The global insurance market is ready to hand repeatable operational work to AI, with 83% of respondents supporting AI execution of routine tasks, according to new research from ISG.
According to the study, a further 75% said they would only allow that execution suing a model built specifically for insurance or governed within their own institutional rules, while just 6% said they would trust a general-purpose AI model alone for high-consequence underwriting and claims decisions.
The study surveyed senior leaders across underwriting, operations, claims, technology and transformation functions across North America, Europe and Asia, covering 20 operational activities spanning submission intake and triage, quote generation, bordereaux processing, claims adjudication and compliance screening.
Carriers estimate that one in nine broker submissions is currently declined or left unquoted not because of appetite, but because operations cannot process them in time, on risks insurers had the appetite to write in the first place. Where AI is already running in operations, 61% of respondents reported productivity improvements and 51% reported faster cycle times, with operating costs expected to fall 16% over two years. Asked what would most improve how brokers perceive them, respondents ranked pricing (64%), ease of doing business (52%) and AI-driven speed and completeness of submission response (51%) as the top factors.
Martin Henley, chief executive of mea Platform, said insurers are ready to hand repeated work to AI on one condition: a model that understands insurance and operates inside their own rules. He said that standard is demanding but the right one, since carriers are turning away business they had the appetite to write, and winning that back is worth more than any cost saving.
Henley, whose AI-native insurance technology business was founded in 2021, has previously told trade press that the productivity AI unlocks matters less in isolation than what insurers choose to do with the resulting capacity, with some clients already using it to expand appetite or enter lines they previously avoided.
The study draws a specific distinction between AI generally and what it calls a governed model: one that draws on controlled policy wording, endorsements, underwriting appetite and claims guidance, separates what it may read from what it may decide, allows a named person to override or stop it, and operates within an institution's own appetite, thresholds and referral triggers.
That distinction sits behind why 86% of respondents said consequential decisions, those that create genuine competitive distinction, should remain with people, even as the same respondents want AI running the repeatable work underneath those decisions.
That framing echoes a concern the Chartered Insurance Institute raised in August 2026, warning that carriers deploying AI in underwriting risk treating "human in the loop" as a formality unless the reviewer is genuinely equipped and empowered to question automated outputs.
The ISG research suggests most of the market already recognises that distinction in principle, since only 6% would trust a general-purpose model unsupervised, but recognising the risk and building governance robust enough to satisfy it in practice are evidently two different challenges.
Ninety-six percent (96%) of insurers have AI-led operational redesign on their agenda, but an advanced operating posture, where AI is central to workflow design, stands at just 13% today, against 52% expected within two years. Fully AI-native operation, where AI executes defined processes end to end while people set policy and manage exceptions, sits below 1% today, with only 12% expecting to reach it within two years.
Ashish Jhajharia, insurance subject matter expert and principal analyst at ISG, said insurers know exactly where they want the line between AI and human judgement to sit, but getting there is the part they haven't solved, calling the gap between 96% ambition and 13% actual advanced adoption the defining challenge of the next two years.
The report attributed that gap to genuine operating-model work, decision rights, referral rules and evidence trails that let a system act on an institution's behalf, rather than a simple technology procurement decision, noting that 27% of the market is still evaluating rather than building for exactly this reason.
The research lands amid a broader, already visible pattern of insurers building the kind of governed AI infrastructure this study describes. mea Platform itself partnered with Lloyd's MGA IQUW in 2024 to deploy generative AI in its D&O underwriting, aimed at converting unstructured submission data into actionable insight, while separate reporting has found only around one in 10 non-standard property submissions currently binds, illustrating just how much capacity genuinely sits idle in submission-heavy lines even when appetite exists.
Read against that backdrop, this study's central finding, that insurers see AI-driven capacity as a growth lever rather than simply a cost-cutting exercise, is consistent with a pattern already playing out in specific corners of the market, even if the fully AI-native operating model the research describes remains, by its own numbers, almost entirely aspirational for now.
