So we crossed 200 million now. 20 months to 100 and then 10 months to 200.
Today we have one of the fastest growing AI companies in the world, ElevenLabs, and I'm so thrilled to welcome their co-founder, Mati, to the hot seat today.
Pre-seed was tough. Pre-seed was hard. I think we spoke with a good amount of double-digit investors, between 30 to 50.
How much did you raise in the pre-seed?
Do you remember the price?
9 million. For us, even from an investor, from a product perspective, it's like especially now, the speed of execution, speed of investing is the only thing you have. Our biggest contract is around 2 million, and they are mostly in a call center, customer support, personal assistance base.
What do you believe that most around you disbelieve?
That you can build a company from Europe at the global scale.
How do you answer the question: why won't OpenAI just do this?
They definitely will, but I think they, you know, they lack the ready to go.
Mati, dude, I cannot wait for this. I've wanted to do this one for a while. So, thank you so much for joining me today.
Harry, thanks for having me. After working on a number of projects together, I'm so happy we can finally speak.
Dude, I am so thrilled we could do this. Now I spoke to so many people beforehand, and I have to say Luke was phenomenal on your team in prepping me so much. But I do want to start a little bit with the origins. So you grew up in Poland, and this was Luke's. How did growing up in Poland impact your mindset towards the world and company building first?
You know, in Poland, it's a very different, much smaller world. So for growing up, I was in a very lucky position where I was in the suburbs of Warsaw, then went to high school in Warsaw, then kind of started seeing more of that world, which was of course incredible because that kind of opens your eyes to what's possible, that there's so much world beyond what you've seen in the smaller cities. And I think in a similar way as you think about building ElevenLabs now, it's the scale of what we don't know is still ahead of us, and this is what's exciting, that we know that if we climb those additional hills, the additional mountains, we will see increasingly more.
And I think the second thing was I saw the transition where in the suburbs of Warsaw, in a public school, so a school where you would have just any kids that would live in the area, and then in high school, and where I met my co-founder, you would have a slightly different crowd, like a crowd of people that would win competitions, need to go through a few steps together, and suddenly that density of talent was the most motivating factor to explore more, learn more and do more, which now we are trying to of course replicate that at ElevenLabs, and any other company I'm sure too, it's like you try to keep that density as high as possible because at the end of the day that's what's most motivating, being part of this incredible set of people.
I was most inspired or given hunger because my dad always said like when we went into Chelsea, like, oh, it's a better life here, and like this is where people who've won live, and I didn't live there, and I wanted to live there, and it fed me with this like insane hunger. Where did your insane hunger come from in those early days?
The early days. I mean, definitely it was, you know, the combination of family. My older brother kind of trailblazed going abroad to study, then motivated us to like, you need to do the same, this is really helpful. And then the second, it was really this community of people in high school, meeting my co-founder, who was extremely smart, meeting some of our close friends, and like kind of going through this cycle of motivating each other. It's like, yes, we want to study at the best university. Yes, we want to go deeper and crush this exam. And I think this was a huge, huge element for all of us, just going after it and knowing that maybe it's possible. At the time, everything felt distant, and it's still, you know, many of the things do, but I think without that, the kind of motivation through the people would not be possible.
And speaking of the motivation through the people, you have a very special relationship with your co-founder, Piotr.
Yeah. So I have to ask then, you come together. How does the idea for ElevenLabs come to be? Was it truly inspired by bad movie dubbing?
There was two things that coincided for us starting ElevenLabs. One was through the years prior, he worked at Google, I worked at Palantir. We would meet for a hack weekend project together. So we would try to explore a new technology. One was in a recommendation space. One during the crypto hype, we built a crypto risk analyzer, which wasn't very easy and didn't work very well. And then once we did the project in audio, which was the idea was, can it analyze how you speak and give you tips on how to improve speaking? And that kind of opened our eyes to what's possible in the technology space there.
And then fast forward later that year, so we did the hack weekend project in early 2021. In late 2021, the moment hit from Poland, which is, wow, all movies are still dubbed where you have all the voices from the original, whether it's male or female voice, narrated with one single character. So you have one voice narrating all the characters in a flat, no emotional way. So it's like an audiobook reading of a movie. It's, as you can imagine, a terrible experience and something that we knew, combining the experience from working on audio and now knowing this is a problem that, okay, this will change a few years from now as you imagine that kind of advancement, all the voices will have original emotions, intonation, sound incredible. And that kind of kickstarted this idea. And of course then it expanded to, okay, we need to fix the research layer to actually make it happen.
Then we started with a lot of our creative platform work around ElevenLabs and then expanded to agentic platform work of how now the interaction is shifting where voice is this big interface for the technology around us. So while it was very dubbing specific originally, it kind of expanded through to everything voice today.
I mean, it must be the most boring movie experience having one single voice for everything.
It's terrible. It's terrible.
So what do you do when you land on the idea then in terms of your next steps? You move into the research layer and determine whether it's possible to actually do this.
