EXL: 76% of Insurers Think They Lead AI, Only 6% Do

Seventy-six percent of insurers rate themselves ahead of their competitors on artificial intelligence. Six percent actually qualify as leaders, and seventy-two percent land in the middle follower tier, the highest follower concentration EXL's third annual Enterprise AI Study found in any industry it surveyed (Carrier Management, August 12, 2026). The gap is widest exactly where pricing and reserving decisions get made.

That seventy-point spread between self-perception and EXL's classification is not a rounding error in a satisfaction survey. It is the headline finding of a study built specifically to separate stated confidence from measured capability across six to eight core business functions, and insurance posted the widest gap of any of the five industries EXL surveyed: healthcare, life sciences, retail, utilities, and insurance itself (EXL, 2026 Enterprise AI Study, August 2026). actuary.info has covered the mechanics of the industry's AI adoption problem from two angles this year, the pilot-to-portfolio bottleneck and the outcome-measurement gap identified by Capgemini. EXL's study adds a third, more uncomfortable layer: insurers are not just failing to scale pilots or measure ROI. Most of them do not accurately know where they stand to begin with.

A Wider Gap Than the Cross-Industry Average

EXL's study surveyed 322 executives across the five industries and classified respondents into three tiers, leaders, followers, and laggards, based on measured capability rather than self-report (EXL, 2026 Enterprise AI Study, August 2026). Across all five industries combined, 76% of executives believe they are ahead of competitors, but only 10% qualify as leaders by EXL's criteria, roughly matching capability in six to eight core functions with enterprise-wide data access and agentic AI embedded in workflow redesign. Insurance's leader rate came in at 6%, four points below the cross-industry average, while its self-belief held at the same 76% every other industry reported.

Run those two figures together and insurance is not simply overconfident in line with everyone else. It is both further behind the average leader threshold and equally convinced it is ahead, which is a worse combination than either fact alone. A sector that knew it was lagging could at least prioritize catching up. A sector that believes it already leads, while sitting near the back of the pack on the metric that defines leadership, has no internal signal telling it to change course. Digital Insurance's coverage of the same release quotes Rup Goswami, EXL's Insurance Growth Office Lead, framing where the industry's attention has actually shifted: "The industry has stopped asking whether AI works and started asking how fast they can scale it" (Digital Insurance, August 2026). Scaling intent and scaling capability are not the same claim, and EXL's tiering is designed to catch the difference.

Where the Gap Concentrates: Actuarial and Underwriting Trail the Front Office

The study's functional breakdown is where the maturity gap stops being an abstract survey artifact and becomes a pricing and reserving problem. Forty-six percent of insurers report agentic AI advancing fastest in actuarial, underwriting, and pricing workflows, putting that cluster behind risk management at 54% and roughly level with customer experience at 45% (Carrier Management, August 12, 2026). Fraud detection and customer servicing both lead the application list at 54% adoption, with claims close behind at 42%. The functions with the most direct line to statutory reserves and filed rates, in other words, are not where insurers have pushed AI hardest. They are running roughly even with, or behind, functions where a model error produces an inconvenient customer interaction rather than a mispriced book of business.

That ordering matters for how skeptically a reader should treat AI language in carrier disclosures. A 10-K risk factor or an earnings-call reference to "AI-driven underwriting" can describe a genuinely deployed pricing engine, a single pilot cohort, or a chatbot bolted onto the front end of a legacy rating system, and EXL's functional breakdown gives no reason to assume the strongest case by default. Forty-six percent of insurers report full deployment in actuarial and underwriting specifically, which means a slim majority do not, even as 96% now list scaling AI as a high organizational priority, up from 86% in 2025 (Carrier Management, August 12, 2026). Priority and deployment are two different lines on the same chart, and the actuarial function is where they diverge most.

Metric	Insurance	Cross-industry average
Believe they are AI leaders	76%	76%
Actually classified as leaders	6%	10%
Classified as followers	72% (highest of any industry)	–
Agentic AI most advanced in actuarial/underwriting/pricing	46%	–
Cite data as an obstacle to AI success	92%	–
Rate own data management as mature	24%	–

Source: EXL 2026 Enterprise AI Study (August 2026), as reported by Carrier Management (August 12, 2026) and Digital Insurance (August 2026).

The data-readiness figures explain part of why actuarial and underwriting adoption trails the front office. Ninety-two percent of insurers cite data as an obstacle to AI success, the highest share attributing the barrier to data silos of any industry EXL surveyed, and only 24% rate their own data management as mature (Carrier Management, August 12, 2026). Actuarial and underwriting models draw on the deepest, most fragmented data estate a carrier holds, historical loss triangles, policy administration records, claims notes, third-party scoring feeds, spread across systems that predate the current AI build-out by decades. Fraud detection and customer servicing tools can often bolt onto a narrower, cleaner data slice. The functions with the least clean data to work from are the functions where the highest-stakes decisions get made, which is a structural reason the actuarial gap does not close on its own timeline.

Reading Carrier AI Disclosures Against a Self-Delusion Baseline

The practical use of EXL's finding for anyone reading rate filings, 10-Ks, or investor decks is as a discount rate. If three in four insurers believe they lead on AI and only one in twelve actually clears EXL's leader bar, then a carrier's own characterization of its AI maturity, made in a proxy statement, an earnings call, or marketing collateral aimed at reinsurers and rating agencies, carries a documented industry-wide bias toward overstatement. That bias is not necessarily deliberate; EXL's leaders-versus-laggards data suggests it tracks data-management self-perception closely, with 91% of leaders rating themselves ahead on data maturity versus 61% of laggards making the identical claim about themselves (Digital Insurance, August 2026). Companies that are actually behind are not lying; they are using an internal reference frame calibrated to their own history rather than to the market's actual leaders.

