Celent: 48% of Insurers Run GenAI in Production, Pushing Past the Early-Majority Threshold

From tracking GenAI deployment disclosures in carrier 10-Q filings and annual vendor surveys over three consecutive years, the acceleration from proof-of-concept to production is now measurable quarter by quarter. Celent's third annual "GenAI-oneers in Insurance" survey, published in Q1 2026 across separate P&C and Life editions, puts the global in-production figure at 48%. That number matters not because it is large in isolation, but because it crosses a threshold on the technology adoption curve: the transition from early-majority to late-majority deployment. In Everett Rogers's diffusion framework, that inflection point changes the competitive question from "should we adopt?" to "what happens if we don't?"

This is exceptional speed for an industry that historically takes five to seven years to adopt new technology platforms. Three years ago, only 8% of insurers had GenAI in production. Two years ago, the figure was roughly 28%. By mid-2025, it reached 44%. The jump to 48% in Q1 2026, combined with Celent's projection that late-majority status will be reached during 2026, compresses what would normally be a decade-long adoption cycle into less than four years.

Yet the 48% headline coexists with data showing that only 7% of insurers have achieved what Sedgwick defines as scalable AI success, and that 68% of organizations have moved 30% or fewer of their GenAI experiments into full production (Deloitte). This article examines what the Celent data actually measures, how it reconciles with lower-scale surveys, where the production use cases cluster, and what late-majority dynamics mean for carriers, vendors, and actuarial practice.

What Celent's Third Annual Survey Measured

Celent, a division of Oliver Wyman, has conducted its GenAI-oneers survey every year since 2023, making it the longest-running continuous benchmark of insurer GenAI adoption. The third annual edition, released in early 2026, covered global insurers across P&C and life segments, with separate regional editions for North America, EMEA, Latin America, and Asia-Pacific.

The survey's definition of "in production" captures insurers that have moved at least one GenAI use case beyond pilot or proof-of-concept into live operational deployment. This is an important distinction from surveys that measure "full-scale" or "enterprise-wide" deployment. A carrier running a single GenAI-powered claims summarization tool in one business unit qualifies as "in production" under Celent's methodology even if the rest of its operations remain untouched by generative AI.

That definitional scope is what produces the 48% figure and simultaneously explains why it can coexist with much lower numbers from other research firms measuring broader deployment criteria. Understanding the survey population and the bar for qualification is essential before drawing competitive strategy conclusions from the headline number.

Key findings from the third annual survey:

• 48% of global insurers now operate GenAI in at least one production use case, up from approximately 44% at mid-2025 and 28% a year earlier.
• 61% of respondents reported increased productivity or improved efficiency as a measurable positive impact from GenAI deployment.
• 22% of participating insurers plan to have an agentic AI solution in place by year-end 2026.
• Claims and underwriting remain the leading production deployment areas, followed by marketing and sales enablement.
• GenAI adoption has initially differentiated carriers "behind the scenes" through employee-facing use cases rather than customer-facing applications.

The Adoption Curve: 8% to 48% in Three Years

Celent's three consecutive annual surveys provide the clearest longitudinal dataset on insurer GenAI adoption available anywhere. The trajectory is striking:

Survey Year	Insurers in GenAI Production	Adoption Phase
2023 (first survey)	~8%	Innovators / early adopters
2024 (second survey)	~28%	Early majority
Mid-2025	~44%	Early majority (approaching threshold)
Q1 2026 (third survey)	48%	Entering late majority

The 20-percentage-point jump from 2023 to 2024 represented the steepest acceleration, which is consistent with Rogers's model: the early-majority cohort moves faster than innovators because they benefit from established vendor ecosystems, proven use cases, and lower implementation risk. The subsequent gains from 28% to 44% and then to 48% reflect the natural deceleration as the easier adopters have already converted and the remaining population includes carriers with more complex technology environments, stricter governance requirements, or smaller technology budgets.

