One in Five Insurers Deploys AI While Cutting Training Budgets: Inside the Covenir Workforce Readiness Gap

From reviewing carrier staffing models across four annual deployment cycles, the pattern is now clear: budget lines for AI tooling scale linearly while training allocations flatline or decline. The Covenir 2026 Insurance Operations Leaders Trends Report, released June 9 and based on a February-March survey of 152 U.S.-based insurance operations decision-makers at carriers, MGAs, and insurtechs, puts hard numbers behind that pattern. Seventy percent of organizations now run AI in live operations, up from 58% a year ago. Yet 20% are simultaneously slashing training budgets, and only 7% have taken steps to protect them. The report reveals an industry that is deploying faster than it is preparing its workforce to use what it deploys.

Most coverage of the Covenir survey headlines the 70% deployment figure and moves on. This article goes deeper into the data, focusing on where the training gap hits customer-facing operations first: the 42% of carriers reporting brand-promise breakdowns at First Notice of Loss, the 91% C-suite strain rate that signals executive capacity is already consumed, and the 54% of advanced AI adopters planning headcount cuts that will remove the experienced workers needed to supervise AI handoffs.

What the Covenir Survey Covers

The Covenir survey is now in its second year, giving the data a year-over-year baseline that single-edition surveys lack. The 152-respondent sample is drawn exclusively from U.S.-based insurance operations decision-makers, which narrows the population but sharpens the signal: these are the people who manage the teams, workflows, and budgets where AI meets day-to-day insurance operations.

The respondent pool spans carriers, managing general agents, and insurtechs, covering companies from fewer than 200 employees to more than 1,000. That size distribution matters because the survey reveals stark differences in AI readiness by organizational scale. The smallest carriers (1 to 199 employees) are the most likely to lack confidence in their ability to deliver on brand promise, while mid-sized carriers (500 to 999 employees) report the highest budget growth alongside the highest team strain.

Covenir fielded the survey between February and March 2026, which means the data captures the state of AI deployment after the major carrier announcements of late 2025 and early 2026, including Travelers' expansion to 20,000 Anthropic users, Allstate's ALLIE agentic platform scaling, and Allianz's global audit-ready AI partnership, but before the mid-year correction in AI spending that several consulting firms have projected. The timing makes this a snapshot of the industry at peak deployment velocity.

The 70/20 Paradox: Deploying AI While Defunding Readiness

The headline numbers tell a contradictory story. Seventy percent of organizations have AI running in production, a 12-point increase from 58% in 2025. That acceleration is consistent with the Celent finding that 48% of global insurers run GenAI in production and the broader pattern of late-majority adoption pushing deployment rates past the 50% threshold in 2026.

But deployment without readiness is not adoption. It is exposure. Twenty percent of organizations are cutting training budgets even as they expand AI tooling. Only 7% have taken deliberate steps to protect training allocations. The remaining 73% fall into the middle: neither cutting nor protecting, letting training budgets drift as AI spending claims an increasing share of the technology line.

David Squibb, Covenir's president and CEO, described the tension in the report's release: "Insurance operations leaders are doing something genuinely hard right now. They're staying optimistic while absorbing real pressure, deploying AI while cutting the training budgets that make it work, and sitting on operational intelligence they know is valuable but haven't yet built the infrastructure to act on."

This pattern is not unique to insurance. Fortune reported in March 2026 that global AI spending is projected to rise 44% this year while training budgets are expected to grow just 5%. Nearly 75% of knowledge workers globally now use AI at work, yet 60% say they have not received formal training on the tools they use daily. The Covenir data confirms that insurance operations follow the cross-industry pattern, with one critical difference: insurance operations involve regulated decisions that affect policyholder outcomes, making untrained AI use a compliance risk as well as a productivity problem.

Advanced Adopters Cut Deepest: The 54% Headcount Signal

The most striking workforce finding in the Covenir data is the correlation between AI maturity and headcount reduction plans. Among advanced AI adopters, defined as organizations running AI across multiple operational functions, 54% plan headcount cuts in 2026. That rate is five times higher than the 11% among less mature peers still in early deployment stages.

