100 Million Calls a Month and Nobody Knows It's AI

01 /The Concept

The voice AI race has been about scale: more languages, more voices, lower generation cost. Rime made a different bet – that the ceiling on enterprise adoption was quality, not price, and that getting there required doing something most AI companies treat as solved: gathering better training data.

The company built a recording studio in San Francisco staffed by linguists. The goal was not volume but fidelity – capturing the rhythm, pausing, and tonal variation of real conversational speech that synthetic pipelines flatten out. Phoneme-level pronunciation controls followed: the ability to guarantee that a medication name, a brand term, or an airport code gets pronounced correctly every time.

The results are measurable. Enterprise customers report 3x latency improvements, with sub-100ms time-to-first-byte in live calls. Over 100 million calls per month now run on Rime's infrastructure. Customers include Mayo Clinic, Dialpad, Upstart, and Asurion – organizations where a robotic-sounding voice is not a minor annoyance but a clinical, legal, or financial liability.

The $24 million Series A – led by M13 Ventures, with Twilio Ventures and Unusual Ventures participating – brings total raised to $29.5 million. The platform supports 600+ voices across 50+ languages. On-premises deployment is available for compliance-sensitive industries. Founders Lily Clifford, Brooke Larson, and Ares Geovanos have grown the team to 35, expanding post-round.

02 /Why It Matters

Voice AI is one of the few enterprise AI categories where quality failure has direct, measurable consequences. A chatbot with occasional errors is frustrating. A voice agent that mispronounces "metoprolol" in a clinical workflow undermines the clinical process it was supposed to support. A financial IVR that sounds robotic during an account dispute creates legal exposure. The enterprises deploying voice AI at scale – healthcare systems, airlines, financial services, insurance – are exactly the organizations where quality failure is most costly.

This creates a segmentation the industry hasn't fully priced in. Low-stakes deployments go to the cheapest provider with acceptable quality. High-stakes deployments go to whoever can prove quality at scale, and they pay significantly more for it. Rime's customer list is consistent with winning the high-stakes segment.

The structural question is durability. ElevenLabs, Play.ht, and OpenAI's speech models have all improved materially in the last 18 months. Rime's advantage compounds in ways that generic model improvements do not. Phoneme controls are a direct function of supervised annotation work. On-premises deployment is institutional expertise. And the 100 million monthly calls generate real-world pronunciation correction data that improves the next generation of models on actual enterprise edge cases, not benchmark datasets.

100 million calls per month is not a traction metric. It's a training data pipeline.

03 /Opportunities

Voice quality is most underdeveloped in languages where training data has historically been thin: regional dialects, lower-resourced languages, and code-switching contexts where a caller alternates between English and Spanish mid-sentence. A platform that solves pronunciation and rhythm in those contexts serves markets Rime isn't addressing and faces no well-capitalized direct competition.

Voice authentication is the adjacent integration that hasn't been fully built. Banks and healthcare systems are experimenting with voiceprint verification for call identity. A platform that handles both conversation and caller authentication creates a single enterprise integration instead of two – and captures a security workflow that comes with regulatory tailwinds. The provider that runs the call and verified the caller's identity occupies a different contractual position than a pure voice quality vendor.

The clearest near-term opportunity is outbound. Most enterprise voice AI is reactive: someone calls in, the system responds. Outbound voice – healthcare appointment confirmations, insurance renewal reminders, clinical follow-up calls – has a different quality requirement because the recipient didn't initiate the interaction. Getting tone, pacing, and brand alignment right in outbound healthcare, where every call runs under HIPAA constraints and clinical protocols govern what can be said and how, is harder and less mature than inbound. A player who cracks outbound quality at Rime's latency level, in regulated healthcare specifically, enters a market with almost no direct comparison and a clear ROI story: fewer missed appointments, measurable cost savings, auditable compliance.