The way we've approached this was we first tried to do two things in parallel. My co-founder was, okay, can we use existing technology, stitch it together and create a dub of a movie in a better way? And it quickly transpired that you get a good effect, but it wasn't brilliant. So to actually fix it, let's take a step back and fix one of those components and make it great. At the same time, when he was doing research, my mission was, does anybody actually want that dubbing product? So I was emailing all the YouTubers, getting the emails, scraping them, and trying to message them, hey, if we had a dubbing product to make your movies available in all languages, would you be interested in that? And there was roughly, it was like 15% reply rate initially from the first batches that we sent. All were personalized. I mean, we sent thousands of them, but the interest was lackluster. All of them were like, oh, I don't fully believe this is possible. Can you send us a sample? It would be great, but how will I operationalize this? YouTube doesn't support it. So it was somewhat of an excitement, but not like a burning problem. I need to solve it.
But then the second thing happened. So we started sending the samples. We started speaking more with the YouTubers, and it quickly came that what they actually want help with is much simpler. They want to post-produce and correct if somebody said something incorrectly. They want to understand how the script will sound before they need to produce it. They might not want to speak at all, and they want to voice over the movie. So a super simple problem which didn't include anything with language engineering. And of course, Piotr, as he dived into the research, realized, and he's an incredibly smart researcher, that you can actually build a completely new text-to-speech model which will be a lot more emotional, innovative, that will make that narration a lot better. So let's put a dubbing problem aside, and as we see, there's this huge other problem with a lot of the creators. Let's solve this problem first and bring that research just on the text-to-speech layer, and then, you know, kind of as usual, as we dive.
And that's only possible by creating your own models?
Yeah, quick answer is yes. All the models at the time, you could tell immediately, it's like uncanny valley, not very good. You cannot replicate voices.
Dude, how should I know as an investor? Again, I use this show just to get better as an investor. How should I know as an investor whether a problem needs its own model or whether you can just leverage alternative existing models?
Wow, I think the answer will definitely change now to like 2022. So that was the year where nobody yet was thinking about AI. This was kind of downfall of metaverse, crypto days. So the attention to the models was not known in the public eye. You know, ChatGPT happened beginning of 2023. That's where it kind of all started spiking across. So at the time, you didn't really have a choice. You knew as a user, as an investor, that what existed out in the market just wasn't very good. And then there's more of a question, will this team be able to solve and create something better?
If you were creating ElevenLabs today, what would you do architecturally differently given the architecture that we have today?
I think still a current question is to what extent you continue on that kind of single modality space where at a time it was mostly you train a dedicated model for speech or dedicated model for image, video, etc., over what is now more of a theme where you kind of train a more of a multimodal approach where you combine reasoning and speech together to create even better speech experience. So that's our most recent generation is effectively that, the Eleven v3. If we had that then, I think we would have probably skipped a few steps and our experience would have been even better.
So off tangent, but I'm just concerned about, bluntly, the plateauing progression of models. And you know, GPT-5 was the embodiment of that. You move from like incredible feature descriptions to a description on cost. And when you have a description on cost and efficiency, it's like, oh, we've hit that. Right? Do you agree? And are we in an element of incrementalism and plateauing in voice?
I think the, so it's, you know, there's a little bit of like, it depends, it depends on the use case, but to make it more concrete, there are use cases like narration. In narration, we think it's plateauing. The new model generations will not make narration of an audio look drastically different. It will still be in a similar quality. But I think in general, the point is right, where if you just do research, eventually it will commoditize, and eventually that advantage you can deliver from research isn't enough. So you really need to build product, and as you think about Eleven, we combine both together.
Do you think we're at a stage where scaling laws no longer equal an equivalent level of progression?
Well, in a biased way, I feel like the progression is still just scratching the surface, where it's just the moment of still seeing that escape curve of AI getting adopted everywhere is just getting started while scaling does continue in the same way. So I do think, but there's a difference there between adoption and development. So I agree with you there in terms of the adoption, but actually the development progression.
I think, you know, it's probably a different answer on LLM space and a different answer in the voice space. I think in voice space, you are still in, it's still not a slower rate. And then there's an interesting variation of like, as they all combine, what does that mean for the world? Like it's slightly deeper understanding of everything around you. But I think in voice, you still will see a pretty quick curve on the LLMs. I agree it's probably a little bit flattening to some extent.
I spoke to Andrew Reed before the show. I had a great chat to him last night, and he said, ask him about like, why can't OpenAI do it? And I remember chatting to Kieran about this. Kieran obviously being my partner, and there was two reasons why, you know, we didn't work together. One is that small check which you very kindly offered. And the other was like, no offense, it was super freaking early, and it was like, why wouldn't OpenAI do this? That, no offense, two very young guys in London are doing. Like, how is this not going to be done by them? How do you answer the question, well, duh, why won't OpenAI just do this?
They definitely will do something, but I think they, you know, they lack the genius that I'm happy my co-founder and our team has. But I think the true, longer answer is one is focus. You really, I think in the early days especially, you know, when we spoke then, there was just so many different things you could do in AI space, and I think we took this bet that we want to really own and win in the voice AI research and product space. All our work is directly tied to voice.
