Every year, carrier CEOs have to make budgetary decisions that include how much to spend on IT infrastructure and, even more importantly, how much of the IT budget should be allocated to AI. In 2026, industry spending is about 4.6% of direct written premium, a barely noticeable increase of a tenth of a point from 2025 and up from 3.7% a decade ago. For a carrier with $5 billion in DWP, that's about $35 million. While that may sound like serious money, it would be enough to cover replacing a policy admin system in a single line of business, not to mention the two years of steering committee meetings thrown in at no charge.
Right now, the 2027 budget is being argued over in tense conference room discussions. This is the right time to ask where the 2026 IT budget went—and what we can realistically expect from it in the year ahead.
Coming into 2026, AI was no longer new territory. Every insurance carrier was factoring it in. A survey by IA Capital Group of national and regional P&C and life and annuity carriers found a clear focus on the core insurance value chain.
82% have AI in use or in evaluation in claims
73% in underwriting
73% in coding/engineering
45% in sales/distribution
45% in customer support
Only 9% said they were broadly integrated across multiple business lines. So how is it actually panning out?
Most of the AI investment in underwriting this year went to commercial submission triage. The reason was that it can increase premium growth, which is more important than just reducing costs. Many commercial insurers were declining to quote 60–70% of inbound submissions, and the problem wasn't the risk. They simply didn't have the time to review them quickly enough.
That is the point: AI is not making underwriters smarter than they already are. What it does is widen the funnel, which turns the capacity problem into a better risk-selection problem The hit ratio and GWP per underwriter are the metrics that won over CFOs. Zurich's reports back this up, with processing-time reductions of 50% to 97%, hit-ratio gains of at least 15%, and GWP per underwriter increasing by as much as 30% across roughly 1.5 million submissions. .
What changes in 2027: Gen AI was the headliner in 2026, but 2027 will see an increased use of agentic AI in underwriting. Commercial P&C carriers running agentic underwriting in controlled environments are now reporting loss ratio improvements of 3 to 5 points and quote-to-bind reductions of 60-99%. Put three points against a $1 billion commercial book, and insurers are looking at $30 million.
Not every insurer is jumping on board, with many preferring to wait and watch. Only 16% of US and Canadian P&C insurers currently use AI to support human underwriting, while 60% plan to prioritize it by 2028. We expect 2027 to be the year when insurers view this new technology as an advantage but by 2028, it is likely to become a standard requirement—and eventually, table stakes. And once something becomes table stakes, it stops being a source of competitive margin.
Nearly every major carrier has applied AI and machine learning to FNOL intake, photo-based damage estimation, and fraud flagging. State Farm, for example, joined OpenAI's Frontier program in February 2026 as a launch partner. The company has more than 96 million policies and accounts. SimpleSolve is seeing similar benefits through its SimpleINSPIRE platform.
But there's a catch. Even if AI can cut costs by 20% to 40%, the regulatory risk changes once the technology moves beyond handling information and starts influencing the outcome of a claim. Once AI touches severity prediction, reserve setting, or anything that influences what a claimant gets paid, things get dicey. You're getting into market-conduct territory.
The NAIC is developing an AI Systems Evaluation Tool to help state regulators examine how insurers use AI, including their governance, risk controls, potentially high-risk AI models, and the data used by those systems. In 2026, the tool is being piloted by 12 states in a mix of market conduct exams, financial exams, financial analysis, and related regulatory inquiries.
What 2027 looks like: Carriers that handle this well will draw the line early. AI does the intake, routing, extraction, and adjuster augmentation; humans handle coverage and reserves, with an audit trail thick enough to hand to a regulator without a lawyer in the room.
Here's the 2026 finding that should make a CEO put down the coffee: 93% of AI investment is going to technology and data, versus 7% to redesigning work and adapting the workforce.
There is a valid logic to this since carrier data is what fuels the AI machinery. Much of the budget can go to streamlining the museum of acquisitions with multiple claims systems, legacy policy platforms and a bordereaux process involving a spreadsheet someone emails every Tuesday. You can't run a model on that anymore.
But infrastructure is easier to approve than changing how people work. And that 7% matters since one in five insurers is deploying AI while cutting training budgets.
What 2027 does to it: Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. The lesson isn't to spend less on AI. It's to fund the technology and the work redesign together and kill weak projects early.
The 2026 AI budget tells a fairly clear story. Insurers spent heavily on the technology itself, while the less visible work of changing processes, preparing people, proving value, and putting the right controls in place often received less attention. That imbalance will become harder to ignore in 2027, when carriers have to show what all that spending actually delivered.
The answer is not simply to spend more. It is to be more deliberate about where the money goes, from the systems and data underneath the technology to the people and processes that determine whether it delivers anything useful. Just as important, insurers need to be willing to walk away from projects that aren't working rather than keep funding them because they made it into the budget.
The carriers that get this right won't necessarily have the biggest AI budgets. They'll be the ones that know where to spend, what to expect in return, and when to stop.