McKinsey lists insurance among the highest-potential sectors for AI value creation, with generative AI alone capable of producing more than $1.1 trillion in annual value across financial services including insurance. That is a top-down number with a lot of assumptions in it. The bottom-up evidence is more measured. Insurance has well-defined documents, well-defined workflows, and an enormous amount of unstructured text moving between adjusters, underwriters, agents, and customers. Those are exactly the conditions where AI tends to land productivity gains. They are also exactly the conditions where the regulatory environment moves slowly enough that adoption lags potential.
What is happening on the ground is less dramatic than the trillion-dollar headline suggests, and more interesting than the press releases describing AI-first claims handling. The work of an insurance professional in 2026 is reshaping around the parts of the job AI can take a first pass at, with humans still owning the decision and the customer relationship.
Where AI is actually being used
Four use cases dominate insurance AI deployments today.
Claims triage is the most mature. When a customer files a claim, AI now routinely classifies it into severity tiers, identifies likely coverage match, flags incomplete documentation, and routes to the right adjuster. The adjuster’s first hour, which used to be spent reading the claim file from scratch, is now spent reading a structured summary the model has produced. That is a meaningful efficiency win without changing who decides the claim.
Underwriting summarisation is the second. Pulling the salient details out of a 60-page broker submission, surfacing prior loss history, and flagging risk factors that match the underwriter’s appetite is now a 10-minute task instead of an hour. The underwriter’s judgement on the actual decision is unchanged. The reading time is collapsed.
Fraud detection is the third, and this is the use case with the longest history. Insurance has been using machine learning for fraud pattern recognition for over a decade. The newer generative AI layer adds the ability to read narrative claim descriptions, claimant statements, and adjuster notes for inconsistencies, not just transactional patterns.
Policy comparison and customer service is the fourth. Customer-facing chatbots and agent-facing knowledge tools that can answer “is this covered” with a citation to the actual policy wording are now standard at most large carriers. The cross-sector data on AI customer service points to roughly 65% of incoming queries resolved without human handoff in 2025, up from about 52% in 2023.
The customer experience shifts (and the worry about it)
What customers experience has shifted accordingly. Faster acknowledgement of claims, faster initial decisions on simple losses, and richer self-service for policy questions. What customers worry about, with some justification, is whether AI-assisted denial decisions are being made carefully enough. Several state regulators in the US and supervisory bodies in Europe and the UK have begun looking closely at this. By the end of 2026, expect the framework around algorithmic-decision disclosure and human-review requirements in claims to harden meaningfully.
For insurance professionals, that is the part of the role that is not going away. AI can summarise. AI can flag. AI cannot, in most jurisdictions, lawfully be the entity that denies a claim or sets a premium without a human decision-maker accountable for the call. The regulator is the structural moat for the people doing this work.
The productivity gain, sourced
The closest peer-reviewed evidence on AI for adjacent knowledge work is the Harvard Business School and BCG study showing roughly 25% reduction in time-to-completion on consulting tasks within scope, and the GitHub Copilot RCT showing 55.8% speedup on a benchmark coding task. The realistic productivity gain on insurance documentation, summarisation, and drafting work sits in the 20 to 40% range on the tasks AI is good at. That is a useful slice of the week, not the whole week.
The “saves 6 to 9 hours a week” figures floating around insurance AI marketing are not from controlled studies of insurance professionals. The honest version is that an adjuster, underwriter, or agent who builds AI summarisation, AI drafting, and AI knowledge-lookup into their daily workflow will get back several hours per week, with the exact figure depending on which sub-discipline they sit in. Adjusters working high-volume property claims see more. Commercial underwriters working complex risks see less.
What the next 12 months look like
The McKinsey State of AI 2025 data shows 88% of organisations using AI in at least one function but only 7% who say AI is fully scaled across the org. Insurance fits that distribution. Most carriers have AI in production for one or two workflows. Very few are running AI across the full claims-to-renewal lifecycle.
Three movements are likely to be visible by the end of 2026. First, claims-handling cycle times for straightforward losses will continue to compress, with several large carriers competitively pricing on speed of payment. Second, the regulatory framework will harden, particularly around algorithmic transparency in pricing and claims decisions, with new requirements coming into effect in EU member states and likely several US state legislatures. Third, the agent and broker channel will start to see real productivity gains from AI tools that handle quote-comparison and policy-explanation work, the parts of the relationship that have always been most time-consuming and least valued by customers.
If you work in insurance, the most useful question is not whether AI changes your role. It does. The question is whether your team has moved past pilot experiments into repeatable, audit-ready, regulator-defensible workflows. The carriers that have done that will be visibly faster and visibly more profitable through 2026. The ones still running pilots will be talking about AI without showing it.