Then the second thing is I actually think the number of researchers in the world working on voice and being exceptional is super small, probably like 50 to 100 people at like this top level. And Piotr and the people around him were able, so Piotr was able to assemble one of the best teams in the space. I think we have five to 10 people that are in the top 100. Like it's a mighty team, which I'm kind of time and time again surprised how incredible they are. And to make it specific, I mean, text-to-speech was one that kind of have blown everything out of the water. Then speech-to-text is beating OpenAI, Gemini on benchmarks. Now music, which is again something that no big company has yet been able to crack or do. So it's an incredibly mighty team.
And then even if I think if OpenAI does actually do something on research, and I think they are trying, then that product layer is where you really need that big advantage. If you are in a creative space, if you do narration, if you do voiceover, if you're a dub, you go through so many additional steps to really make it perfect. We do that well. If you're building a voice agent or conversational agent, you need to bring knowledge base integrations, functions. You need to then deploy, test, evaluate, monitor. All of these pieces are coming together in a platform. And I think OpenAI is not investing as much time. They could, but they aren't. So combination of mighty research team, speed of execution, and then actually focusing on the use case in a deeper way.
You mentioned that kind of the mighty team. Those people are very valuable people, and an OpenAI or an Anthropic or a Meta would pay a lot of money, it would seem, for people like that. Do you worry about the war for talent, and how do you think about retaining talent when there's hundreds of millions of dollars going across the table for them?
I think the talent, especially on the research side, the impact that those people can create is out of scale across any of those companies, especially in the early days. Of course, to some extent, as I think about Meta and others, they are of course paying for the know-how as much as they are paying for the raw talent too, where getting those early people gets you some insight into the models and architecture that you can then bring across and accelerate. So in the early days, it's more valuable than late.
But I think in our case, the reason I think it's still valuable is one, the upside for ElevenLabs is just getting started. So I hope we'll be able to compete with any of those companies in the future at the same scale. Two is what you do have, which I think is important as I think about any of those companies, is like how close you are from developing research to actually deploying research. And at ElevenLabs, if you are actually creating new models, they get to production are some of the most important things that our products work on almost immediately. So the gap is super quick. Well, you know, bigger companies, you go through the usual corporate red tape to get any of that work done.
And then three, I think we now have like a very small and mighty team that can learn from each other and move pretty quickly. I don't think there's a guarantee you get that in some of the other companies. I think they are optimizing for a lot of people, but not necessarily the right people.
We're going to go to team size later. I do want to stick with some chronology. So many questions that I want to ask you, but some chronology. We mentioned that kind of the V1 and building to beta, and I think it was, where was it, 2023 that you raised the pre-seed, Jan 23. How did the fundraise process go? You're two young guys saying, hey, we're going to do this. How did the pre-seed fundraise go?
Pre-seed was tough. Pre-seed was hard. It was, you know, similar question that you asked was like, how will we fix research? The second question, which was very interesting, we think the market is very small for what you are solving, which at the time was like, you know, it's like a combination of AI voice, like nobody was thinking about that. So that was surprising to us, which we were very disagreeable. And third, it was defensibility. Will it actually be better than what incumbents from the Big Tech companies will actually solve, or how will you outcompete on long term? So these are the three. So it was tough. I think we spoke with a good amount of double-digit investors.
Double digits of investors, and they said no?
Yeah. It was like with between 30 to 50, and it was double hard at the time because we, early 2022, we got an offer from one of the accelerators in US, not YC, and we were thinking like, should we take it or not? And we decided, no, we are rejecting. We think we can create something more valuable, and we don't need that help at the time. So decided to go independent, and this triggered this more stressful time where suddenly now you need to raise money. We started spending a little bit more on GPUs. We hired the first few people. It's like all of that was going from our savings, from which we are lucky to have from Google and Palantir. But I was like, okay, now it's actually getting a little bit more risky, and we want to double down. We want to invest even more. So we need money, and yeah, it was.
How much did you raise in the pre-seed?
2 million. Do you remember the price?
9 million post. The amount was exactly 11% of the equity that the first investor was buying, and then the other ones were layered in.
So it's like just over a million. And this was 2023?
This was not. No, but dude, I'm so used to like stories like this on the show where it's like HubSpot, and you're like, when was it? Oh, 2006 or 2008 or 9 or whatever. But this was recently.
Oh yeah. Yeah, it was only four years ago, which I mean, it was so, you know, in the end we found, we had Concept Ventures here in UK, formerly Episode 1, we had one of friends from Oxford, Peter Czaban, who invested, was a co-founder of Polkadot, which is the cryptocurrency. So we had like a good set of early believers, but to get there, it took us a bit longer.
So you raised the 2 million, then what happens? Then you start building out the team, you start investing in GPUs?
Exactly that. The main reason we raised was to accelerate what we would want to do. First one was building a small data center, so we built one, and back then it was I think in Poland that we invested a few, and then quickly after moved and started buying some in US. And then team, we started hiring, not many, but it felt like a lot at the time. It was like additional two people, I think.
Don't go too far there. Mati, did you, because you kind of did the pre-seed raise and the beta launch at the same time. Did you have product-market fit on the beta launch? Were there early signs that it was working?
