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

Covenir's 2026 Insurance Operations Leaders Trends Report, released June 9 from a February and March survey of 152 U.S. insurance operations decision-makers, puts AI in live operations at 70% of organizations, up from 58% a year earlier.

The same survey finds 20% cutting training budgets and 7% protecting them. Those two numbers describe an industry deploying faster than it is preparing the people who have to supervise what it deployed.

Key Takeaways

• 70% run AI in live operations, a 12-point rise from 58% in 2025, against 20% cutting training budgets and only 7% taking deliberate steps to protect them.
• 54% of advanced AI adopters plan headcount cuts in 2026, five times the 11% rate among organizations still in early deployment.
• 42% name First Notice of Loss as the function where brand promise breaks down most often, the same function where AI triage is most attractive.
• 91% of C-suite respondents say their teams are at maximum strain, nearly double the rate reported by managers and individual contributors.
• 47% either do not use their operational data or cannot translate it into decisions, even though four in five call operational insight critical to business success.

What the Survey Found

The sample is narrow by design: 152 U.S.-based operations decision-makers at carriers, MGAs and insurtechs, spanning organizations from fewer than 200 employees to more than 1,000. These are the people who own the teams, workflows and budgets where AI meets daily insurance operations. It is a second-year survey, so the year-over-year comparison is real rather than reconstructed.

Deployment is up 12 points, from 58% to 70%, consistent with Celent's finding that 48% of global insurers run GenAI in production and with late-majority adoption pushing past the halfway mark.

The training side moves the other way. Twenty percent are cutting training budgets while expanding AI tooling. Seven percent have taken deliberate steps to protect them. The remaining 73% are neither cutting nor protecting, which in practice means drift as AI spending claims a growing share of the technology line.

Covenir president and CEO David Squibb framed it as leaders "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."

Insurance is not alone in the ratio. Global AI spending is projected to rise 44% this year against 5% growth in training budgets, with roughly 75% of knowledge workers using AI at work and 60% reporting no formal training on it. The difference is that insurance operations produce regulated decisions affecting policyholder outcomes, so untrained use is a compliance exposure as well as a productivity one.

Where the Gap Turns Into a Loss Number

The sharpest finding is the correlation between AI maturity and workforce reduction.

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

54% of advanced adopters plan headcount cuts, against 11% among organizations still piloting. That is a concentration effect rather than broad displacement: the carriers furthest along the deployment curve are removing the most human capacity.

The problem is which capacity. The experienced staff in claims, underwriting and service functions are the ones who handle exceptions, supervise AI output, train new hires and intervene when the model returns a wrong answer. Cutting them while letting training budgets drift for whoever remains removes the supervision layer and the means of rebuilding it in the same move.

The survey locates where that surfaces first. 42% identify First Notice of Loss as the function where brand promise breaks down most often. FNOL is high volume, which makes it the obvious automation target; it is the policyholder's first contact after a loss, which makes errors immediately visible; and it depends on judgment about when a standard workflow should be overridden.

For a reserving actuary the consequence is specific rather than reputational. If FNOL triage quality degrades at a material share of carriers, initial case reserves set at intake become less reliable, claims get misclassified at the point where severity is first recognized, and the subset requiring human reclassification after AI triage develops on a longer pattern than the rest. Development factor selections and IBNR built on a pre-deployment intake process will not reflect either effect, and neither shows up as a data quality flag.

It reaches pricing through the same door. Carriers cutting headcount alongside AI deployment will project expense ratio improvement in rate filings, and Morgan Stanley's 200 basis points of AI-driven expense savings assumes a clean transition. The Covenir numbers price a rougher one: error rates during the training deficit, the oversight exposure the NAIC's focus on AI claims handling is aimed at, and 91% executive strain against the change management the cuts require.

The Data They Cannot Read

The second gap compounds the first and is harder to see from outside.

Eighty-one percent of respondents say operational insight is critical to business success. 47% either do not use their operational data at all or cannot translate it into decisions. The dashboards exist and the pipelines run; the organizational capacity to convert either into a decision does not.

That extends the Capgemini measurement gap, where 42% of P&C insurers had never measured AI outcomes, from AI-specific measurement to operational data generally. It also sits alongside 62% of IT leaders naming data quality as their single biggest challenge across hybrid environments.

For anyone downstream of that data the risk is invisible in the data itself. A claims or underwriting extract can be complete, correctly formatted and internally consistent while being produced by workflows where AI and human processes interleave differently across months, with no one at the carrier positioned to notice the change. A model fitted across that boundary reads a process shift as an experience shift.

The capacity to fix it is where the survey ends up least encouraging. 91% of C-suite respondents report maximum team strain, nearly double the rate among managers and individual contributors, and mid-sized carriers of 500 to 999 employees report the highest budget growth alongside 83% strain: investing hardest with the least organizational slack to absorb the transition. Deloitte's parallel finding is that 90% of insurance executives agree on the need to reinvent employee value propositions for human-machine collaboration while 25% have taken tangible action, which is what a 91% strain rate produces at the decision layer.