Celent projects that GenAI adoption will firmly reach late-majority status during 2026, meaning more than half of global insurers will be running at least one production use case by year-end. If that projection holds, GenAI will have traversed the early-majority and late-majority phases in roughly two years, a pace that Celent itself characterizes as "exceptional in the annals of insurance" technology adoption.

Where Production Deployments Cluster: Claims, Underwriting, and Beyond

Both the Celent data and corroborating evidence from Evident AI's Q4 2025 deployment tracker confirm that production GenAI use cases concentrate in two areas of the insurance value chain, with a third area growing rapidly.

Claims management leads. Evident's analysis of publicly disclosed AI deployments found that claims management represented 37% of GenAI and agentic AI projects in Q4 2025, more than any other function. Within claims, the dominant production applications are document summarization, first-notice-of-loss intake automation, severity triage, and fraud detection signal extraction. Agentic AI systems, which can coordinate multi-step workflows autonomously, accounted for 21% of publicized deployments in Q4 2025, with 56% of those agentic implementations focused on claims. The pattern aligns with what we see in carrier earnings commentary: claims processing involves high-volume, semi-structured data that is well-suited to current GenAI capabilities.

Underwriting and pricing follow closely. Underwriting and pricing each accounted for 21% of next-generation AI use cases in the Evident data. These deployments include submission intake parsing, risk scoring augmentation, and pricing model copilots that help underwriters evaluate complex accounts faster. The Datos Insights ILTF 2026 survey separately confirmed this pattern, showing that carrier AI production deployments jumped from 37% to 61% in one year, with underwriting overtaking claims as the primary AI differentiator in the latest wave of implementations.

Customer engagement emerges as the third vector. A notable shift in Q4 2025 data: 36% of AI use cases involved direct policyholder interaction, compared with a historical average of just 7% (Evident). This five-fold increase signals that carriers are growing confident enough in GenAI outputs to deploy them in customer-facing contexts, from personalized policy recommendations to AI-assisted claims status updates. Celent's survey corroborates this, noting that while initial GenAI differentiation has been "behind the scenes" in employee-facing use cases, carriers will increasingly deploy customer-facing GenAI as accuracy and compliance controls improve.

Function	Share of GenAI/Agentic Deployments (Q4 2025)	Source
Claims management	37%	Evident AI
Underwriting and pricing	21% each	Evident AI
Customer engagement	21%	Evident AI
Direct policyholder interaction	36% (vs. 7% historical avg)	Evident AI

P&C insurers are the most active adopters by segment, accounting for 50% of all AI deployments in Q4 2025 according to Evident. Asset management and investment functions showed the fastest growth rate, tripling year-over-year from a smaller base.

The 22% Agentic AI Frontier

Celent's survey introduced a forward-looking dimension that was not present in the first two annual editions: agentic AI planning. Of the insurers surveyed, 22% stated they plan to have an agentic AI solution in place by year-end 2026. Celent projects that agentic AI adoption will rise from approximately 14% today to 70% by 2028.

The distinction between GenAI copilots and agentic AI systems is critical for understanding where the technology trajectory leads. Copilots respond to queries and assist human workers with specific tasks. Agentic systems can autonomously execute multi-step workflows, coordinate between different AI tools, and make decisions within defined parameters without human intervention at each step. For insurance, this distinction maps directly onto workflow complexity: a GenAI copilot can summarize a claims document, but an agentic system can read the document, extract relevant fields, cross-reference them against policy terms, run a severity estimate, flag potential fraud indicators, and route the claim to the appropriate handler, all as a connected sequence.

The 22% figure for planned agentic deployments by year-end 2026 indicates that roughly one in five insurers with existing GenAI production experience is preparing to move from task-level assistance to workflow-level automation. Celent's companion report, "Shedding Light on Agentic AI in Insurance," frames the current moment as still focused on advancing intelligent virtual assistants (IVAs) and AI copilots, with the transition to specialized autonomous agents supporting specific parts of the insurance value chain coming next.

