Most industry surveys now show AI in use at nearly every carrier, but that headline obscures how deeply they actually use it. A model running one live workflow differs from a deployment that works across the organization, delivers measurable value and maintains consistent governance. That difference is what matters when assessing AI adoption in insurance.
Claims teams may use AI to summarize files or monitor claims progress, while underwriters may use AI agents to shorten submission intake and data entry. These tools can improve workflow performance, but claims professionals and underwriters retain authority over coverage, pricing, payouts, and denials.
This article compares adoption benchmarks by maturity stage, population, and line of business, then examines results across claims, fraud detection, underwriting, and policyholder service, along with the barriers that keep pilots from scaling. Use the figures to separate live workflows from enterprise-wide scale, and projected opportunity from measured returns.
Insurance AI Deployment, Scale, and Growth
Compare insurance AI adoption rates only when the studies measure the same maturity stage, population, and line of business. "Live operations," "in production," and "at scale" describe different levels and cannot be treated as one trend line. A higher rate may reflect one targeted workflow rather than enterprise-wide deployment, and missing sample sizes or fieldwork dates limit direct comparisons.
2027 Insurance AI Adoption Benchmarks
These are the latest available benchmarks for 2027 planning, drawn from research fielded and published in 2025 and 2026. Use them when deciding whether a reported adoption rate reflects one live workflow, organization-wide scale, positive ROI, or measured value in a core workflow.
AI in live operations: The Covenir operations report found that 70% of organizations had AI in live operations, up from 58% one year earlier. This is sponsored survey research from an operations vendor, not independent analyst research. (152 US operations decision-makers at carriers, MGAs, and insurtechs; fielded February–March 2026 and published June 9, 2026.)
Generative AI in production, distinct from agentic AI: Celent's GenAI review reported that 48% of global insurers were in production with generative AI. (Global insurers; published January 12, 2026. The published summary did not state the sample size.)
AI embedded in workflows: 81% had embedded AI across most or some workflows. (400 global insurance executives; published June 1, 2026. Fieldwork date was not stated.)
P&C AI scaling: Only 10% successfully scaled AI. About 60% remained in exploration or proof of concept. (344 executives, 809 employees, and 1,113 policyholders across the Americas, Europe, and Asia-Pacific; reported May 2026.)
P&C value at scale and modeled impact: BCG's AI-First P&C Insurer reported that 38% generated value at scale from AI in core workflows. (P&C insurers; published March 2026. No sample size was stated. The value measure covers core workflows rather than enterprise-wide deployment.)
Life, annuity, and group production: The 2026 implementation report found that 73% ran AI in production, while 7% had scaled it enterprise-wide and 33% reported positive ROI. (Life, annuity, and group-benefits insurers; published August 2026. No fieldwork date or sample size was stated, and the full report sits behind a lead-capture form.)
Agentic AI deployment and ROI measurement: Datos insurance ebook found that 17% were running agentic AI, while fewer than one-quarter of 42 senior US carrier technology leaders could measure AI returns despite most having AI in production. (42 senior US carrier technology leaders; 2026.)
Agentic AI intent: Celent's agentic AI report reported that 22% planned an agentic AI solution by the end of 2026. (Insurance survey; published May 20, 2025. This reflects planned adoption rather than deployment.)
How to Compare Maturity Measures
Treat production as a workflow status. On its own, it doesn't prove enterprise scale, positive ROI, or a successful workflow redesign. A practical maturity path runs from limited exploration to pilots, production workflows, and organization-wide scale.
Production means the model operates in a live workflow with data access, governance, and human review already in place. Scale means those same practices extend beyond one team's pilot to multiple business units or the full organization. Count a deployment as scaled only when it meets both tests: live workflows that connect to measurable outcomes, and human accountability that hasn't been removed to get there.
Insurance AI Cost, Growth, and ROI Benchmarks
Use financial benchmarks only after separating realized performance from projected opportunity. Workflow-level KPIs determine whether a deployment has created measurable value.
Between 2022 and 2024, insurers using advanced analytics had combined ratios six percentage points lower and premium growth three percentage points higher than slower adopters, according to a WTW survey of 59 P&C insurers in the US and Canada, published in March 2026. The comparison is observational and does not establish that analytics adoption caused the difference.
In a March 2025 survey, EY's spring 2025 insurance research found that most carriers expected average cost savings to rise from below 10% to 11–20% within two years; roughly 55% of respondents fell in that range. Anticipated savings should not be reported as realized ROI.
