ElevenLabs' Carles Reina: Building Moats In AI
VP Revenue on positioning as a vertically integrated AI company, moving upmarket, and advice for founders
👋 Hey friends, I’m Akash! Software Synthesis is where I connect the dots on AI, software and company building strategy. You can reach me at akash@earlybird.com!
Today we’re speaking to Carles Reina of ElevenLabs.
Carles is an operator and angel investor. He was the first investor in ElevenLabs and joined as employee #4 to build the go-to-market function; he currently serves as VP of Revenue for the startup. Carles was an early international employee at Uber, the 6th employee at London unicorn Tractable AI, and built GTM at Sonantic (acq by Spotify). Carles is also a prolific angel investor with +60 investments and founder of the Baobab Ventures syndicate.
In this wide-ranging discussion we cover how ElevenLabs approaches defensibility, moving upmarket to sell to the enterprise, community building, and Carles’ advice for early stage founders.
Listen to an audio version of this conversation (produced with ElevenLabs) here:
One of the most interesting things about ElevenLabs is that you're both a research and an application company, and one of the few vertically integrated players. Can you talk us through why ElevenLabs follows this approach compared to other companies who either just build models or applications?
You start with the research piece - figuring out a really hard problem. Then you realise later that it's not enough to build a good moat over time.
OpenAI and Anthropic have been driving a similar conversation. You can still be the best at the research side, but it gets to a point that research in the next 2-5 years will get commoditised in many ways with open source. The only option you have is working on a full end-to-end spectrum that enables you to build moats around yourself.
The way we think about moats is: you have the research component (which is fundamentally important today), but then you start accumulating data and having product integrations that are end-to-end. You still allow companies to build their own applications, but you want to power some of those products yourself to make it easier for your customers.
Then you start switching towards things like our voice library, where people can make money from their own voices because we share revenues directly with them. You then move from a single product use case to a product suite scenario. That's how you win and stay embedded within the customer ecosystem.
Research companies have all realised the same thing: you start with the earliest, easiest, lowest-value entry level use cases, allowing anyone to build applications on your infrastructure. Then you realise you want to go much deeper than that. That's essentially what we've been doing at Eleven Labs - very focused on voice initially and now on the AI audio spectrum.
This has implications for relationships with your customers who are both consumers of your APIs but then potentially threatened by the applications you're building. How do you see that dynamic?
I've always said that we're building a community of people and companies that love our products and want to build applications on top of it. I see the same scenario with NVIDIA - being the foundational hardware company that anyone is using these days, but essentially driving the majority of the revenue for the big cloud providers.
For us, at the application level, there are thousands or tens of thousands of applications you can build that specifically cater to regions, use cases, and verticals. There's a very small pool of applications that we can build ourselves because we get feedback from customers. We also want to stay focused.
The main reason is you need such specific knowledge per region within each vertical that we don't have and won't have in the coming years. There are a ton of opportunities for other companies to build applications for logistics, healthcare, accounting, or anything else. That's the big opportunity - pick an area and build end-to-end applications leveraging different foundational models to power it.
I do see this evolving in the future - just as Amazon is trying to build their own chips to replace NVIDIA chips, at a practical level, customers are going to end up using a mix. Some customers will still use NVIDIA chips through AWS, and some will use Amazon-built chips. It won't be 99.99% through NVIDIA - maybe 75% NVIDIA, 25% custom-made chips. The same with ElevenLabs - companies will take a risk mitigation approach where a large volume will still run with us, and some volume will run with other companies. Although I would love if everything runs through ElevenLabs; and we can do preferred pricing for that. Our goal is to provide the best customer experience any company can provide and continue delivering the best models in the market with top research; customers will then choose.
Another recent development for ElevenLabs is the move upmarket. There's a lot of debate about how much of the growth trajectories that AI companies have had is in true enterprise versus other tech scale-ups and the mid-market. What have you seen?
Over the past 6 months at Eleven Labs, we've started seeing a big switch. Initially, in Q4 2023 and earlier this year, it was primarily tech companies and startups integrating our technology. What we're seeing now is that non-tech companies are seeking end-to-end products so that they don’t have to build from scratch.
Previously, our ICP was tech companies that wanted to combine components - ElevenLabs for this, Anthropic for that, Deepgram for other things, etc. They would build their own orchestration platform. Today, the majority of companies we're interacting with don't want to build their own applications. They want out-of-the box solutions because they don't have the expertise to do it themselves.