This is precisely the discrepancy the NAIC's newest oversight mechanism is built to test. Twenty-five states had adopted the NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers as of mid-2026, requiring a written AI Systems Program with board-level accountability, documented model validation, and oversight of third-party AI where the insurer remains legally responsible for outcomes (NAIC, AI Model Bulletin implementation tracker, 2026). The NAIC's companion AI Systems Evaluation Tool, running as a twelve-state multistate pilot from January through September 2026, gives examiners a structured framework to test an insurer's self-reported AI governance maturity against actual documentation during market conduct exams. actuary.info's earlier coverage of the bulletin's rollout laid out the state-by-state adoption mechanics; EXL's study supplies the empirical reason examiners have grounds to be skeptical of what they will find in the self-attestation column. A carrier whose AIS Program narrative describes itself as an AI leader, without matching third-party assessment or documented validation results, is describing the industry norm EXL just measured, not an exception regulators should take at face value.

The Governance Debt the 72% of Followers Are Carrying

The follower tier is where a second EXL finding compounds the disclosure-skepticism point into a balance-sheet one. EXL's leader-versus-laggard profile shows leaders achieving 40% more revenue growth and 37% more cost reduction than laggards in the use cases where AI is actually applied (Carrier Management, August 12, 2026), a gap wide enough that boards have pushed hard on scaling: 96% now rank it a high priority. That pressure lands on the 72% follower cohort as an incentive to buy capability quickly rather than build and validate it, and buying quickly is exactly where governance debt accumulates. The CAS's AI Primer, published in April 2026, frames the risk directly for practicing actuaries: responsible deployment requires validation, human-in-the-loop review, and documentation showing AI outputs are "appropriate, explainable, and aligned with business and regulatory expectations," regardless of whether the underlying model was built in-house or licensed from a vendor (Casualty Actuarial Society, AI Primer, April 2026).

A follower carrier that licenses a vendor's pricing or triage model to close the scaling gap fast inherits the NAIC bulletin's third-party oversight obligation on a compressed timeline, with less internal capability to interrogate what it purchased than a carrier that built the equivalent tool over several validation cycles. This is the same substitution dynamic actuary.info has tracked in EXL's own second-quarter results, where AI-led vendor revenue grew to 61% of the company's total while the labor-intensive services line it replaced contracted, a trade that shifts a carrier's actuarial documentation burden from processing volume to validating a vendor's model provenance. The urgency the study's own scaling-priority numbers create, 96% ranking it high versus 86% a year earlier, is the same urgency that makes the shortcut of buying without validating more likely, not less, across a follower population EXL just sized at 72% of the industry.

What a Governance-Readiness Review Actually Finds

The gap between EXL's follower classification and a carrier's own self-report is not evenly distributed across a follower company's AI portfolio. In practice, the shortfall concentrates in three places a governance-readiness review consistently surfaces: model inventories that list vendor tools by name but not by validation status, documentation that describes what a model does without evidence of testing for the unfair-discrimination outcomes the NAIC bulletin requires, and change-management logs that stop updating once a vendor pushes a model refresh the carrier did not independently retest. None of those gaps show up in an executive survey response, which asks whether AI capability exists, not whether its governance paperwork would survive a market conduct exam. That is precisely the layer EXL's tiering cannot see and a state examiner applying the AI Systems Evaluation Tool is built to test directly against an insurer's own AIS Program file.

The economics of closing that gap favor acting before an exam finds it rather than after. A carrier that commissions an internal or third-party validation review of its actuarial and underwriting AI stack, cataloguing which models are vendor-sourced, which have documented bias testing, and which lack a defined revalidation cadence, is doing on its own schedule what a twelve-state pilot examiner will otherwise do on the regulator's schedule. For the 72% of insurers EXL classifies as followers, most of whom are simultaneously reporting AI scaling as a top-three board priority, that review is cheaper as a planned governance exercise than as a remediation plan filed in response to a market conduct finding. The gap between the two is largely a question of who does the counting first.

What the Numbers Do Not Say

EXL's tiering methodology is proprietary; the study does not publish the specific rubric weightings that separate a leader from a follower, and 322 surveyed executives across five industries is a meaningful but not exhaustive sample of the US insurance market. The 62% pilot-to-production rate EXL reports for insurance, the highest of any industry it studied, sits in some tension with the 6% leader classification, since scaling a pilot into production is a necessary but not sufficient condition for the enterprise-wide capability EXL's leader tier requires. Read together, the two figures suggest insurers are unusually good at graduating individual pilots but unusually poor at compounding those graduations into the six-to-eight-function breadth that separates a leader from a follower, a distinction a single percentage cannot carry alone.

Further Reading on actuary.info

• Insurance AI Hits the Pilot-to-Portfolio Wall - The adjacent finding on why individual AI pilots stall before reaching enterprise scale.
• 42% of P&C Insurers Never Measured AI Outcomes: Inside the Capgemini Measurement Gap - A companion data point on why insurers cannot reliably judge their own AI performance.
• EXL's AI-Led Revenue Hits 61% While Its Legacy Operations Line Shrinks - How the same vendor's growth reshapes the build-versus-buy calculus this piece's governance-debt section extends.
• AI Regulation in Insurance 2026: The NAIC Model Bulletin, State Adoption, and the Federal Preemption Battle - The full state-by-state mechanics of the oversight regime discussed above.
• When Your AI Governance Controls Are Vendor-Patented - What follower carriers risk when governance tooling itself comes from the same vendor supplying the model.