This is not a broad automation displacement story. It is a concentration effect. The carriers furthest along the AI deployment curve are the ones removing the most human capacity from their operations. The implicit logic is that AI has progressed far enough in these organizations to replace human workflows. The Covenir data suggests that logic is premature.

Consider the sequence: a carrier deploys AI across claims, underwriting, and customer service. The AI handles routine transactions with acceptable accuracy. The carrier sees an opportunity to reduce headcount in those functions. But the experienced staff being cut are the same people who handle exceptions, train new hires, supervise AI outputs, and intervene when the AI produces incorrect results. Removing them while simultaneously cutting training budgets for remaining staff creates a compounding readiness deficit.

The Acrisure headcount reduction of 2,250 positions, the largest disclosed AI-driven workforce action in insurance distribution, illustrates the pattern at the broker level. The Covenir data shows a similar dynamic emerging at the carrier operations level, but distributed across dozens of mid-to-large carriers rather than concentrated in a single announcement.

AI Maturity Level	Headcount Cut Plans	Training Budget Status	Implied Risk
Advanced (multi-function AI)	54%	20% cutting, 7% protecting	Removing supervisory capacity while underfunding remaining staff
Intermediate (single-function AI)	~25% (implied)	Budget drift, no protection	Scaling deployment without proportional training
Early (pilot/exploration)	11%	Budget stable, lower AI spend	Limited exposure but limited preparedness

91% Strain at the Top: The Executive Capacity Paradox

The Covenir survey asked respondents at different organizational levels whether their teams are more stretched than they have ever been. The results reveal a strain gradient that intensifies with seniority. Among C-suite executives, 91% report their teams are at maximum strain. That figure is nearly double the rate among managers and individual contributors. Across all respondent levels, 65% report teams more stretched than ever.

The 91% C-suite figure deserves careful interpretation. It does not mean that 91% of insurance operations are failing. It means that 91% of the executives responsible for managing AI deployment, workforce transitions, and operational performance simultaneously feel they have no remaining capacity to absorb additional change. This is a leading indicator, not a lagging one. When the people responsible for steering the AI transition have no bandwidth left, the quality of their decisions about what to deploy, when to train, and where to cut deteriorates.

Mid-sized carriers (500 to 999 employees) present the most concerning profile in the Covenir data. These organizations report the highest budget growth, suggesting they are investing aggressively in AI and operational expansion. They also report 83% team strain, nearly as high as the C-suite rate. The combination of aggressive investment and near-maximum strain means these carriers are scaling into new capabilities without the organizational slack needed to manage the transition well.

The Deloitte 2026 Global Insurance Outlook corroborates the pattern: 90% of insurance executives agree on the urgent need to reinvent employee value propositions for human-machine collaboration, but only 25% have taken tangible action. The gap between recognizing the problem and acting on it is exactly what produces a 91% strain rate. Executives know what needs to happen; they do not have the capacity to make it happen while simultaneously managing everything else.

Where Brand Promise Breaks: The FNOL Exposure Point

The Covenir survey asked where brand-promise breakdowns happen most frequently in insurance operations. The answer, at 42%, is First Notice of Loss. FNOL is the first interaction a policyholder has with their carrier after experiencing a loss, and it is the function most exposed to the collision between AI deployment and workforce strain.

FNOL sits at the intersection of three pressures the Covenir data identifies. First, it is a high-volume function where AI automation is attractive because of the potential to reduce average handle times and accelerate triage. Second, it is a customer-facing function where errors are immediately visible to policyholders and directly affect their perception of the carrier. Third, it depends on experienced staff who can recognize when an AI-assisted triage is wrong, when a claimant's situation requires human judgment, or when the standard workflow should be overridden.

When carriers deploy AI at FNOL while cutting training budgets and reducing headcount among experienced staff, the failure mode is predictable. AI handles the routine claims efficiently, but the exceptions, the complex losses, the claimants in distress, the situations requiring nuanced judgment, fall through because the remaining staff lack the training to manage AI handoffs and the experienced staff who would have caught problems have been cut.