Carrier Q1 2026 earnings commentary provides concrete evidence of this transition in real time. AIG's CEO Peter Zaffino reported GenAI deployment outcomes "beyond expectations," with the company's AIG Assist platform delivering a 30% quoting lift, 55% time-to-quote reduction, and 40% binding improvement across eight lines through multi-agent orchestration. Travelers launched an agentic AI claims assistant built with OpenAI that cut call center staffing by a third and achieved 50% straight-through processing. Allstate's proprietary ALLIE platform now codes one-third of its software and processes 10 million emails annually. These are not pilot results; they are production-scale deployments generating measurable financial outcomes.

Reconciling 48% "In Production" With 7% "At Full Scale"

The most important analytical question this data raises is how 48% of insurers can be "in production" (Celent) while only 7% have achieved "scalable success" (Sedgwick) and 68% of organizations have moved 30% or fewer experiments to full production (Deloitte). The numbers are not contradictory; they measure different things, and the gaps between them reveal where the industry actually stands.

Celent's 48% counts any insurer with at least one GenAI use case in live production. The bar is a single workflow operating beyond pilot status. This captures everything from a claims document summarization tool running in one regional office to an enterprise-wide multi-agent orchestration platform.

Sedgwick's 7% measures scalable success: AI operating at enterprise scale across lines of business and claim types, with measurable improvements in processing speed, accuracy, and cost. The gap between Celent's 48% and Sedgwick's 7% quantifies the distance between "we have something running" and "it works everywhere and we can prove it."

Deloitte's 68%/30% metric adds a third dimension. Even among organizations that are deploying GenAI, more than two-thirds have converted fewer than 30% of their experiments into full production. This suggests that many of the carriers counted in Celent's 48% have a limited number of live use cases surrounded by a larger pool of stalled pilots and proofs-of-concept.

A fourth data point reinforces this layered picture. An AM Best survey of approximately 150 rated carriers and MGAs, published in Insurance Journal in May 2026, found that 53% describe themselves as "cautious pacesetters rather than first movers," with only 20% reporting that their AI implementation is at an advanced stage. The top implementation challenges were data readiness (45%), security and privacy (43%), and legacy system integration (41%). Only 13% felt very confident in their ability to accurately measure AI return on investment.

Survey / Source	Metric	Finding
Celent (Q1 2026)	At least one GenAI use case in production	48% of global insurers
Sedgwick (March 2026)	Scalable AI success across enterprise	7% of insurers
Deloitte (Q3 2025)	Moved 30%+ of experiments to production	Only 32% of organizations
AM Best / Insurance Journal (May 2026)	Self-described advanced AI stage	20% of ~150 carriers/MGAs
Grant Thornton (2026)	Could pass independent AI governance review in 90 days	Only 24% of insurance leaders
EY (2026)	Early or full GenAI adoption	55% of insurers

The reconciliation reveals a consistent pattern: a large and growing share of carriers have initial GenAI production experience (48-55%), a smaller group has reached meaningful deployment depth (20-30%), and a much smaller group has achieved enterprise-scale success with demonstrable governance (7-24%). The Celent 48% and Sedgwick 7% are both accurate; they simply measure different points on the maturity continuum.

What Late-Majority Dynamics Mean for Competitive Strategy

The shift from early-majority to late-majority adoption changes the competitive calculus in ways that most trade press coverage of the Celent survey did not address. In Rogers's diffusion framework, the late majority adopts primarily because not adopting becomes riskier than adopting. The motivation shifts from "we could gain an advantage" to "we cannot afford to fall further behind."

For insurers, this dynamic plays out across three dimensions.

GenAI becomes table stakes, not a differentiator. When half of your competitors are running GenAI in production, the technology no longer provides a sustainable competitive edge to adopters. Instead, the absence of GenAI creates a growing operational gap for non-adopters. Consider the Evident data showing that 40% of insurers now report tangible business benefits from AI, with 77% of those benefits tied to productivity gains. If your competitors are processing claims 37% faster or quoting submissions 55% faster (AIG's reported figure), not having those capabilities becomes a measurable competitive disadvantage in cycle time, expense ratio, and customer experience.

