Why insurance carriers shouldn't chase 100% automation

Everyone is measuring autonomous completion rates in insurance right now: How much of a claim can AI complete, start to finish, without a human at all?

Here's the thing I always come back to when it comes to autonomous completion rates. If you make automation the main focus of your business, you'll likely ruin the one thing policyholders are paying for: Trust. Instead of focusing on how AI can handle 70% of the calls, pay attention to the 30% it should leave alone.

Take Frontline Insurance, the company uses voice AI to handle 70% of their FNOL calls and responds in about one second. What Frontline did right was to carefully design its workflow. The last step wasn't more automation, but a transfer to a human when the situation is too complex or emotional. That step is easy to treat as a leftover, as the exception, the part you try to shrink over time, but it's actually the opposite. 

The pressure to automate is intense, and it isn't going away. In 2025, insured losses from natural catastrophes topped $100 billion for the sixth straight year, according to the Swiss Re Institute. At the same time, the people who answer those calls are retiring faster than we can replace them.

Don't treat AI like a wall

AI is critical to our future, but the temptation is to treat it like a wall, protecting not just against the future but also against our customers. The theory goes that if you keep perfecting your AI systems, you'll keep getting closer to 100% automated calls, with no human contact at all. 

In J.D. Power's 2026 U.S. Property Claims Satisfaction Study, satisfaction rose as repair and payment cycle times shortened, but speed wasn't the only lever. Customers also cited how easily they could reach their insurer and get a straight answer. The average homeowner still waits more than 40 days from first notice of loss to final payment. Across those weeks, what people remember isn't the seconds AI shaved off. They remember when a human was there to help them when they needed it.

So here's the number I would put on the wall instead. Not "what percent did AI complete," but "how fast did a frightened person reach the right person on calls that needed one?" Automation rate measures capacity. Time-to-human measures care. Escalation should be a feature, not a failure. But this is much harder than it sounds. 

Be honest about what your system shouldn't touch

Building AI that knows when to step back is more complex than building AI that just keeps going. It means engineering the handoff, not just the deflection. It means being honest about what your system shouldn't touch, including a total loss, a grieving policyholder, and a claim where the facts don't fit the form. That work stays with people because the alternative produces a worse outcome.

The carriers getting this right aren't using AI to replace adjusters. They're using it to redirect them. The adjuster, who previously spent three hours entering data, now has time to be with a family that just lost its home. This is not a smaller task; it's the core of the job.

Insurance has always wanted to be there for people in their worst moments. For a long time, the cost of doing so at scale made it impossible. AI finally makes it possible, but only if we're clear about what we're automating and why.