Oh, yeah. That's a... Okay, so got the good timeline piece there where we actually raised the round in mid, like Q3 of 2022, but we announced it in Q1 of 2023. And the reason we do that, and we do that all the time, is that we don't want to announce a round just for the sake of announcing a round. Our philosophy was always the round should have another purpose, which is bring the product out and help you get the product into the users, celebrate a set of customers to show that you have arrived in a specific sector, bring a new research model into play. So every round that we would do would always tie into a product announcement. So we actually held on for announcement for a few months until we had our beta release, and then we triggered, which was January 2023.
But to your question, initially when we worked on dubbing and that first early days, and we were like emailing, we didn't have product-market fit. It was like very clear that, you know, people were slow to reply, then we sent samples, they weren't engaging. So we didn't through the 2022 period. And then when we shifted from dubbing into narration, voiceovers, then the product-market signal started to hit. Maybe it's there. We had a first, and I remember this thing, three things that happened. First, we did a blog post around the first AI that can laugh, and we sent samples that got picked up by newsletters, and people were like, wow, this is incredible. The next day we had like a thousand people on our waiting list.
Then the second thing happened. We invited the first 100 or so users to ElevenLabs to test it out. And we had this audiobook author who joined the platform. Our platform is like this small text box where you can type, tweet length, and narrate it. And he would copy-paste his entire book 500 times, download it, stitch it together. Then he released it. He released it on the platform. At the time, AI was banned. It passed through as human content, and then it started getting reviews that it's great. And then he came back saying, I want my other book author friends to do it too. So I was like, okay, we are onto something. And then we launched in January 2023 publicly. We knew we started getting more of that signal that creators, narrators love the work. And to be honest, I think from that moment onwards, we saw like a clear momentum.
And then there was like, you know, more events after the media picked it up, then more creators picked it up. But the moment of release, we knew that it's valuable. And then maybe last thing on that, it's like the product-market fit concept was, you know, for us, we never like fully, I think, know what this actually means. Where we knew users just loved our product at the time, but I wouldn't call it at the time we had product-market fit as we were thinking of like, okay, how do we make sure it's providing value for the next 5 to 10 years? Then we'll decide this will be like something where we would call, okay, this is clear product-market fit. We think this is like self-sustaining for as long as we go into the future. So we wouldn't go by that definition. I think we are now closer to that, but we are still like, you know, still know we can create so much more value.
So many things I want to unpack there. You said about kind of the timing of your announcements and aligning it to actually kind of material on news items. Do you have any big lessons on announcements, how to do launches that you found particularly work well that other founders should know?
So that's the first one, 100%. Like, for us, the making announcement close to something big you want to announce from the product, from the user, from hiring perspective, tie it with that. I don't think celebrating the number itself is the right way. I mean, in a way, it's like you're giving away the company. So it's like almost like, what are you doing it for? And you want to show as much of that as possible. So that was the first one.
Second is, I think this is very easy to optimize for, and maybe I'll cover it up front, depends of course on whether you are self-serve as creators, developers, or optimizing for enterprise and sales. But second thing would be you really want to focus on how you actually get the users rather than just the media PR element. I don't think the latter is actually as valuable as it seems to a first-time founder.
For us, I remember when we did the beta, we were speaking with some of the bigger publications, and one rejected us, the other one accepted us, and we felt like it's a big deal. Then we did all the prep for the interview, we did the actual news, and then it was posted. We held like everything around this post that they told us the deadline for, and it had no impact. I think it, I mean, it like probably a few more people read about us from the investor scene, but we didn't really care about it at the time. We were like all about users.
What actually worked was working with the newsletters that were talking about AI, working with our friends from YouTube community that then posted on Discord that we are opening this up. Our Discord community was one of the most valuable. Reddit, those people picked it up quicker than anyone else. Hacker News posting there, like all of those were immensely more valuable. And we since then always spend a lot more time on any of the actual forums where our users are rather than where they are not. I think we grossly overestimate the importance of traditional press.
Oh, 100%. I remember when I was first in TechCrunch, I was like, I'm gonna be like viral. Everything's gonna work. Yeah. And then it happens, and you're like, you get like three followers. And this was when it was much bigger, and it's like, so I completely agree with you there. The grassroots. The other element kind of tied to that is how do you think about fundraising, and this is advice to founders, fundraising pre or post a launch?
What frequently happens if the launch goes well is you will get more interest, and I think you'll get into it from investors, from events, from media to a large extent. This is like the time you should really focus on getting the product, and kind of all of that is like a distraction to a large extent. Like, you know, the first time we launched, that kind of happened. We started getting so much of that, and I'm sure I did the mistake then of like picking up a lot of those and doing more events than we should, where like the best course of action would have been to just spend even more time with users.
And then kind of early enterprises started getting interested, so more of enterprises. But I think on the investor side, I think you want to keep, you want to line up investors. So you do want to tell them like, okay, I'm not raising now, I'm going to reconsider whether we need more capital in Q3 or Q2, whatever is the time period. So you want to line them up, and then when you actually need the money, then is the best time to actually re-engage with them. But I would not, I think it's a waste of time to kind of be in this like continuous fundraising mode. It's, you know, distractive. You need to have conversation. It's not useful.
And do you choose three that you'd like to have that you engage with in between cycles, or do you not even do that? You just say, hey, I'm heads down on building. I'll come to you when I'm ready.
