The voice layer of AI

ElevenLabs builds text-to-speech models that sound human. Klarna uses it for first-line phone support for thirty five million U.S. customers. Deutsche Telekom, Cisco, Adobe, and governments also rely on it.

Six hundred million in revenue

ElevenLabs is pacing at six hundred million dollars in annual recurring revenue. The company is now reportedly valued at twenty two billion dollars by its backers, despite being just four years old.

Enterprise and beyond

Fifty five percent plus of ElevenLabs' revenue is classic enterprise. A big percentage of the remaining forty five percent are small and medium businesses, developers, builders, and creators.

Competing with customers

ElevenLabs increasingly competes with its own customers. Decagon trained its voice product on ElevenLabs and now runs queries through its own models. CEO Mati Staniszewski says the lines between model, platform, and application companies are becoming blurry.

Open source versus frontier models

ElevenLabs customers can choose reasoning models from a menu of options. For informational customer service, open source models work well. For financial services requiring authentication and accuracy, frontier models still lead.

Government deployments vary

ElevenLabs works with U.S., European, and other governments. Each deployment depends on the country's requirements. In Poland, ElevenLabs deployed agents that call patients to remind them of appointments, reducing no shows from eighteen percent.

Tell customers they are talking to a bot

Staniszewski believes businesses should disclose when someone is talking to an AI agent rather than a human. Currently people are not used to it and do not want to feel cheated. In five years, he expects this norm to shift.

“I think there should be disclosure at this time.”

Margins and market share

Staniszewski declined to discuss ElevenLabs' gross margins in detail. He said the company does not mind margins getting squeezed further if it means expanding market share and proving value to customers over the next five years.

Training data and annotation

ElevenLabs created custom models with some customers. The company has thousands of people on a contracting basis annotating customer service calls, noting what was said, how it was said, and what emotions were used. Voice coaches help detect accents accurately.

IPO timing remains vague

It has been reported that ElevenLabs is looking at twenty twenty eight for an IPO. Staniszewski said the company is preparing the foundation to do it in the next years, but whether it will depends on the time and place.

Pacing frontier AI development

Staniszewski said everybody is aligned to work together on finding a way to pace frontier AI development. He said the company should take right precautions as it deploys technology, though the role of public discussion and regulation is separate.

ElevenLabs' safeguards

ElevenLabs does not train text models or the intelligence side of models, which is the core of the AI safety debate. The company does not allow agents to create more agents. Every customer goes through KYC. Cybersecurity risk exists, but the company has precautions in place.