The holdout penalty compounds over time. Late-majority dynamics create a compounding cost for carriers that have not yet deployed. Each quarter that passes, the carriers in production accumulate more training data, refine their models, optimize their workflows, and build organizational muscle memory around AI-assisted operations. The carriers still in pilot or proof-of-concept are not standing still; they are falling further behind a moving target. Morgan Stanley's projection that AI will generate $9.3 billion in P&C operating income by 2030 through 200 basis points of expense ratio compression assumes broad adoption. Carriers that miss the adoption window will face the cost of late implementation (higher vendor prices as demand peaks, scarcer AI talent, less proprietary training data) while their competitors are already capturing those savings.

The vendor ecosystem shifts toward late-majority needs. As adoption crosses the midpoint, AI vendors will increasingly design products for the constraints that held the late majority back: legacy system integration, lighter governance requirements, faster time-to-value, and lower technical prerequisites. Guidewire's ProNavigator, Duck Creek's Agentic AI Platform, and Cytora Autopilot are early examples of this shift. These products embed AI directly into core systems that carriers already operate, removing the integration overhead that blocked many carriers from moving beyond pilot. For actuaries evaluating build-versus-buy decisions, this vendor maturation changes the calculus. Embedded AI from existing core system vendors may offer a faster, lower-risk path to production than custom implementations.

Carrier-Level Evidence: Q1 2026 Deployment Disclosures

Carrier earnings calls and SEC filings in Q1 2026 provide ground-level evidence that corroborates the survey data. From tracking these disclosures over the past two years, the shift from vague "AI initiative" language to specific production metrics is itself a signal that deployments have matured.

AIG reported GenAI results "beyond expectations" on its Q1 2026 earnings call. CEO Peter Zaffino described a "massive shift in the company's ability to process submissions without additional human resources." AIG's Atlanta innovation hub, which opened in 2026 with more than 600 new roles spanning underwriting, claims, operations, data engineering, and AI, signals the company's commitment to scaling from isolated use cases to enterprise deployment. The company's Palantir Foundry-powered multi-agent orchestration system coordinates knowledge, adviser, and critic agents across 370,000 submissions, with 30-hour autonomous cycles and an 88% AI-adjuster agreement rate on fraud detection disclosed on the Q1 call.

Travelers launched an agentic AI claim assistant developed with OpenAI for live auto damage claims in February 2026. The system processes calls in real time, achieving 50% straight-through processing with 66% customer adoption. The company consolidated call centers from four facilities to two, backed by a $1.5 billion annual technology budget and 20,000 employees using AI tools. Separately, Travelers deployed 10,000 Anthropic-powered AI assistants to its engineering and data science workforce.

Chubb appointed Kevin Rampe as its first Global Claims Officer in April 2026, structurally integrating claims operations across 54 countries under one mandate. CEO Evan Greenberg has consistently described an 85% claims automation target, with Q1 2026 results showing an 84% combined ratio and $1.79 billion in underwriting income. Chubb's approach differs from AIG's agent orchestration model: it emphasizes incremental, function-by-function automation rather than full-workflow agentic systems.

Progressive continues to invest in its proprietary machine learning pricing models and telematics-driven underwriting, with 21 million active Snapshot policyholders providing continuous driving behavior data. While Progressive's AI disclosures are less granular than AIG's or Travelers', its record media spend and ML-driven pricing precision contribute to consistent combined ratio outperformance.

Allstate built ALLIE, a proprietary agentic AI platform that now generates one-third of the company's software code, processes 10 million emails annually, and has cut billing escalations by 50%. Allstate is testing AI-powered direct sales in three states, making it one of the few carriers deploying GenAI in a customer-acquisition context rather than purely in operational efficiency.

