistic," a position consistent with the more measured framing in his comments here.
S&P's data and Sinclair's operational framing point to the same practical conclusion for insurers assessing AI vendors and internal build decisions. The meaningful risk differentiator isn't whether AI touches a claim, but which part of the claims process it's allowed to influence, and how auditable and deterministic that influence is.
An AI system that gathers information, flags anomalies, or answers routine policy questions carries a fundamentally different risk and governance profile than one making or materially influencing a payment decision.
As S&P's survey suggests insurers are set to significantly expand AI budgets over the next three years, the distinction between those two categories of use case is likely to become a standard line item in how insurers, regulators and rating agencies alike evaluate AI risk going forward.
