ElevenLabs’ CEO on margins, IPO timing, and telling customers they’re talking to a bot

ElevenLabs develops the voice layer of AI—models that convert text into humanlike speech—and its technology is used extensively in customer service. The company’s voice systems power first-line phone support for firms such as Klarna (for 35 million U.S. customers) and are deployed by customers including Deutsche Telekom, Cisco, Adobe, and several governments. ElevenLabs also serves creators who use its platform for audiobooks, dubbing, and music. CEO Mati Staniszewski told TechCrunch that ElevenLabs is pacing at about $600 million in annual recurring revenue and that more than half of that comes from traditional enterprise customers, with much of the rest from small and medium businesses, developers, and creators. He said the company’s reported investor valuation is roughly $22 billion, despite the firm being four years old. Staniszewski addressed several industry trends in the interview. He noted that audio-model differentiation still exists but expects gaps to narrow over the next three to five years; his long-term goal for the company is to combine intelligence and emotional awareness to pass a Turing-like test for conversational AI. He acknowledged that model and application boundaries are blurring as some customers build competing voice products—citing Decagon, which trained its voice on ElevenLabs and now offers its own model—but said investors have not appeared worried. On model sourcing and deployment, Staniszewski said choice depends on use case: open-weight models can be fine for informational calls where errors are low-risk, while frontier closed models remain preferable for sensitive tasks such as financial-service interactions that require authentication. He also described how deployments for public-sector clients, like a Polish healthcare reminder system, can use a mix of model types and prioritize data residency and fine-tuning to local needs. Regarding transparency, Staniszewski argued businesses should currently disclose when callers are speaking with an AI agent, because most people are not yet accustomed to it and may feel deceived otherwise. He predicted social expectations will shift over time as personal agents become common. On costs and margins, he declined to detail gross margins but said ElevenLabs focuses on fine-tuning and efficiency, and is willing to accept lower margins if that helps expand market share and deliver value to customers. He also described significant annotation work—contractors and voice coaches—to build models that capture timing, emotion, and speaking characteristics for clients who requested tailored voices.