Insurers are moving AI from general efficiency projects into underwriting, claims and capital allocation. The 2026 Evident AI Index describes a sector under pressure to show financial value from systems that influence its core ability to select and price risk.
That transition raises the standard for evidence and governance. Saving minutes on administration is useful; changing a policy decision can improve loss performance or introduce unfairness at scale.
Talent and leadership are moving closer to production
Across the 30 insurers tracked, overall employment declined 2.2% while AI-specialist headcount rose 32%. AI specialists now represent roughly one in 50 employees, with more hiring directed toward development and implementation as foundational data work matures.
Nearly 40% of indexed insurers have a senior leader explicitly responsible for AI, most appointed during the previous year. That structure can connect investment, model risk and business ownership instead of leaving experiments distributed across departments.
Agentic use is rising quickly: one in four newly disclosed cases shows orchestration across workflow stages, compared with one in twenty six months earlier. Coordinating policy, claims and service tasks can reduce hand-offs, but it also lets one error propagate further.
Executive oversight needs action-level controls, independent validation and a clear owner for customer outcomes. A central AI office should not become a substitute for accountability in underwriting and claims teams.
Shared platforms can turn experiments into governed capabilities
Zurich rose from twelfth to fourth in the index after emphasising a shared platform model. ZurichIQ supports underwriting, claims, legal and service operations, with modules for contract comparison and enforcement of underwriting guidelines.
A common environment reduces duplicated integrations and allows consistent monitoring. Zurich combines it with an investment and model-risk committee and employee training, including an AI apprenticeship programme.
The platform approach also makes performance comparable. Teams can reuse data and controls while measuring whether a use case changes cycle time, leakage, selection or customer experience.
Standardisation should not erase local market and product differences. Models need validation against each jurisdiction, portfolio and customer group, especially when recommendations affect eligibility or price.
Human underwriters remain responsible for exceptions and judgement. Interfaces should show which evidence influenced a recommendation and allow specialists to document why they departed from it.
Core risk workflows make ROI visible and consequential
Claims commonly consume 60–80% of premium income, so small improvements in fraud detection or risk selection can outweigh broad administrative savings. That economic leverage explains the shift toward core decisions.
Manulife, Generali and Intact Financial have disclosed projections that together exceed $1 billion in AI-driven value. Allianz reports hundreds of use cases and the sector’s largest AI talent pool, while AXA also retains a leading position.
Boards should examine how value is calculated. Avoided losses, faster claims and revenue gains need baselines, confidence ranges and adjustment for model, integration, training and oversight costs.
Customer measures belong beside financial ones. False fraud alerts, adverse-selection bias, appeals and claim delays can reveal damage hidden by a headline return.
Dynamic climate and cyber risk make continuous monitoring essential. A model trained on yesterday’s loss patterns may become least reliable when conditions change fastest.
Insurers should retain routes for brokers and customers to correct underlying data. Better underwriting cannot rest on profiles that are detailed but wrong, especially when errors follow a person across renewals.
The sector’s AI maturity will not be defined by the number of deployed agents. It will be visible in disciplined underwriting, accountable claims and transparent economics.
Insurers have begun to treat AI as part of the operating model rather than a technology initiative. That can generate substantial value, provided every automated insight remains connected to evidence, a responsible decision-maker and the promise that insurance will respond fairly when a customer needs it.