Today, less. At the time, at the time we, no, good question in the sense of we do have a few investors that we think would be valuable that we might not have today in the kind of the space. I mean, I'm very happy we are kind of working together.
But no, that's the truth is 100%. Through the time, we would have few investors that we wouldn't try to proactively reach out, but if they reached out through any of the conversations, we would try to engage. But the engagement there wouldn't be like, hey, we might be raising. It would be much more of like, hey, can you help me with introduction to X? Hey, can you help me with this hiring problem? And this has worked tremendously well through the work where I find like the investors are genuinely keen to help if you find the right person and you have a very clear problem and you don't try to like overuse their time too. Like you come with one, two things that are concrete, useful, and then it's, they know as well that if you are growing as a company, it's good for them to show the help too. So we've done that.
It's also a good litmus test to see if they genuinely are interested in being there or if it's just platitudes of niceness.
Before, it's like kind of the second thing where you do get the interest from investors. I think this is the best time to actually test whether they can be helpful. And before you accept any term sheet, any money, it's like, now can you help me with angels that you want on the cap table? Do you want the introduction then? That's the best time.
Any advice on angel selection for founders in the early days?
The way we approach this is like, so we had standard of course the venture capital to whether it was the help of their network, with the brand, with making sure we arrived in a specific region. But on angels, we would usually optimize for, do they have domain expertise that we don't have? That's one category. Second, can they help us validate ourselves in specific circles we might not have access to? So like if you are an AI founder, it may be valuable to have a set of AI founders in there. So you are part of the same events, the same conversations. And then of course in our case, it was can we have some of the go-to-market expertise where we didn't have. We had a lot of self-serve, some volunteer experience, but the sales-led, how can we bring that in place as well.
You mentioned starting with the author, suddenly it was going well, and you were seeing this great groundswell and great adoption. And this is kind of January to June 23, I believe. And then in June 23, you raised like $19 million from Brian Koo and Andreessen, and then Nat and Daniel from, I can't remember what it's called, NFDG or something.
Yeah, NFDG. It's Nat Friedman and Daniel Gross.
NFDG, wonderful, brilliant name. And that was quite a shock because it wasn't a London fund, it wasn't Index, it wasn't Accel. How did that round come to be?
Yeah, NFDG is now, I feel like it's a good name, but I was pitching on changing that to like a different name for a while. Maybe now it's not possible anymore. But so it was like the interest started around kind of March 2023. We had a lot of the investors that were in close conversation in the past like re-approach us. We effectively would be a little bit more in the waiting mode, like let's see how we get through and let's optimize for the partners that we truly want.
And what we truly wanted was a combination of making sure that people globally in SF trust and know that we have arrived, like we are a company that can be trusted, is building something ambitious. And then of course people that we would admire to be able to work with that created something special. And that was definitely in that latter category. And it was a very fun thing because he approached us. We previously...
Does he just DM you? Like how does he just approach you?
He emailed us. He emailed us, but then I met him here in London. He flew in, and it was like a weird setup where he emailed me, but it wasn't clear whether it will happen, and it happened at the last time. And then I see him in London in one of the hotels where he was staying, and the first thing he was saying was, which was the only investor ever to do that, was, so I tested your APIs, and this thing doesn't work, this works, voice here wasn't good, the stability wasn't very clear parameter. So he was the only person that tested our APIs, decided that it was actually valuable what we are doing, gave me feedback on like what we should change, and then told me like, and I'm keen to invest, which of course triggered a conversation of what's the vision, what we are after.
And what's special about that is that you can tell him very little, and he kind of has this beautiful ability of infilling all the context around it, comparing that to any of the past experiences, and kind of bringing any of those experiences that then might be relevant to you. But he gets it immediately. So from the first conversation, he knew kind of exactly what we are after and challenged us with the right questions from the start. And then he moved very quickly. He was very keen to partner with us. We were keen to partner with him. We were trying not to show it too much, but we were.
And then at the same time, we know that Nat and Daniel were more perceived as the angels, while we admire them from their GitHub and summer days. That was like still like an angel category for many of the externals. So we knew this is our chance to bring a huge investor to actually position us as a top company. And that's, we spoke with all the classic tier one funds, and a16z was just one of the most...
We spoke with few funds in London. Yes.
And then Brian flew to London.
And Brian flew to London. So Brian, I mean, a16z was phenomenal. They did exactly what we spoke about, which was they've shown us that they cared before they invested. They introduced us to incredible people, to some celebrities to work on the voice licensing. So they were like on it for two weeks prior or like a week prior already to help out in any way they could. And then Brian flew to London, met with us, and then we signed a preliminary term sheet, which was interesting. He flew in. We're negotiating also on my phone call to co-founders like, hey, what about these terms? Can we accept this? And Piotr would give guidance of where is the hard line, what we can do. And then on speaker, we would sign together and finalize.
Do you think speed really is a differentiator as an investor in winning?
Oh yeah. Speed is like for us, even from an investor, from a product perspective, it's like you need to, especially now, the speed of execution, speed of investing is the only thing you have.
