Grant Thornton Survey Exposes the Insurance AI Proof Gap

Grant Thornton's 2026 AI Impact Survey, fielded February 23 to March 18 across 100 insurance respondents from 950 business leaders, found 52% reporting measurable AI revenue growth and 24% very confident they could pass an independent governance review within 90 days.

That 28-point spread is what the firm calls the AI proof gap. Set against four other surveys published the same quarter, it is the only one measuring provable governance rather than adoption, and every other survey lands in the same place from a different direction.

Key Takeaways

52% report AI revenue growth against 24% audit confidence, a 28-point proof gap, with the revenue figure sitting 15 percentage points above the cross-industry average.
56% name regulatory or compliance uncertainty as the top barrier to scaling, against 29% citing talent or upskilling, so the constraint is the framework rather than the workforce.
Bain's gap is 74 points and Capgemini's is 50, measuring operational scale rather than governance, which means four independent surveys locate the same bottleneck.
68% say AI controls exist but the evidence is fragmented across departments, platforms and documentation systems, which is the specific failure mode the NAIC exhibits are built to find.
Only 7% of insurance respondents believe their workforce is fully ready, and COO confidence runs 32 points below CIO and CTO confidence on the same question.

What the Insurance Subsample Says

Among the 100 insurance executives, 52% report AI has contributed to measurable revenue growth, 15 percentage points above the cross-industry average. Sixty-two percent report improved decision-making, 50% report direct cost reductions, and 62% rate their maturity as scaling across multiple functions rather than piloting.

Against that, 24% describe themselves as very confident of passing an independent review of AI governance and controls within 90 days. The remaining 76% acknowledge fragmented controls, incomplete documentation, or untested incident response.

The failure modes inside that 76% are specific. Sixty-eight percent say controls exist but evidence is scattered: model risk holds validation records, compliance holds bias testing, IT holds data lineage, and nothing aggregates them into a record an examiner could read as one governance history. Forty-four percent say governance or compliance problems have already contributed to an AI project failing or underperforming, which is retrospective rather than anticipatory.

The barrier ranking is the finding that reframes the rest. Fifty-six percent name regulatory or compliance uncertainty as the single largest obstacle to scaling AI across more functions, against 29% citing talent or upskilling gaps. The technology and the people are available; what is missing is a settled framework worth committing to.

Four Surveys, One Bottleneck

The value of the proof gap is that it can be triangulated, because the other Q1 2026 surveys measure adjacent things on different samples.

Survey (2026)	Sample	Adoption/Revenue Metric	Governance/Scale Metric	Gap
Grant Thornton	100 insurance execs	52% report AI revenue growth	24% audit-ready in 90 days	28 pts
AM Best	150+ rated insurers	63% formal AI policy	47% robust governance	16 pts
Capgemini	344 P&C execs	60% exploring/POC stage	10% successfully scaled	50 pts
Datos Insights	36 carrier tech leaders	61% AI in production	Document processing focus	Scope gap
Bain	160 global insurers	78% adopted gen AI	4% scaled meaningfully	74 pts

AM Best surveyed over 150 rated insurers and MGAs and found 63% with a formal AI policy but only 47% describing governance as robust, a 16-point gap of the same shape. Its top challenges are all upstream of documentation: 45% cite data readiness, 43% security and privacy, 41% legacy system integration. Sridhar Manyem's framing is that AI produces unreliable outputs when the underlying data is fragmented across legacy systems or insufficiently governed, which is the same fragmentation Grant Thornton measured, observed at the data layer rather than the evidence layer.

Capgemini's 19th World P&C Insurance Report, from 344 senior executives, puts 10% of P&C insurers at successful scale with 60% still exploring or in proof of concept, and 42% tracking no AI metrics at all. Its spending split explains why: seventy-two percent of carrier AI investment goes to technology and infrastructure against 28% to change management, training and organizational readiness, the category that contains governance program development. The 10% Capgemini calls trailblazers are roughly four times more likely to invest in change management and three times more likely to have deployed explainable AI.

Bain surveyed 160 global insurers and found 78% adopting generative AI in some form against 4% scaling it meaningfully across claims, a 74-point gap. Datos Insights adds the scope dimension: production AI moved from 37% to 61% of its 36 respondents in a year, but concentrated in document processing, with underwriting at 56% and claims at 50% both at limited scope.

Taken together, three-quarters of carriers have adopted AI, fewer than one in ten have scaled it, and fewer than one in four could evidence that what is deployed is governed. The metrics differ; the bottleneck does not.

The Gap Becomes a Filing Requirement

What changes the exposure is that the examination the 76% say they would fail is being standardized.

The NAIC's AI Systems Evaluation Tool pilot launched March 2, 2026 across 12 states, runs through September, and is expected to be adopted at the November 2026 fall meeting. Its four exhibits amount to the independent governance examination the survey question was probing: Exhibit A quantifies AI usage across functions, Exhibit B applies a governance risk assessment, Exhibit C requires documentation on high-risk systems in claims, underwriting, pricing and fraud detection, and Exhibit D covers data details.

The tool creates no new governance requirements. It creates a standardized way to test whether existing ones are met, and it asks for exactly the aggregated, coherent record that 68% of respondents say is currently scattered. A carrier reporting AI revenue growth is by definition running AI in one of the Exhibit C categories.

Third-party exposure compounds it. The NAIC Model Bulletin has been implemented by over half of all states, a model law on third-party AI vendor oversight is anticipated in 2026, and the IA Capital Group survey puts OpenAI in 90% of carrier stacks. Carriers will have to evidence not only their own governance but the governance of vendor systems embedded in their workflows.

The workforce numbers say who does that work. Only 7% of insurance respondents believe their workforce is fully ready for AI, 47% call it mostly ready, and thirty-nine percent say frontline employees need the most support. Cross-industry the readiness view splits sharply by seat: thirty-nine percent of CIOs and CTOs think the workforce is fully ready against 7% of COOs. In insurance the COO typically owns claims operations and underwriting workflow, which is where the Exhibit C systems live. The 32-point confidence gap runs in the direction that matters, with the people closest to the high-risk deployments the least convinced anyone is ready to document them.

Grant Thornton Survey Exposes the Insurance AI Proof Gap

Key Takeaways

What the Insurance Subsample Says

Four Surveys, One Bottleneck

The Gap Becomes a Filing Requirement

Further Reading

Sources