92.2% of Hotel Callers Who Don't Book Leave Zero Trace. OpenAI's New Voice Model Is Built for the Reservation Line.

OpenAI released GPT-Live-1 on September 10, a voice model that listens and speaks in a single pass instead of the stacked speech-to-text-then-LLM-then-text-to-speech pipeline most phone bots still run on. OpenAI says it cuts interruptions by roughly 80% compared to turn-based systems, and the company names reservation lines and customer-service telephony directly as target use cases, with full-duplex handling designed for exactly the kind of overlapping, interrupt-heavy conversation a live phone call is. Pricing is $0.05 per minute for the voice layer, available now in the API.

The hospitality phone channel this targets has a well-documented, expensive gap. A Revinate secret-shopper study of 308 calls across 135 hotel properties found 92.2% of callers who didn’t book on the call left absolutely no digital trace — no email, no phone number — meaning they simply vanished as retargeting prospects the moment they hung up. Only 4.7% of calls captured an email address; 5.4% captured a callback number.

Restaurants show the same pattern from a different angle: SevenRooms reports 64% of diners still call to book, but 40% of those calls go unanswered during service — and separately, 75% of diners say they’re comfortable letting AI handle the reservation entirely.

Put those together and the numbers argue for the same fix from two directions: a full-duplex voice model that can hold a natural phone conversation removes the excuse for an unanswered line, and if it’s capturing structured booking data by design, it closes the exact trace-and-follow-up gap the Revinate data measured. The technology arriving this week doesn’t guarantee any property implements it well — but it does mean “our phone line just doesn’t scale” is no longer a credible reason to keep losing more than 9 in 10 callers who never left a name.