Insurance AI's Next Test Isn't Answering the Call

For the past few years, the industry's question has been about capability. Can AI answer the call? Can it understand a policyholder who is upset, talking fast and not following a script? Increasingly, the answer is yes.

The question for 2027 is about execution. Can it complete the work?

A night-time hurricane call shows the difference. An AI agent that answers and takes a message has replaced the voicemail. One that takes the report, captures the details and starts emergency mitigation while the storm is still passing has changed the claim. 

Our new paper, Insurance 2027: 7 Predictions That Will Shape Claims, Sales and Service, lays out seven predictions. Three of them show most clearly what completing the work will mean.

Prediction: FNOL becomes the first action, not the first form

First notice of loss has long been treated as intake: collect the information, create the claim, send it downstream. Digital adoption is already rising. J.D. Power's 2026 U.S. Property Claims Satisfaction Study found that 38% of homeowners customers who filed a claim reported it digitally.

The bigger opportunity isn't faster intake. It's closing the gap between reporting a loss and acting on it. The same study found that for customers using direct repair programs, repairs on higher-severity claims averaged more than two weeks shorter.

One executive put it simply: "If it's a simple glass claim, you go ahead and just schedule it for them right there."

That changes what carriers measure. Cost per FNOL matters less than what the first interaction set in motion: complete loss information, the right routing, mitigation started and less follow-up work for the adjuster.

Prediction: The handoff becomes part of the AI's job

The debate over whether customers want a person or AI misses the point. They want the problem solved. Sometimes AI can do that. Sometimes a licensed agent or an adjuster needs to step in.

The experience breaks when the customer has to start over. J.D. Power's 2026 U.S. Auto Insurance Study found that 21% of customers were forced to switch channels to resolve a single inquiry. Those customers were significantly less satisfied and less likely to renew.

A transfer moves the call. Orchestration moves the conversation, the context and the work already done. In 2027, I expect carriers to judge AI less by containment rates and more by a different set of questions. Did it know when to escalate? Did the right person get the work? Did the customer have to repeat themselves?

Escalation won't be evidence that the AI failed. Escalating well will be part of what the AI is for.

Prediction: "Show your work" becomes the standard

Regulators have been clear that using AI doesn't change an insurer's obligations. According to the NAIC, existing state insurance laws apply whether a decision is made by a person, an algorithm or a third-party vendor. Its Model Bulletin sets expectations for governance, documentation and testing, and in 2026 the NAIC began piloting an AI Systems Evaluation Tool with 12 states.

Carriers will need to show not just that AI completed a task, but how it did it: what happened in the interaction, what actions it took, when it escalated and where people stayed in control. One leader described the environment they want: "Everything is logged, tracked, auditable."

That matters for compliance. It matters just as much for operations, because supervisors need to see where AI succeeds and where it struggles. Auditability can't be added later. It has to be part of the architecture.

Where to start

None of this requires transforming the enterprise at once. It requires starting with the right workflow and asking a few questions before it moves toward production:

• Is the workflow clearly defined, including what a successful outcome looks like?
• Are the systems, APIs and data sources it needs mapped?
• Are the operational details understood, from authentication to transfer rules and field mappings?
• Has escalation been designed, including what context follows the work?
• Has it been tested beyond the happy path?
• Is governance built in, and is there a plan for what happens after launch?

The dividing line in 2027 won't be between insurers that use AI and insurers that don't. It will be between those running isolated experiments and those building AI into how the work gets done.

Answering the call was the first test. Completing the work is the next one.

The full set of seven predictions, with the research behind them, is in Insurance 2027. Download the paper →