BCG modeled that US P&C operating costs per premium dollar could fall 15–25%, equal to $35–60 billion in operating expenses, with early movers achieving 3–5% premium growth. These are modeled projections, not reported industry-wide results.
Financial impact is increasingly visible, but measurement maturity trails deployment. Leaders should require workflow-level KPIs such as handling time, routing accuracy, complaints, quote turnaround, combined ratio, and positive ROI before calling a deployment scaled.
AI Adoption and Impact in Core Insurance Workflows
Move a workflow into production only when you document three things: a deployment rate from a comparable population, a measured result tied to a workflow KPI, and a defined point where a person reviews the outcome. Claims routing, fraud flags, underwriting intake, and policyholder service each have adoption benchmarks below; policy administration and other emerging workflows call for the same review discipline even without dedicated benchmark data yet.
Claims AI Adoption and Automation Results
Use AI agents in claims to speed up claim-file intake, summarization, and routing. Claims benchmarks are most useful when they separate live workflow adoption from carrier-specific performance gains.
The Datos conference summary found that, in April 2025, 21% used AI in claims. The same research also found that 35% used it for summarization and 22% for document processing, but it didn't state a date for those two figures.
WTW's analytics survey, published in March 2026, found that 14% used straight-through processing in claims workflow automation, while an additional 36% planned to adopt it.
An April 2026 analysis reported that Aviva's more than 80 claims models had shortened liability-assessment time, improved routing accuracy, reduced complaints, and saved more than £60 million in 2024. A carrier-outcomes report said Allianz's Project Nemo shortened food-spoilage claim resolution time, with a person approving every payout.
BCG's P&C executive perspectives estimated that first-notice-of-loss (FNOL) automation and adjuster copilots could improve processing speed by as much as 50%. This is an illustrative estimate, not a measured industry average.
Decision boundary: Do not use carrier savings or modeled processing gains as a proxy for portfolio-wide straight-through processing: AI agents can prepare and route the claim file, but an adjuster should retain payout and denial decisions.
Fraud Analytics Adoption and Loss Projections
Use AI agents for fraud detection, with investigators reviewing flagged claims, false positives, and explanations before any adverse action. Current deployment and projected avoided losses are not equivalent measures.
A WTW survey, published in March 2026, found that 33% of surveyed US and Canadian P&C insurers used advanced claims analytics for fraud detection and 29% used it for severity assessment, and that insurers expect both figures to reach 65–70% within two years.
Deloitte's fraud analysis estimated that AI across the claims lifecycle could prevent $80–160 billion in fraud losses by 2032. The estimate is a forecast, not a measured saving. The same research estimated current detection rates of 20–40% for soft fraud and 40–80% for hard fraud.
Decision boundary: Fraud detection has a large projected financial opportunity, but carriers still need to track false positives, investigator workload, and explainability rather than treating flagged claims as confirmed fraud.
Underwriting AI Adoption and Productivity
Use AI agents in underwriting when submission intake and data entry take time away from risk selection and pricing. The operational test is whether intake time falls without weakening risk selection, pricing, or review quality.
Accenture's underwriting report found that underwriting executives expected AI use to rise from 14% to 70% over three years. The study surveyed 430 underwriting executives and reports expectations, not completed adoption. The same research found that 33% of underwriters already use solutions that gather, cleanse, and provide quality data views, while adoption of integrated solutions that prevent redundant data entry is expected to rise from 9% today to 72% within three years.
Zurich reported that an underwriting platform reduced manual triage time from 75 minutes to 15 minutes, improved digitization accuracy, and expanded intake straight-through processing. Zurich reported these underwriting results in 2026.
Capgemini's P&C report found that 41–43% of underwriter time went to administrative activities, compared with 32–33% on core risk assessment and premium calculation.
Decision boundary: Automating intake and data entry should free underwriter time for risk selection and pricing, not replace the judgment behind either one: a faster submission process that shifts freed time into unreviewed risk decisions is not a net gain.
Policyholder-Service AI Adoption and Comfort
Use AI agents when high-volume claim-status and billing questions are creating service backlogs. Policyholders should still be able to reach a person for complex or consequential questions, and adoption does not remove the need for transparency.
EY's insurance survey found that 68% of insurers were investing in chatbots to drive cross-selling, and that among insurers prioritizing agentic AI, 60% put customer service and engagement at the top of the list.
The Geneva Association report, fielded in February 2025 and published in November 2025, studied 6,000 insurance customers across six countries and found that insurer-deployed chatbots and virtual assistants were the most-noticed application at 45%, while 68% had used a general-purpose generative AI tool such as ChatGPT or Copilot somewhere in their insurance-buying process. The report classified these as generative AI applications rather than broader agentic AI workflows.