The NAIC's intensifying focus on AI claims handling reflects exactly this concern. State regulators are examining whether AI-assisted claims decisions meet the same standards as human decisions, and whether carriers have adequate human oversight at the points where AI and policyholder interests intersect. The Covenir data on FNOL brand-promise breakdowns provides empirical support for the regulatory premise: the point of greatest AI exposure is the same point where customer experience is most at risk.

The remaining brand-promise breakdown locations in the Covenir data, while not as concentrated as FNOL, follow a pattern. Functions that combine high volume, customer visibility, and AI deployment are the ones where trained human supervision matters most and where the training deficit is felt first.

Ninety percent of leaders say they are extremely or very confident that their operations deliver on brand promise. Yet 42% identify FNOL as the point where that promise breaks most often. The gap between confidence and performance is itself a signal that leadership may not have visibility into how the training deficit is manifesting at the operational level.

The Data Translation Gap: 47% Cannot Act on Their Own Intelligence

The Covenir survey reveals a second structural gap that compounds the training deficit. Eighty-one percent of respondents say operational insights are critical to business success. Yet 47% either do not use their operational data at all or cannot translate it into actionable decisions. They are collecting dashboards they cannot read and building data pipelines that terminate in reports no one acts on.

This finding connects directly to the Capgemini measurement gap, where 42% of P&C insurers reported never measuring AI outcomes. The Covenir data extends the problem beyond AI-specific measurement to operational data broadly. Carriers are investing in data infrastructure and AI tools while lacking the organizational capacity to convert either into decisions.

Carrier Management reported on June 3, 2026, that "AI is not creating new data problems. It is exposing existing ones," citing research showing 62% of IT leaders identify data quality as their single biggest challenge in managing data across hybrid environments. The Covenir data confirms that this is not merely a technology problem. The 47% who cannot translate data into decisions include organizations that have the data and the tools but lack the trained personnel, established processes, and decision-making frameworks to use them.

For actuaries, the data translation gap has direct implications. Actuarial work products depend on operational data flowing from claims, underwriting, and customer service systems. When 47% of carriers cannot act on their own operational data, the data feeding actuarial models may be incomplete, inconsistently defined, or poorly maintained, not because the systems are broken, but because the organizational capacity to manage data quality was never built alongside the technology investment.

Outsourcing the Problem: 29% Expect Partners to Deliver AI Readiness

Twenty-nine percent of respondents expect their outsourcing partners to embed AI solutions into the services they provide. This figure signals that a meaningful share of carriers have decided, implicitly or explicitly, that they will not build internal AI readiness. Instead, they will externalize the problem to third-party operations providers.

This strategy has precedent. Carriers have outsourced claims administration, policy servicing, and call center operations for decades. Adding an expectation that the outsourcing partner will also manage AI integration is a natural extension of existing relationships. But it introduces a new category of dependency: the carrier is not just outsourcing labor; it is outsourcing the AI competency that will increasingly determine operational performance.

The risk is that carriers who externalize AI readiness lose the institutional knowledge needed to evaluate, govern, and course-correct their AI operations. When the outsourcing partner's AI makes an error in claims triage or underwriting support, the carrier lacks the internal expertise to identify the error, diagnose the cause, or specify the fix. The carrier becomes dependent on the partner's AI governance, which may or may not meet regulatory expectations.

Former XL Group CEO Mike McGavick, now non-executive chairman of mea Platform, quantified the operational waste at stake in a Carrier Management feature published June 5: "$32 billion annual sector inefficiency loss," citing Accenture research. McGavick noted that underwriters spend 35 to 40 percent of their time on non-core administrative tasks. His blunt assessment of the industry's AI readiness: "We're genuinely at the beginning. When you watch your companies make announcements about AI work, I'm sure you snicker a little because you know it isn't that much yet."

McGavick's perspective frames the 29% outsourcing figure as rational but risky. Carriers that cannot build internal AI readiness may be better off partnering with organizations that can. But the $32 billion inefficiency will not be recaptured by carriers that have externalized the capability needed to capture it.