This is what pisses me off. I'm happy in that world. I can move supremely fast and invest a lot of money. But what I find more and more with founders is they want to run a roadshow and they want to do a 10-day process meeting investors and then have another week to decide which term sheets they want to take. I'm willing to give you a term sheet tonight, save 17 days. But it's more and more common that founders actually want to run roadshows. How do you think about that?
Yeah, this is, I'm now in a lot of great companies get built on ElevenLabs. So I get to be in the conversations now more frequently where they would raise money from our investors. Like recently we had a company raise a Series A that has built a healthcare voice bot which would effectively help patients calm them through the conversations in an incredible way. Then there's like another company building for healthcare customer support. So quite a few, and now I'm in the conversations of them raising the money, which is an interesting one.
And then usually it's clear that some of the people don't know fully what they are actually optimizing for, and it's good to talk for, are they optimizing for the valuation? Are they optimizing for dilution? Are they optimizing for the right brand name to help them out? Is it the network of people that this brand name has? Is it the specific partner? Is it something different? And I think it's like pretty distributed. So when I speak usually with founders, or our investors ask me to speak with the founders, it's usually that question like, what is it actually that you care about for the thing? What do you miss? And then it surfaces pretty quickly whether they are going after just the roadshow is there to bump it up because they think the value that the term sheet they got is just not valuable, or is it that they actually want someone else but they want to keep the optionality of having the first one while running the process. But you know, all the traditional simple things.
I'm a romantic. I hate this idea of a roadshow where it's like I'm going to meet everyone and then I'm going to compare term sheets and then I'm going to decide. It's like, what happened to the partnership?
Well, but it's, you know, the truth is the other way is true where a lot of the investors want to give the term sheet, but that is before any show of partnership. So how do you know that this is a good partner? How do you know whether they are giving you good terms? Especially you're a first-time founder, you never run this, you don't know how much you are valued. You need to have some relative comparison, and this can happen for conversations with other companies, other founders, or comparing yourself to other companies in the field, or running some form of a process.
Roadshow probably sounds pretty bad because it sounds like a lot. You probably like, the way we approached fundraising was always, you know, especially in the early days, like cue a few investors we really care about, and first speak with maybe one or two that maybe aren't a priority so we understand whether we are at all kind of perceived in the, that the kind of the messaging that we want to put across makes sense. If it does, move straight to the top ones.
How do you feel about exploding term sheets?
I don't like them, but we got a few. We had a few through.
I do. I think I do. At the time I didn't, but I do. It's like, I mean, I don't like them still, but we, I won't name the people, but so one example was a, let's say a smaller fund, and they felt that if they get us the term sheet, we'll use that to bump other term sheets.
Very often happens, and that it does happen.
So I got it. It was like, you know, and they came from the approach of like, if it goes too high, they won't be able to invest, and they don't want to be part of that. So that happened.
And of course, being an investor, being the first term sheet's the worst because you're used as the stalking horse where they then go, I've got a term sheet, I've got a term sheet from a great fund. They never name the fund, and you get no credit for being the first, and then you're just trampled on by everyone else that come.
That's what I don't like, founders that are trying to do that too much, like that if they are trying to get the first term sheet to bump the other ones, I think it's a wrong approach. I think you should like, it's, you know, okay, if it's like order of magnitude off, it's probably wrong, or it's like, you know, 5x, then yes, but if it's like plus minus 20%, that's like a wrong optimization. There's just so much more value that investors, partners can give than the money itself that I think should be optimized for.
You've got Andreessen introducing you to celebrities, and you've got Nat and Daniel, two of the brightest minds in AI who are querying APIs that no one else does. Are the Americans just playing a different game to the European VCs?
We love our partners. I mean, we have, and team is so incredible, the network. I mean, now SEV with Andrew is just another level too. Iconic team, also crazy help. NEA too is just like bringing...
They are all US. So I think they are playing a different game. I think, you know, they are all from our experience so far a lot more keen to take the risk. Like, you know, even our conversations are always like, how do we bet bigger rather than optimize for the negatives.
The one check I always try to do with Piotr for any of the partners that we bring beyond the capital, network, brand, is always for the person you partner with, how will they behave if it's not going well? So we usually try to get any of the back references for how did they behave when the company was failing. And all of the ones that we had are in, and they're very positive, like they actually were really on the side of the founder, helpful when it was not going well, even more so than when it was going well.
And I was doing the same check with some of the checks with other companies, and maybe this is getting used to that, like, you know, in US, the kind of the set of companies that went up and down is so high, but in some of the investors we spoke in Europe, that check didn't yield the good results.
I had one with Brian Koo that didn't go well, and I have to say Brian was the best. He was the lead investor in the round, and he got the money back for everyone, and that's rare, and he really put in the work to do it well and to do it quickly. I was really impressed with him in that.
No, exactly. Brian was one of the ones on our checks came up strongly. Jennifer, too. It's like all of them. It's just crazy how...
Okay. $19 million in the account. Most people suddenly then go on a hiring spree — and not a two-person hiring spree, but like a proper hiring spree. But you favor very small teams. You've said this to me before, but Luke said it to me. Why do you favor really small teams?