The Governance Constraint on Scaling

If 48% of insurers are in production but only 7% have achieved scale, the obvious question is what blocks the other 41%. The evidence points overwhelmingly to governance infrastructure, not technology capability, as the binding constraint.

Grant Thornton's 2026 AI Impact Survey of 100 insurance executives found that 44% cite governance or compliance challenges as contributors to AI project failures. Only 24% expressed confidence they could pass an independent AI governance review within 90 days. The AM Best survey corroborates this, finding that 45% of carriers identify data readiness as a top challenge and 41% cite legacy system integration barriers.

The NAIC's regulatory trajectory compounds the governance challenge. Twenty-three states plus Washington, D.C. have now adopted the 2023 AI Model Bulletin in some form. The NAIC's Spring 2026 meeting advanced a multi-state pilot program using an AI evaluation tool that systematically assesses carrier AI systems, data sources, governance practices, and high-risk use cases. A proposed vendor registry would give regulators visibility into the third-party tools carriers rely on.

For carriers moving from initial GenAI production to enterprise scale, governance is not merely a compliance checkbox. It is the architectural layer that determines whether AI tools operating across different business functions, lines of business, and vendor platforms can be monitored, validated, and defended during regulatory examination. As Celent's own survey notes, the path to late-majority adoption requires building governance infrastructure in parallel with technical deployment, not after the fact.

Why This Matters for Actuaries

The late-majority shift and the 48%/7% gap have specific implications for actuarial work across pricing, reserving, and enterprise risk management.

Expense ratio assumptions need a bifurcated framework. For the 7% of carriers at enterprise AI scale, expense ratio improvements of 100-200 basis points are credible and supported by carrier disclosures (Chubb's 1.5-point automation savings, Morgan Stanley's 200-basis-point projection). For the remaining 93%, prospective AI-driven expense improvements remain speculative. Pricing actuaries should model expense ratio trajectories differently for carriers in the scalable tier versus those still in early production. ASOP No. 29 requires that expense assumptions in rate filings be supported by evidence, not aspirational technology plans.

Loss adjustment expense patterns will shift unevenly. Claims AI deployed by carriers in the 48% cohort should begin producing measurable changes in allocated loss adjustment expense ratios within two to four quarters. The Evident data showing 37% of GenAI deployments in claims management suggests that LAE is the expense category most directly affected. Reserving actuaries should monitor whether FNOL acceleration and AI-assisted triage are compressing development patterns in early reporting periods, particularly for personal lines where claims volume and standardization make AI integration most straightforward.

Model governance scope expands under ASOP No. 56. The NAIC's AI evaluation tool and proposed vendor registry directly affect actuaries responsible for model governance. When GenAI tools from multiple vendors feed into underwriting or claims decisions, the actuary's ASOP No. 56 responsibility extends to the entire chain of AI tools, including vendor-provided models, third-party data enrichment layers, and agentic orchestration platforms. The vendor fragmentation that AM Best identifies (68% of carriers source AI from third parties, but only 18% consider third-party model risk a challenge) suggests that most carriers have not yet scoped their model governance obligations to match their actual AI deployment footprint.

The 22% agentic AI planning figure signals a reserving methodology question. Agentic AI systems that autonomously handle multi-step claims workflows will change the distribution of case reserve accuracy. If an agentic system consistently produces earlier and more accurate initial reserves (as AIG's 88% agreement rate on fraud detection suggests), the historical development patterns that underpin chain-ladder and Bornhuetter-Ferguson methods may shift. Reserving actuaries at carriers deploying agentic claims systems should track whether AI-era development triangles show materially different patterns than pre-AI triangles, and document any divergence for the Statement of Actuarial Opinion.

Competitive benchmarking requires survey literacy. Actuaries participating in strategic planning or rate adequacy discussions should understand the methodological differences between the 48% (Celent), 7% (Sedgwick), 55% (EY), and 20% (AM Best) figures. Citing a single number without qualifying the survey scope and adoption definition will produce misleading conclusions about where a given carrier stands relative to its peers.