Cross-Referencing Against Industry Benchmarks

The Covenir data does not exist in isolation. Benchmarking its findings against other 2026 surveys helps calibrate whether the training gap it identifies is a U.S. operations-specific phenomenon or part of a broader structural pattern.

Celent (April 2026): Forty-eight percent of global insurers run GenAI in production, with 22% planning agentic AI deployment by year-end. The Covenir figure of 70% is higher because it measures all AI (not just generative) and focuses exclusively on U.S. operations where adoption rates lead global averages. The two surveys are directionally consistent: deployment is accelerating across the industry.

AutoRek (early 2026): A survey of 250 insurance managers found that 82% believe AI will dominate the industry's future, but only 14% have fully integrated AI into financial operations. The average insurer feeds 17 separate data sources into premium processes, and 14% of operational budgets go to fixing manual process errors. This supports the Covenir finding that deployment velocity outpaces organizational integration.

Insurance Business (May 2026): GlobalData reported 63,293 active job postings for AI expertise in insurance in 2025, a 50.9% increase over 2024 and the highest annual total on record. Spike Lipkin, Willis's Chief AI Officer, noted that "many organisations are moving forward without fully understanding the systems they rely on." The hiring surge alongside the training deficit the Covenir survey documents suggests carriers are trying to buy AI talent externally rather than develop it internally.

Deloitte 2026 Global Insurance Outlook: Ninety percent of insurance executives agree on the need to reinvent workforce strategies for human-machine collaboration. Only 25% have taken tangible action. This 90/25 intention-action gap mirrors the Covenir 70/7 deployment-protection gap. The industry recognizes the workforce readiness problem at the executive level but has not allocated the resources to address it.

Fortune / Gallup (March 2026): Gallup estimates that disengaged and stressed workers cost the global economy nearly $9 trillion annually. Applied to insurance operations where the Covenir survey shows 91% C-suite strain and 65% overall strain, the productivity losses from an overstretched workforce may offset or exceed the efficiency gains from AI deployment.

Survey	Sample	AI Deployment Rate	Workforce Readiness Signal
Covenir 2026	152 U.S. ops leaders	70% in production	20% cutting training; 91% C-suite strain
Celent 2026	Global insurers	48% GenAI in production	22% planning agentic AI with limited governance
AutoRek 2026	250 insurance managers	14% fully integrated	14% of budget fixing manual errors
Capgemini 2026	344 P&C executives	60% in pilot/POC	42% never measured AI outcomes
Deloitte 2026	Global insurance execs	N/A	90% recognize need; 25% acting

The Optimism Disconnect

One of the more puzzling findings in the Covenir data is that 88% of respondents are optimistic about the insurance industry's future, up from 83% in 2025. Sixty-eight percent of organizations saw budget increases. The optimism coexists with the strain, the training cuts, and the brand-promise breakdowns documented elsewhere in the same survey.

This is not cognitive dissonance. It is a product of the industry's financial position. U.S. P&C insurers posted their strongest Q1 underwriting performance in 25 years. Premium growth, investment income, and surplus levels provide a financial cushion that supports executive confidence. The optimism is about the industry's macro trajectory; the strain is about the micro-level operational challenge of managing the AI transition while running the business.

For actuaries, the optimism disconnect matters because it affects how organizations prioritize training and readiness investments. When executives are optimistic about the macro trajectory, they are less likely to treat workforce readiness as an urgent capital allocation priority. The 20% cutting training budgets are, in many cases, executives who believe the AI tools will work well enough without extensive training, that the financial cushion will absorb any short-term service quality issues, and that the market will reward AI deployment regardless of workforce readiness.

The 42% FNOL brand-promise breakdown rate suggests that assumption is already being tested.

Why This Matters for Actuaries

The Covenir workforce readiness gap has specific implications for actuarial practice across pricing, reserving, and operational risk.

