ElevenLabs CEO: Why Voice is the Next AI Interface

We partnered almost three years ago, and it's great to hear all the kind notes. When we partnered, the infrastructure team had only three people. Now, as ElevenLabs' founder, I can say our company's infra team has grown to 11 people. We love the number 11, and it seems to be everywhere here, especially with the companies present having raised $66 billion in total funding. To start, I believe the smartest person I know is my co-founder, Piotr. He has been the research brain behind creating many models and assembling incredible researchers in the voice space. Together, we created the first text-to-speech model that truly understands context, turning it into emotional intonation.

It's very tricky. Many of you likely face the same dilemma: should you build a product when research innovation might displace it? We encountered this early on. For example, we had a model at work, and a common request was for different voice speeds—an additional slider to modify audio generation speed. We were strongly against sliders or toggles, not wanting to become like previous generations of editing suites. Instead, we aimed to solve it at the research level, where the system would inherently know the appropriate speaking speed based on the voice. We resisted this for about nine months, but couldn't solve it through research. Ultimately, a simple product solution quickly satisfied all users.

It was so bad. In post-communist Poland, it was a cheaper way to produce audio. You didn't need to hire as many people; you just had one monotone audiobook reading of a movie. That's where the company began. We started in Europe, but quickly realized that to solve what was then a research problem, we needed to hire the best people, wherever they were. We couldn't limit ourselves to San Francisco or the West Coast. We needed to find talent across Europe and Asia and bring them into the company. So, we started fully remote, focusing on finding those individuals. For engineering, we also rejected traditional hiring methods, like relying on LinkedIn or conventional backgrounds. We sought a different approach to hiring, which led to some very interesting hires.

The first step is to spend time with the industry, understanding their priorities and incentives. This can be tricky; sometimes you might even be starstruck. We had the honor of working with Jared on his incredible work, learning what's important and where AI can be effectively used in the production process, and where it's best to maintain human involvement. This is a crucial thesis for all partnerships in this space. In our case, we focused on the voice space, considering how voice acting will evolve with technology. To cater to diverse use cases, you need a vast array of voices, languages, accents, and styles. We launched Voice Marketplace, allowing users to create and share their voices, earning money in return. Today, we have almost 10,000 voices and have paid $10 million back to our community.

Hiring legal counsel was one of our trickiest roles. We initially hired a couple of people who weren't a good fit. Our third hire came from Fortune 500 companies, with no startup or venture experience. Every conversation with them focused solely on risks, making it difficult to innovate. We understood the risks, but their advice always led back to avoiding decisions. Now, we've hired someone who understands the risk equation much better. They've worked in various companies and are not just identifying risks but also suggesting solutions based on industry practices. They've become a true thought partner, which has made a tremendous difference.

When we launched, we received significant early inbound interest. As we began with a product-led growth model, much of this inbound came from enterprise clients. I recall discussing this with a 60Z team when they joined us. Our initial thought was to be an engineering company, avoiding traditional salespeople. We wanted engineers to handle sales. We hired one traditional salesperson and one non-traditional salesperson, an engineer, and instructed them to do sales. As you can imagine, this approach didn't work out. We learned our lesson and now invest in a combined approach: 80% sales and 20% engineering. This was a crucial factor in understanding our customers and their needs, allowing us to work closely with them and bring that feedback into our product and research. This collaboration revealed what we truly needed to develop.

I have many mini-realizations, of course. My co-founder, whom I've known for 15 years since high school, is my best friend. I'm incredibly lucky to have that combination. Jennifer and all our partners have also been instrumental in helping us through transitions, which has been incredible. A recent realization occurred as our company grew to 350 people. This meant our go-to-market team and its incentive structure evolved significantly. Initially, everyone operated on passion, doing what they thought was best for the company. However, as our go-to-market team expanded, we realized the incentive structure truly matters. When building this machine, the people helping create it become part of it. These incentive structures will eventually drive behaviors that might differ from your initial vision if not made extremely clear. Quotas and commissions are effectively lagging indicators of strategy, while strategy leads to future outcomes.