Travelers sends AI claims assistant nationwide on OpenAI Realtime API

Travelers now handles auto property damage first notice of loss across the entire United States with a fully autonomous voice agent built on OpenAI's Realtime API and frontier models, and 85–90% of customers who use the AI Claim Assistant complete their claim filing through it.

The insurer did not move slowly. It launched the assistant in eight states, then expanded it countrywide within two months. The system guides customers through the first notice of loss process for auto property damage claims in natural conversation: it answers policy questions, gathers accident details, and submits the claim on the customer's behalf.

The operational stakes are large. Catastrophe events can generate more than 100,000 claims in a matter of days, according to the company. Last year, Travelers handled more than 1.5 million claims and paid out more than $23 billion in losses. A voice channel that absorbs surges at that scale, without queues, changes how an insurer staffs and structures its claims operation during precisely the moments when customers are most stressed.

Patrick Gee, SVP of Auto and Property Claims at Travelers, said the choice of model came down to performance under real-world pressure. "What set OpenAI's real-time model apart was the ability to perform in that environment," Gee said. That environment is a customer calling immediately after a car accident, needing help and confidence that the claims process will go smoothly.

Behind the scenes, the deployment required more than an API key. Travelers connected OpenAI's models to its claims infrastructure, orchestration systems, and internal tools so the assistant could operate safely at enterprise scale. That integration layer — connecting a conversational model to policy data, claim submission pipelines, and internal tooling — is where enterprise AI deployments typically succeed or fail, and Travelers treated it as the core engineering problem rather than an afterthought.

The company frames the customer benefit in concrete terms. Customers reach support 24/7 with no wait times, including during catastrophe events. Claim professionals, in turn, focus on more complex cases that require human expertise. That division of labor — autonomous AI for high-volume, structured intake; humans for judgment-heavy work — has become the standard playbook for large-scale enterprise AI adoption, and Travelers' numbers offer one of the clearest insurance-sector data points to date.

The 85–90% completion figure matters because voice agents often lose callers mid-flow. If roughly nine in ten customers who start with the AI finish their filing with it, the assistant is not a deflection screen in front of humans; it is the primary intake channel for this claim type. Combined with the two-month path from eight-state pilot to national rollout, the deployment shows how quickly conversational AI can move from experiment to production infrastructure once it clears enterprise reliability and safety requirements.

For the broader market, the story signals that real-time voice models have crossed a threshold in regulated, high-stakes customer service. Insurance claims intake involves sensitive personal information, policy interpretation, and strict process compliance — not a sandbox for fragile demos. Travelers' willingness to run the assistant nationwide, and to publish completion rates alongside its $23 billion annual loss payout, indicates the technology now meets the bar for that environment.

The open questions are the ones the announcement does not answer: how the assistant performs on edge cases, how customers who opt out of AI are routed, and how the system behaves under a genuine catastrophe surge rather than steady-state volume. But with national coverage, 24/7 availability, and completion rates approaching 90%, Travelers has set a reference point that competitors writing auto property damage claims will now be measured against.