Expense ratio assumptions in rate filings. Carriers planning headcount cuts in conjunction with AI deployment will project expense ratio improvements in their rate filings. Pricing actuaries need to evaluate whether the projected savings account for the transition costs: increased error rates during the training deficit period, potential regulatory penalties from inadequate AI oversight, and the productivity losses associated with 91% executive strain. The Morgan Stanley projection of 200 basis points of AI-driven expense savings assumes smooth deployment. The Covenir data suggests the transition will be rougher than the projections imply, particularly for the 54% of advanced adopters cutting headcount before workforce readiness catches up to deployment velocity.

Claims reserve development in AI-affected lines. The 42% FNOL brand-promise breakdown rate signals that claims handling quality may be degrading at the intake stage for a material share of carriers. Reserving actuaries should monitor whether FNOL error rates are producing downstream effects: delayed severity recognition, misclassified claims, or longer cycle times on the subset of claims that require human reclassification after AI triage. If FNOL quality degrades, initial case reserves may be less reliable, affecting IBNR estimates and development factor selections.

Operational risk modeling. The combination of 91% C-suite strain, 20% training budget cuts, and 54% headcount reduction among advanced adopters creates a correlated risk that enterprise risk models may not capture. Traditional operational risk modeling treats workforce capacity and technology deployment as separate risk factors. The Covenir data shows they are mechanically linked: reducing workforce capacity while increasing technology deployment produces a single compound risk, the inability to supervise AI operations at the scale required for quality and compliance.

Vendor dependency risk. The 29% outsourcing AI readiness to partners creates a concentration of operational capability outside the carrier's direct control. Actuaries evaluating counterparty risk in outsourcing arrangements should consider whether the partner's AI governance meets the regulatory standards the carrier is obligated to maintain. The regulatory pushback against automated claims decisions applies regardless of whether the AI is operated by the carrier or by a third-party service provider.

Data quality for actuarial models. The 47% data translation gap means that nearly half of carriers are generating operational data they cannot interpret or act on. Actuarial models that depend on claims, underwriting, or customer interaction data from these carriers face a quality risk that is not visible in the data itself. The data may be complete and properly formatted while being inconsistently generated, irregularly validated, or produced by workflows where AI and human processes interleave in ways that introduce systematic biases the carrier cannot detect because it lacks the trained personnel to identify them.

The Path Forward: Training as Infrastructure, Not Overhead

The Covenir data points to a reframing that the 7% of carriers protecting training budgets have already made: treating training as operational infrastructure rather than discretionary overhead. When a carrier invests $50 million in AI tooling and $0 in training the workforce to use it, the $50 million is not a technology investment. It is a technology purchase with no plan for returns.

Three specific actions emerge from the data for carriers that want to close the readiness gap before the AI J-curve becomes a permanent cost structure rather than a transitional phase.

Protect training budgets proportionally. For every dollar of AI tooling spend, allocate a defined percentage to training. The Covenir data shows that the current ratio is implicitly 93/7, with only 7% protecting training. The Capgemini data on trailblazers, which identifies 72/28 as the industry average and shows that top performers invest materially more in change management, provides a benchmark. Even moving to an 85/15 split would triple the training investment that most carriers currently make.

Sequence headcount reductions after readiness thresholds. The 54% of advanced adopters planning headcount cuts should define readiness gates: measurable criteria for when remaining staff have achieved the competency levels needed to supervise AI operations at the reduced headcount level. Cutting before those thresholds are met is how the FNOL brand-promise breakdown rate reaches 42%.

Build internal data translation capability. The 47% data gap will not close by purchasing more dashboards. It closes when carriers invest in the analytical staff, both actuarial and operational, who can interpret data and convert it into decisions. This is a workforce investment, not a technology investment, and it competes for the same budget dollars that AI tooling currently claims.

The insurance industry's 88% optimism rate is not misplaced. The financial fundamentals are strong, the technology capabilities are real, and the deployment velocity is impressive. But deployment velocity without workforce readiness produces exposure, not advantage. The Covenir data quantifies where the gap exists, how wide it is, and which functions will break first when it widens further.