I think the first piece is that more people frequently doesn't fix the problem. It's like you don't need that many people to do something special. That's the first thing. Second is, by keeping — and as an organization today we are, you know, 250 people, but really it's more like 20 teams going to market of like five to 10 people that are just executing on specific projects where they have higher ownership. They can move extremely quickly. They see the results and iterate with reality a lot quicker to improve that. And I think this just works in such a beautiful, beautiful way. Of course, there are other challenges with this, but...
Are those 20 teams organized by function or by project?
A little bit depends. In the product, by like a product area. So we have a team working on our studio interface team that will be responsible for all the core experience when you log in, a team responsible for the entire voice agent suite, and within that there's a team working on some of the more enterprise components, another team working on some of the self-serve elements. So all of the teams will be organized there on the product area, and then the other parts we try to shard it pretty quickly. So, you know, we will have a separate team for talent, of course, then the team for people, and they will have a high degree of independence when they actually execute, which helps.
You mentioned 250 there. It's small given the size of company in many ways, but it's also still 250 people. When was the company culture the worst and what did you learn from that?
You know, there was like moments where I could feel this kind of tension between the go-to-market, research, engineering. And this is like a specific story where — so I mean it's a super lucky position where we develop our models and bring them to the customers, which of course makes it very special. We frequently will show for the very first time to the world something that was not possible and is possible now in 2023, late 2023. So we did a text-to-speech, we then did voices so you could recreate a voice, then we created like a basic way for speech-to-text internally. We had all the components that we originally were speaking about to create dubbing, and we haven't done dubbing yet publicly, and we gave all the components to some of our customers so they could use text-to-speech, the voices, and there was like a lot of, you know, pressure that like it's a valuable technology we can give to our enterprise clients. And one of the enterprise clients, we told them that we are planning to launch our dubbing solution combining those components later in that — it was, I think, September, later in that month. And they took the components and released dubbing two weeks before we did. And that was one of like the low moments for me, for my co-founder, for the entire team, because dubbing was our story. We told that to everyone. We knew that this is the thing we want to solve first. We had all the components to be able to do it. We're just waiting to optimize it to make it perfect. And then suddenly this partner released it and got all the attention. So all the media, Twitter, all the users were like, 'Wow, this is the most incredible thing that has happened. How amazing that you can speak in another language and still sound the same.' And you could feel the morale in the company was low. People were like, you know, we spoke about this. We knew for like almost two years that this is something we want to solve, and how did it happen that the customer had this and solved this first? So research engineering wasn't happy, like how did they do it before with the components we gave them. The go-to-market wasn't happy because like how did we sell this to the customer? Now we are like all the potential clients we could have had isn't good. And of course me and Piotr were like, hey, this was our idea. Like why would you do any of the partnerships if all of that is like kind of given to another company?
I don't know if I can mention. And to their credit, I don't think the time like that we told them it was actually a factor. I think they were just trying to do something quick, and like we had a lot of calls with them those days because we were thinking like, what do we do? Does the contract give us flexibility to not continue? So we spoke through with them of like, hey, why did you launch? I don't think it was dictated by our timeline, but the timeline was very close to each other, and they told us that they thought it's a good hack weekend idea. So they gave it to their intern, and the intern has built that project, and then it exploded. But it did give them like tens of millions of revenue over that period of time. So it was significant.
What's your biggest piece of advice to a founder who has a moment like that where you just feel the air come out of the company? How do you inflate the company again after such a damaging blow?
I don't think the first reaction should be that like, hey, everything is fine. I think you should be authentic and tell the team what you are feeling, like how has this happened, and actually go through like what have we done wrong. And you know, that frustration was clear. I don't think we — I think it was valuable for us to talk through like we are angry at ourselves, this is wrong, and then as we move from that stage, like yes we are angry, what are we actually going to do about this, and move to that second stage very quickly. But I think, you know, I think I've done a mistake sometimes of just going to the second stage, like kind of it's not a problem, let's just go through and solve it. I think it's actually super valuable to talk through what has happened. But then you know you need to go into the second thing, and then in the long term that will work out. You know, then the relentless execution will prove itself out and show it to the world. And the worse if you repeat the mistake, then someone needs to feel the repercussions. But if you've learned from that and haven't repeated the mistake, it's fine.
Given the commoditization of a lot of technology today, do you feel speed of execution is the core differentiator between those that will win versus those that won't? Or is it quality of research, access to GPUs? How do you think about that?
I think it's both. I think it's both. I think it's, you know, the way we approach this in the company is we have research, product, and the ecosystem that we built, which is a combination of distribution and brand. But research for us is a head start. We are investing in research. We'll continue investing in research. We want to be the best across the voice technologies. But all this gives to us is advantage over the competition for the next year, two, maybe three.
How far ahead are you compared to the competition, do you think?
I think it depends on the use case, but six to 12 months, I would say, depending on the space that we are in.
Do you think that's a lot or not?
I think that's a lot. I think that gives us — so that research piece of six to 12 months is enough for us to then do the second thing well, which we do in parallel from the start, is build phenomenal product experience.
What percent of your revenue do you think you spend on compute?
So we've built our own data centers for training, and then for inference we of course use some of our great partners across the traditional...
