The Voice on the Other End of That Pizza Order Wasn't Human

01 /The Concept

The voice that took your Domino's order by phone this year wasn't human. Same for Wingstop. The reason you didn't notice is not an accident of convergent feature parity with human agents – it is the result of a data strategy most voice AI companies aren't pursuing.

Most competitors train on publicly available audio: podcasts, audiobooks, YouTube. That corpus is fluent but it doesn't sound like a customer service call. The register is different. The pacing of someone trying to place an order on a phone line is different from someone performing for an audience. Rime built a recording studio in San Francisco and hires voice talent to speak in the specific register that transactional phone conversations require – not polished narration, not podcast performance, the acoustic texture of someone routing an insurance claim or scheduling a medical appointment.

The resulting models handle high-volume enterprise calls at Domino's, Wingstop, Asurion, Upstart, Dialpad, and Mayo Clinic. Thirty-five people built them. A $24 million Series A, led by M13 Ventures with participation from Twilio Ventures, closed in July 2026, bringing total funding to $29.5 million.

02 /Why It Matters

The crowded end of voice AI – ElevenLabs, Cartesia, Play.ht – is competing for content creation: voiceovers, entertainment, publishing. Rime is not competing there. Enterprise phone calls have different evaluation criteria. What matters is not naturalness in a studio. What matters is latency consistency under telephony conditions, handling of edge cases according to company-specific escalation rules, and whether the model can hold a conversation that sounds right inside a specific industry context.

Twilio's participation in the Series A is the signal worth examining. Twilio operates a significant portion of the telephony infrastructure that enterprise customer service runs on. A strategic position in a voice AI company winning at Mayo Clinic and Asurion – institutions with real compliance requirements and high call volumes – suggests Twilio sees enterprise conversation handling as a layer that sits on top of its own infrastructure rather than threatening it. That alignment creates distribution leverage that content-creation competitors can't easily replicate.

03 /Opportunities

The product that should exist and doesn't: a voice persona system. Each enterprise has a brand voice – specific vocabulary, acceptable escalation paths, handling rules for every edge case. Codifying that voice spec into a deployable system that trains consistently across inbound calls creates switching costs that rival a CRM. Once a healthcare system or insurer has built its call-handling logic into a voice AI workflow, moving to a competitor isn't a technology decision. It's an integration and retraining project.

Rafael Valle's hire as Chief Scientist – formerly at Meta's Superintelligence Labs and Nvidia's applied deep learning audio research – signals that Rime is treating this as a model-quality arms race. The data advantage compounds: each enterprise call the models handle teaches them something about the actual conversational register of that domain. Competitors trained on public data can't replicate that loop without equivalent recording infrastructure.

The vertical most underserved right now: mid-market healthcare clinics. They have the call volume to justify automation and the absence of internal infrastructure to build it themselves. Mayo Clinic is the proof point at the enterprise end. The next hundred accounts are the clinics between it and the individual physician's office.