Giving employees AI tools is the easy part. The returns come from embedding AI into decision-making, building confidence in its outputs, and combining AI with human expertise to improve decisions, with people remaining accountable.

Capgemini’s World Property & Casualty Insurance Report 2026 found that nearly half of surveyed property and casualty (P&C) employees said they are comfortable relying on AI recommendations for moderately complex decisions. But comfort is not performance. And 42 percent of insurers track no AI metrics at all, so many cannot say whether those decisions are improving outcomes.

Note: Moderately complex decisions include evaluation of standard risks, premium calculation and adjustment, retention or upsell offers, payout or reserve adjustments, product or carrier selection, cross-sell offers.

In insurance, the experts (underwriters, claims adjusters, and agents) remain accountable for every decision. The higher the stakes, the more judgment, validation, and oversight they need. If they cannot understand a recommendation, they will not act on it, and adoption stalls.  

The challenge, then, is not simply to build employees’ confidence in AI, but to embed it in ways that measurably improve decisions and their outcomes. Three shifts can help P&C insurers do that. 

1. Redesign workflows to embed AI into decision-making  

When AI is used in a standalone app, a chat-style sidebar, or an add-on to the insurance core platforms, experts must go to it manually, switching apps, gathering context, and interpreting outputs before they act. Under tight turnaround times, that demands extra work, and many employees will revert to established workflows instead. 

The better approach is to redesign the workflow around expert-AI collaboration, with AI operating inside the insurance core platforms where decisions are made. In underwriting, it can combine broker submission data with external risk information, highlight material risk factors, and surface guidance in the underwriter’s workbench. In claims, it can summarize the file, identify missing evidence, and flag cases needing a specialist adjuster. 

Embedding also allows review checkpoints to be built into the workflow, scaled to the level of risk; routine decisions may only need to be validated, while complex or high-consequence cases should require explicit expert review before proceeding. This allows insurers to scale AI-supported decision-making without treating every underwriting or claims decision as carrying the same level of risk. 

2. Design for awareness, not uncertainty 

Experts must know how far to rely on what they see. A confident-sounding answer invites overreliance, and an unclear response gets ignored. A Q&A-style sidebar can produce both: fluent answers, without the case context or supporting evidence experts need to judge them. Recommendations embedded in the workflow, alongside that evidence, let experts calibrate how far to rely on them.  

Good design starts by labeling AI-generated outputs clearly and showing how they were produced by revealing key risk factors, supporting evidence, data sources, and material uncertainty. Not just an answer or a confidence score. Labels should distinguish information from required action. Like road signs that each prompt one clear response – stop, look, and proceed – a clearly labeled AI-generated value or recommendation nudges experts to pause, review, and reconsider before acting.  

For underwriters and claims adjusters, this distinction matters because an AI recommendation should inform professional judgement, not obscure where accountability for the decision sits. Experts should be encouraged to review, challenge, and refine outputs, and their feedback should shape the design.  

The goal is not maximum trust, but appropriate reliance: knowing when the evidence supports a recommendation, when to investigate further, and when expertise should prevail. For insurers, this means enabling experts to distinguish between decisions where AI can help accelerate outcomes and those where risk, complexity, or uncertainty requires specialist judgment. 

3. Measure decision quality 

Adoption rates, time saved, and tasks automated show usage, not value. Insurers need to identify whether AI is improving the decisions that ultimately lead to results. 

In underwriting, that means risk selection, pricing accuracy, and referral quality, judged against subsequent loss performance. For claims, measures include leakage, settlement accuracy, cycle times, and appropriate escalation. 

This also changes how experts relate to technology. As AI recommendations become more embedded in underwriting and claims decisions, the technology itself becomes part of what is evaluated. Experts and AI should therefore be assessed together on decision quality, with the expert remaining accountable. 

Joint measurement controls two opposing risks: low adoption, where useful recommendations go unused, and confirmation bias, where they are accepted without challenge. Regular reviews of accepted recommendations, overrides, and outcomes show which is happening.  

When experts overrode AI, was their judgment sound? Are errors clustering in particular risk or claim types? Are claims recommendations improving settlement outcomes? Questions like these help pinpoint where and why decisions may be breaking down. 

Make Expert-AI collaboration an operating capability 

Insurers must define the decisions AI can execute or only recommend, and which require expert approval. Experts need a clear route to challenge recommendations and escalate disagreements. Business, frontline, technology, and risk leaders should review joint performance regularly and feed findings back into models, workflows, and training. 

Effective adoption requires ongoing training and support, not a one-off launch. And yet, our research found that 72 percent of P&C insurers’ AI spending goes to technology, compared with just 28 percent to change management. This imbalance risks leaving insurers with capable technology but without the workflows, skills, and governance needed to translate AI capability into better decisions. 

The advantage will go to insurers that make expert and AI judgment work as one, connecting better underwriting and claims decisions to stronger insurance performance, better policyholder outcomes, and tangible business value.