ElevenLabs creates the voice layer of AI — the models that change text into speech that sounds human. Most people come across it while dealing with customer service, often without knowing it. Klarna uses it for first-line phone support for 35 million U.S. customers. So do Deutsche Telekom, Cisco, Adobe, and a rising number of governments. ElevenLabs also sells to creators, who apply its platform to audiobooks, dubbing, and music.
The firm reports an ARR of $600 million, or annual recurring revenue. It is said to be valued by its investors at $22 billion, even though it has existed for only four years. That makes it a young unicorn with a large addressable market, and a CEO who is willing to discuss the rough edges of the business.
Staniszewski on the Turing Test
The co-founder and CEO of ElevenLabs, Mati Staniszewski, was interviewed at Nrth in Toronto, a conference for entrepreneurs that used to be called Elevate. Two years earlier, at TechCrunch Disrupt, he had forecast that audio models would become commodities soon. Today he believes the difference between companies is shrinking, though it has yet to vanish.
“There is still a lot of work to be done,” he said. “The quality delta you can achieve just on the model level is still significant. If we think longer term, probably three, five years from now, those differences will be smaller.”
His bigger ambition is the Turing test for conversational AI. “You need to combine intelligence, but you also need emotional intelligence,” he said. “You need to understand the emotions of the other side, to be able to slow down or speak up.” He admitted that hasn’t been done yet.
Enterprise vs. Creators
ARR for ElevenLabs comes from two distinct customer bases. More than half, fifty-five percent plus, comes from classic enterprise customers. The rest supports small and medium businesses, developers, builders, and creators.
The combination matters for pricing. Some buyers pick the reasoning layer from a list of choices, deciding between frontier lab models and open weight. Staniszewski said the choice isn’t binary. Open source models work well for informational calls. But for financial services, where authentication and transactions carry weight, frontier models still lead.
Competing With Customers
Customers are training their own voice products on ElevenLabs’ technology and competing against it more and more often. One such customer is Decagon, a conversational AI platform that built its voice product using ElevenLabs and now processes queries through its own models.
Staniszewski explained how the boundaries between model companies, platform companies, and application companies were fading, and “In the past you’d have very clear splits where one starts and ends,” he said. “Today that line is much more blurry.”
He cited Anthropic as a case in point, saying it was already happening there. “What was a model company is definitely a platform and increasingly a wide set of applications,” he said. He believes the same pattern will keep unfolding.
Margins Under Pressure
When asked about gross margins, Staniszewski was cagey but honest. He wouldn’t discuss them in detail, but he made clear he doesn’t mind them getting squeezed even further if it means expanding market share.
“I’m going to give a vague answer,” he said. “Given we have that research element, we’re able to fine-tune and constrain models in extremely smart ways. But if we can pass on any savings to the customer, we do that.”
He framed the trade-off around proving value. “The biggest thing is still proving the value and being there with the customer,” he said. “So if we can invest and prove that value, we don’t mind the margins going lower to actually benefit together as the value gets created in the next five years.”
Disclosure on AI Calls
Staniszewski believes companies ought to let customers know whether they’re speaking with a representative or a person, though he says it’s too soon for that disclosure. “I think there should be disclosure at this time,” he stated.
The problem is cultural. People aren’t used to it. “Currently, people aren’t used to it, and the common pattern is you don’t want to feel cheated on that call,” he explained. “But in five years, when everybody has their own agent working on their behalf, you’ll be calling in and expecting an agent.”
He offered a practical solution. “There are good ways of doing it — if there’s a 30-minute wait for a human, offer the customer a choice,” he said. “In almost all cases, they choose the agent and then they’re surprised by how good the experience is.”
Training Data
ElevenLabs has millions of hours of customer service calls. Some of that goes into training, but not all. Staniszewski said the bottleneck isn’t volume — it’s annotation.
“In certain companies, we created the models together. They wanted a specific model for their use case,” he said. Otherwise, the work has been labeling data. “We have thousands of people internally on a contracting basis helping us annotate not only what was said, but when people were speaking, how they said things, what emotions were used.”
He brought in voice coaches to help detect those subtleties.
Who Uses ElevenLabs
ElevenLabs’ customer list reads like a who’s who of global business and government. The company works with:
- Deutsche Telekom
- Cisco
- Adobe
- A growing list of governments
- Klarna’s 35 million U.S. customers
The company also serves creators building audiobooks, dubbing, and music.
Key Numbers
- $600 million ARR
- $22 billion valuation by backers
- 35 million U.S. Klarna customers
Staniszewski makes his view plain: customer service will come to rely on AI that mimics human speech, and firms will eventually demand it. The company is willing to let the question of whether the profit margins can stand the change remain open.
Source material: “ElevenLabs’ CEO on margins, IPO timing, and telling customers they’re talking to a bot,” TechCrunch.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.