Insurer AI Returns: What the Evident AI Index Measures and What an Actuarial Scorecard Should

The 2026 Evident AI Index scores 30 major insurers against 68 capability indicators, and 3 of the 30 disclose a comparable enterprise-level financial return: Manulife at $217 million, Generali at $116 million and Intact Financial at $145 million of 2025 AI-attributed value. The other 27 are ranked on process proxies, meaning headcount, job postings and use-case volume, rather than on anything that appears in a loss ratio.

Key Takeaways

• 3 of 30 carriers disclose an enterprise AI return figure. The index is built on public data, so the ranking measures what carriers say they have built, not what it earned.
• 49% of documented use cases remain narrow, aimed at speed and cost reduction, and only 8% show agentic reasoning. Speed does not reach a loss ratio unless it changes risk selection or severity.
• 42% of insurers track no AI performance metrics at all and more than 55% report unclear returns, on Capgemini's survey of 344 senior executives.
• 72% of AI investment goes to infrastructure against 28% to training and change management, which is why 47% of employees with AI access report an essentially unchanged workday after 18 months.
• A 5.4 point industry loss ratio decline from 2023 to 2024 overlaps the deployment window, so any carrier attributing its full improvement to AI is counting the rate cycle inside the technology.

What the Index Can and Cannot Show

The capability signals are real. AI-specialist roles across the 30 carriers grew 32% year over year and now run at roughly 1 in 50 employees, while broader insurance headcount contracted 2.2%. Twenty of the 30 publicly report at least one use case with documented outcomes, up 8 from the prior year, and agentic deployments moved from 5% to 25% of newly disclosed use cases in six months.

The design has a ceiling built into it. Allianz tops the 2026 ranking with an AI talent pool 28% larger than AXA's, which is visible in job postings. Whether it is moving Allianz's combined ratio is not in the index, because the index reads public disclosure and that disclosure does not exist.

The three carriers that did put a number on it project more than $1 billion of AI-attributed value within two years: Manulife to $723 million by end-2027, Generali to $407 million, Intact to $361 million by the end of the decade. Intact's 2025 figure was itself revised up 33% to $145 million.

None of the three splits the figure between underwriting, claims and operations, or says whether it represents loss ratio improvement, expense ratio reduction or both. These are boardroom metrics. An actuary pricing a book or reserving a portfolio needs to know which line of the P&L the dollars sit in before the number can enter a technical analysis.

What a Return Metric Has to Survive to Be Credible

The measurement deficit is partly a choice of metric. Carriers report what is easy to count: claims straight-through processed, processing speed, manual touches removed. Those are operationally valid and financially unattributable. Five metrics carry the actuarial load, each with a credibility floor and a known confounder.

Metric	What AI Should Move	Minimum Credibility Threshold	Primary Confounder
Loss ratio (AI-screened segment)	Selection quality; adverse risk exclusion	3,000+ exposure units, 2+ policy years	Underwriting mix shift
Hit ratio	Submission binding quality; new business selection	1,000+ submissions per period, matched control period	Risk appetite or distribution shift
Claim leakage rate	Closure accuracy; reduction in payments above reserve	500+ closed claims; documented pre-AI leakage baseline	Claim settlement timing shift
Expense ratio by function	Cost per unit of underwriting or claims volume	Full-year expense reclassification; shared-service allocation rule	Shared-service cost allocation methodology
Claims cycle time and ALAE	Days-to-close; allocated loss adjustment expense per claim	1,000+ closed claims per line; severity-stratified comparison	Complexity mix across claim severity bands

The expense ratio is the most traceable of them. The P&C industry expense ratio fell to 25.3 in 2024 from 27.7 in 2014, a 2.4 point move driven mainly by remote work and operational consolidation rather than AI, and Morgan Stanley projects a further 2.0 points by 2030 attributable to AI, worth $9.3 billion of operating income across its carrier cohort (Carrier Management). On a $2 billion expense base, 2 points is $40 million.

The loss ratio side is where the credibility floor binds. For a carrier writing $1 billion of premium with AI-screened renewals at 40% of the portfolio, a 3-point loss ratio improvement on that segment is $12 million of underwriting benefit. Getting to that number requires mix, pricing level, limits profile and territory distribution to hold between the pre-AI and post-AI periods, and they rarely hold together.

Volume is the other floor. A 500-claim pilot carries roughly 30 to 40 percent credibility weight on a pure premium indication at moderate severity, which is why a 10-point loss ratio improvement reported off a pilot that size is inside the range of ordinary variance. The table's 3,000 exposure units and two policy years for the loss ratio metric are that credibility standard, not a stylistic preference.

Three Confounders Sitting Between the Spend and the Signal

Mix shift is the most common. A carrier deploying AI submission screening usually tightens appetite or shifts distribution toward lower-hazard segments at the same time. The loss ratio improves and the driver is partly the mix. Isolating the AI share requires holding the risk classification distribution, territory exposure, limits profile and pricing tier composition constant, which is exactly the data layer the 72%-to-28% infrastructure split has not funded.

The cycle is the second. The U.S. P&C loss ratio fell 5.4 points from 2023 to 2024, driven mainly by earned premium growth outpacing loss emergence after two years of rate action. Carriers deploying underwriting AI into that environment show improvement that includes the tailwind. The reverse holds too: AI deployed during an adverse development cycle can be working and show nothing.

Settlement timing is the third, and the one actuaries are best placed to catch because it lives in development patterns. AI-accelerated closure moves claims out of IBNR and into paid earlier than historical lag factors predict, which compresses the reported loss ratio without reducing the economic cost of the claim. Both periods have to be restated on the same settlement-lag basis, or the improvement partially reverses as development matures.

Carrier self-selection sits underneath all three. Capgemini's trailblazer cohort, the top 10% scaling AI successfully, showed 21% higher revenue growth and a 51% greater share price increase over three years (Capgemini, May 2026). Those carriers also entered the period with better underwriting discipline, cleaner data and higher profitability, which is what made the deployment tractable. The differential is real; the share of it attributable to AI cannot be read off the correlation.