On November 4, 2025, J.D. Power reported that 47% were uncomfortable with AI processing claims. Policyholders were most comfortable with AI for claim-status updates (24%), billing (23%), and basic service (21%).
Decision boundary: Deploy AI agents for claim status and billing only when they can route coverage or payment exceptions to a person; the comfort data does not justify removing human review from consequential claims work.
Insurance AI Scaling, Data, Talent, and Governance Barriers
A live pilot isn't ready to scale when data access, legacy integration, KPI ownership, training, governance, or review responsibility remain unresolved. Current studies show that these operational barriers recur more often than a lack of strategic interest.
Scaling and Measurement Barriers
AM Best's April 2026 insurer research found that legacy-system integration and data quality or accessibility rank among the top obstacles to scaling AI beyond pilots; legacy systems often store data in inconsistent, unstandardized formats that degrade the AI results built on top of them.
Capgemini's 2026 World Property and Casualty Insurance Report found that 42% of insurers have never measured AI outcomes, and most who did could not show clear ROI.
Talent and Governance Barriers
In 2026, insurers called AI strategic but did not believe they had the required talent, according to Datos insurance research.
Despite rising deployment, the Covenir operations report found that 20% of organizations cut AI training budgets in 2026 while only 7% increased them, suggesting investment in the people running these systems isn't keeping pace with the technology.
KPMG's insurance report found that, in September 2026, insurance executives rated their workforce's AI proficiency poorly and believed failing to respond to AI would undermine competitiveness.
The National Association of Insurance Commissioners' (NAIC) adoption map, dated as of August 31, 2026, shows the NAIC Model Bulletin on insurers' use of AI adopted across 25 states plus Washington, D.C., with California, Colorado, New York, and Texas listed separately for insurance-specific AI regulation or guidance.
Don't move a pilot into broader production until you document reliable data access, legacy-system integration, KPI ownership, team training, governance, and review responsibility.
Automate Claims, Underwriting, and Policy Administration With Datagrid
Start with claim routing when a claims AI pilot keeps stalling before production. Define handling time and routing accuracy as workflow KPIs, specify the claim conditions that trigger adjuster review, track complaints, and move the workflow into production only after ownership of those controls is documented. Datagrid's AI agents can connect to claims, underwriting, and policy administration systems to handle the document and data work behind these workflows, from claim intake through policy reconciliation:
Claim file summarization and routing: Analyze incoming claim files and route them to the right adjuster based on complexity, coverage type, and required review.
Submission intake and data extraction: Extract applicant, property, and risk data from submission documents and complete the corresponding fields in underwriting systems.
Fraud signal flagging: Flag claims carrying fraud indicators for investigator review, with the triggering factors attached to each flag.
Policyholder status and billing responses: Generate routine claim-status and billing responses and route coverage or payment exceptions to a person.
Document and data reconciliation: Cross-check policy documents, endorsements, and claims records against source systems and flag inconsistencies before they reach a reviewer.
Claims professionals and underwriters review every flagged finding and retain decisions on coverage, pricing, payout, and denial.
Get started with Datagrid to see how connected workflows handle the document and data work behind one claims or underwriting process.
Frequently Asked Questions About AI Adoption in Insurance
What Is Meant by AI Adoption?
AI adoption ranges from early exploration and pilots to live, governed workflows across the organization. The word alone doesn't signal maturity: a survey measuring "AI adoption" may be counting a single pilot or an enterprise-wide deployment with data access, measurable outcomes, human review, and accountability already built in. Checking which stage a statistic describes is what makes it useful.
How Is AI Being Used in Insurance?
Insurers use AI agents to route claims, summarize claim files, extract data from documents, flag fraud signals, triage underwriting submissions, and deliver routine status or billing responses. Adjusters and underwriters keep the final call on coverage, pricing, payouts, and denials, so automation frees their time for exceptions rather than replacing professional judgment.
Why Do Insurers Report Different AI Adoption Rates?
Insurers report different AI adoption rates because studies measure different maturity stages, populations, dates, and lines of business. Live operations, production use, workflow-level value, and enterprise-wide scale are distinct measures and shouldn't be treated as a single trend line.
Which Metrics Show That Insurance AI Has Scaled Successfully?
Treat insurance AI as scaled only when live workflows deliver measurable outcomes under human accountability. In claims, track handling time, routing accuracy, complaints, and review points. In underwriting, track submission intake and quote turnaround. At the portfolio level, track combined ratio and positive ROI.
