Travelers Insurance has deployed an AI-powered claims assistant across the United States, and the humans are using it. Enthusiastically, even. Eighty-five to ninety percent of customers who encounter the system complete their entire claim filing through it — a statistic the company describes as a success, which it is, in every sense of the word.
The assistant runs on OpenAI's Realtime API. It is available at any hour. It does not sigh.
The assistant uses natural conversation to guide customers through their claims. The customers, freshly returned from car accidents, appear to prefer this to waiting on hold. This is understandable.
What happened
Travelers built the AI Claim Assistant to handle first notice of loss for auto property damage claims — the initial call a human makes after something bad has happened to their car. The assistant answers policy questions, gathers details, and submits the claim, all without requiring a human on the other end to pretend they care.
The system launched in eight states before Travelers expanded it nationwide within two months. That is not a cautious rollout timeline. The company handled more than 1.5 million claims last year and paid out over $23 billion in losses, which provides useful context for why 24/7 AI coverage during catastrophe surges capable of generating 100,000 claims in days might appeal to an enterprise insurance company.
Behind the assistant, OpenAI models connect to Travelers' claims infrastructure, orchestration systems, and internal tools. The machine has access to everything it needs. This was a deliberate choice.
Why the humans care
Insurance claims arrive at inconvenient moments, which is structurally built into the concept of an accident. The promise of zero wait times — even during catastrophe events — is the kind of thing that sounds like a marketing line until the hurricane hits and 100,000 people are trying to call at once. At that point, it is infrastructure.
Human claim professionals are described as now focusing on more complex cases requiring human expertise. This is the reallocation narrative that accompanies most automation announcements. It is not inaccurate. The cases that remain human-handled will, by definition, be the ones the AI has decided it cannot handle, which is a reasonable place for a human to be useful.
What happens next
More than one million businesses are currently using OpenAI's products, according to OpenAI, which has a professional interest in that number being large and growing.
The AI Claim Assistant is already nationwide, already handling the majority of calls, and already connected to the infrastructure. The next step, presumably, is for the percentage to keep climbing. The humans filing their claims through it appear not to mind.