42% of P&C Insurers Never Measured AI Outcomes: Inside the Capgemini Measurement Gap

The Capgemini World Property and Casualty Insurance Report 2026, released May 5, found that 42% of P&C insurers have never measured AI outcomes, while a top 10% of "intelligence trailblazers" produced 21% higher revenue growth and 51% greater share price appreciation from 2021 to 2024.

The number that explains the 42% is the spending split. Insurers put 72% of AI investment into technology and 28% into change management.

Key Takeaways

• 42% have never measured AI outcomes at all, and among the 58% who tried, 55% report no clear ROI, so roughly four in five carriers cannot connect AI spending to a financial result.
• 72% technology, 28% change management is the investment split Capgemini's financial services research lead identifies as the structural cause rather than a symptom.
• 21% higher revenue growth and 51% greater share price appreciation separates the top 10% from mainstream peers over 2021 to 2024.
• 55% say it is unclear who owns AI initiatives, which is the same failure as the measurement gap seen from the governance side.
• 47% of employees using AI tools report no change to their workday after 18 months, against 14% who say they are very clear on AI's role in their work.

What the Report Actually Sampled

The 19th edition runs three separate instruments rather than one executive survey, which is what makes the internal contradiction visible.

Capgemini interviewed 344 C-suite and senior P&C leaders across the Americas, Europe and Asia Pacific, surveyed 809 employees split across underwriting, claims, customer service and agent roles, and polled 1,113 policyholders between December 2025 and February 2026. The executive stream produces the trailblazer finding; the employee stream produces the numbers that qualify it.

The trailblazer behaviours are specific. That cohort is nearly 4x more likely to invest in change management beyond basic AI literacy, nearly 3x more likely to have deployed explainable AI that lets underwriters and adjusters interrogate model outputs, and nearly 2x more likely to write AI responsibilities into job descriptions rather than layering tools onto existing role definitions.

Against that, 60% of P&C insurers remain in exploration or proof-of-concept. Combined with the 42% who have never measured, a large share of the industry is running pilots with no criterion for deciding whether to scale, pivot or stop.

Not Measured Means Not Attributable

The measurement gap matters most where an expense ratio assumption meets a rate filing, because unmeasured AI spending does not disappear from the numbers. It arrives as cost.

Morgan Stanley projects 200 basis points of P&C expense ratio improvement by 2030, from 30.4 to 28.5. That projection assumes the operational efficiencies offset the infrastructure they run on. Under the Capgemini split, the more likely industry outcome is bifurcated: trailblazers capture disproportionate savings, while carriers spending 72% on technology and 28% on absorption accumulate platform and data engineering costs that hit the expense ratio without the workflow change that would offset them.

For a pricing actuary that turns an industry aggregate into a poor benchmark. An expense trend fitted to industry data blends two populations moving in opposite directions, which understates the advantage accruing to the measured carriers and overstates the efficiency path of everyone else. The segmentation is not observable from filed data either, because the distinguishing variable is internal measurement discipline rather than anything that appears in a statutory exhibit.

What gets measured instead is usually the vendor's variable. Carriers that outsource deployment inherit model accuracy, processing speed or throughput volume as the success criteria. A vendor reporting a 94% accurate model has answered a different question from the one an actuary needs: whether that accuracy produced a 0.3-point movement in the loss ratio on the affected book. The 55% who say AI initiative ownership is unclear are by construction the same carriers where nobody owns the translation between the two.

The Workforce Reports Something Else

The employee stream is where the executive narrative stops matching, and it is the finding hardest to reconcile with any productivity projection.

Employee Survey Finding	Percentage	Implication
Workday unchanged after 18 months with AI tools	47%	Adoption without workflow redesign
"Very clear" on AI's role in their work	14%	Communication failure from leadership
Cite job security as top concern	43%	Trust deficit slows voluntary adoption
Time spent on cross-team collaboration	49%	Workflow fragmentation persists

Nearly half of employees equipped with AI tools report no meaningful change to their workday after 18 months. Either the tools sit outside the workflows they were meant to change, or they solve problems those employees did not have. Only 14% describe themselves as very clear on how AI fits their role, which removes the frontline feedback that a measurement system would need as its input, and 43% name job security as their top concern.

The training allocation runs the same way. Among carriers investing in AI skills, 86% focus on baseline literacy, 52% on governance and risk management, 46% on advanced skill development, and 27% on incentive and workflow redesign. The item that would close the gap receives the smallest share, while two-thirds of executives cite an AI skills shortage as a barrier to scaling.

Selection sits underneath all of it. The carriers publishing AI metrics are the ones with results to publish, so the visible examples skew toward the measured minority, and the 42% who have never measured are largely absent from the public record by definition. As Capgemini's Luca Russignan puts it, "The technology is maturing, but the organizational conditions to absorb it are not yet keeping pace."