Why build your own data centers? Most people will just use a CoreWeave or use Nvidia or whatever that is. Why build your own data centers?
So we did the math. In our case it was if we assume we continue training the models the way we want to, so continuously a lot of those models. Then second, for the data transfers and as we think about continuously bringing more data, we will likely on a two-year horizon break even for having our own and not renting. Assuming of course, you know, you need to assume some improvement in the GPU infrastructure, but assuming that wasn't too high of an improvement, we would have been successful, and we were. I think the ROI made sense there, and now it pays dividends because we can just do more experiments quicker. And you know, this can still be wrong at some point, there might be innovation that kind of breaks that equation, but for the current time it just makes more sense, more control.
A lot of investors are talking about the poor unit economics of a lot of companies that we see today, whether it's your Replit's or your Lovables or any of companies like this, well, any kind of application-layer companies period really, they generally have pretty tough unit economics generally speaking. Do you think that is a fair criticism or do you think it is simply shortsighted given the changes we will likely see in the cost structure?
Of course, the unit dynamics in most of those cases that you mentioned is pretty poor, but I think the strategy is that yes, one, the models will optimize in cost, and then two, they will be the brands that customers trust, and then they can actually use a lot of the signal back. And I think to ElevenLabs' example too, I think our unit economics is much healthier than most of those companies. We control our research, our product and distribution, but it's still if we had a new model that we need to release ourselves, we would optimize less for the cost structure so we can get that magic out quicker than other competitions even can think about creating their own model.
But to be clear, you think the concerns around margin are overblown and not justified.
Well, it's a risky business where, you know, like there will be a winner from some of — whether it's, I love what Lovable is doing with Anton, but Replit, Vercel are all incredible companies too, or products too, that are doing some great work. I do think at least one, maybe more than one, will create something special. The market is so big that all of them can create something special. But yes, there will be definitely a loser in that mix too, and that company and the margins that they are carrying does just not make long-term sense, and probably the same for coding apps. But I think it's a cool bet. I think it's cool not even like cool from the sense of like it's, you know, it's a nice application, but it's like it's an ambitious bet that they are trying to win with the biggest companies in the world that are trying or are trying with Google, with Firebase — sorry, I don't remember their name of their product — but they are trying, and the Lovables of the world are winning.
Goes to your point on brand. Neither of us can remember Google's product name, so I mean it's just interesting.
Anton is a phenomenal marketing person too, like the founder, the way he's doing is amazing.
One thing that's also really interesting is you were very horizontal in your customer from day one, which respectfully I would advise you very much not to be. I think Philippe was telling me he was advising you the same, be much more targeted in your ICP and have a clear mind of who that customer and user is, and you were much more broad and horizontal. And if you're advising founders, how do you tell them when they're launching from day one whether to be horizontal versus whether to be vertically specific?
I think if you're launching something very new, very different, and you don't yet fully know, you know, like a subset of customer base, but you know that there is a bigger one, I think horizontal is completely fine. If you know and you have domain expertise and this is the category you're trying to win, I think go vertical.
You say go vertical, that normally when you have verticals you have kind of a maturing of organization, you have titles. Titles is something that you decided to get rid of. This is very counter to traditional structures. Why is it better to not have titles?
Yeah, there are some incredible companies that of course did similar, Stripe being one example. But in our case, it was especially then a super small team. So we have a team of five, a lot of people join, and what we wanted to optimize for is like one, the main thing that matters is the impact. You can join and you can be the most impactful person from day zero in any of those teams. So the title shouldn't define what is your level of decisions. Second thing with the small teams that happens is that you have a lot of those small teams. So if you start looking at who gets the title or not, it just becomes a distraction, given that teams are like those small units that just keep executing. So that's the second thing. And the third thing, we wanted to also make it very clear in that mix, where and still do, if you are joining ElevenLabs today, you can transition to being a leader of any team of any function super quickly, and titles felt limiting to that. We felt like people joining in and seeing the kind of the wider set of organization already having a set of roles defined for them would make this a little bit harder, and that's not — like you can have a lower tenure and be a manager of people with much longer tenure if you are the right person. So titles we felt were taking away from that. Well, at the same time, we have a good structure internally within the sub-teams, like who is the current lead of that team that will make the decision if people cannot agree on what's the right path. But that person can, and it's not guaranteed that the person will remain a lead forever. And titles in a way, when you give a title, it stays usually forever in any other company.
Speaking of titles and a common criticism of European tech and scaling is that we don't have these titled people who've seen scale significantly. Your VPs of sales who've seen a billion in ARR, you name it. How did you think about broaching that? Do you want to get top US talent here or do you want to grow talent here?
We want to grow talent here. We match a lot of our current talent with the advisors from our network of US investors, and the goal is to grow people. We love growing people, like our approach is almost like if we can, we want to take a bet on the person growing rather than bring them externally, but we know that to grow you need sometimes mentorship from someone that has done it. So we try to tap into all the people that have done it and match them to people across the organization.
Anton said on a show with me recently from Lovable that building in Europe is building on hard mode. Do you agree with that?
I think it is. I think it is building on hard mode. But there is some great advantages too.
I think the first one is the talent here is incredible. And I think you just need to know how to get it.