This is a good sign that the market is maturing. Companies have realised that if they don't deploy today, they'll fall behind quickly. Compare it to cloud adoption - there are still companies moving to the cloud today after 15 years(!).
This AI transition is happening within 2 years. Companies that traditionally would have waited 8-10 years to switch are saying, "No, I won't survive if I don't start adopting generative AI now."
We're seeing mid-market logistics companies, hospitals, and 200-employee e-commerce companies all wanting to implement AI. They see the benefits but don't have the resources to do it themselves. This changes how I, at Eleven Labs, approach customers but also gives long-term moats, contracts, and success. Once you secure one of these non-tech customers, they're likely to stay with you - they don't want to change, unlike big tech companies that can easily switch APIs.
The cloud analogy is really stark because compared to the transition to cloud, not adopting Gen AI is a more existential threat.
If you don’t act now, you won't survive as a company. There's no alternative - either you adopt it or you don't. When you see news about AI not delivering ROI or value for enterprise, they're missing the point. The value is there. Companies see it because they are using it, and their competitors know they are using it. So, they are also motivated to deploy generative AI..
The adoption rate increases exponentially with every new company that adopts it. It's different from the early internet days when companies could spend 6-7 years without a website or not being on Google AdWords.
Today, that's not possible with generative AI.
It’s also why the hyperscalers are investing so much in capex because they know the stakes are too high.
Similarly, as an enterprise it's worth over-investing despite early limited ROI signals because everyone else is going to do it.
One example for us is Bertelsmann. Today, about 38 different companies within Bertelsmann use our technology. This is the largest media and entertainment brand in Europe - productions, books, education, movies, all of that. Two years ago, this would have never been possible.
Previously, these enterprises focused on deals that would take 12-18 months with peers at their level. Now they're seeing that working with startups is good because they can move much quicker and see return on investment straight away. They don't want to waste a year negotiating a contract - they want to get it done in 3-5 months.
More companies also want to go public about using ElevenLabs or generative AI, in general, because they see the value in accessibility, content creation tools, and enhancing their teams. You have these big changes in the market because everyone knows everyone is adopting it; and the big players in each industry are adopting it too.
Yes, this is especially true for public companies and IPO candidates like Klarna who need to form a strong narrative of realising margin expansion from AI.
As part of that transition upmarket, you've had to develop partnerships and think about ecosystem building. What does the ecosystem for Eleven Labs look like when serving the enterprise?
I think about ElevenLabs having two sides of a coin.
One side is the businesses, prosumers, or users that require the technology for specific use cases, automations, or applications.
The other side is voices - how do we ensure we have a very large, ideally unlimited number of voices so that depending on what you're building, you can leverage voice X or voice Z.
Building both sides of the equation is what we're focused on today. Nothing is more important than ensuring people who voluntarily put their voices on the Eleven Labs platform get properly compensated. We're the only one doing this today - we have a payment system where anyone can define how much money they want to get paid based on 1,000 credits of usage. Usually, 50,000 credits usage is about one hour of voice content. So artists or anyone can define their rate and get paid weekly once they reach certain thresholds.
That's fundamentally the most important side because you create the right incentives for people to monetise their voices. They feel engaged and included in the benefits of generative AI. The extra benefit is that companies can use thousands of voices to generate unique content and experiences for their customers.
There's also a third layer sitting on top of it - the developer ecosystem. Developers are constantly innovating and building applications and use cases that no one imagined yesterday.
As a company with a PLG foundation, how do you marry that with the sales-led motion as you go upmarket?
It's about momentum. If you want a strong go-to-market motion, you need momentum, which doesn't come without giving people a way to test the product and ensure the experience is simple and automated. We launched on the B2C motion first - anyone can sign up, understand the value proposition, test it and decide if it fits their requirements. Then we can upsell them to different plans - enterprise, business, whatever is relevant for them.
For some customers, we'll help them navigate our products and hand-hold them along the way. That's why we now have two different teams - one focused on B2C/prosumer (which we call Growth), and one purely focused on Enterprise. Even within enterprise, we have segmentation.
Both teams are connected in the middle - PQLs are automatically flagged to the enterprise team, indicating when a company might be ready for a more custom plan. We also do outbound and prospecting for companies that haven't heard of ElevenLabs.
Last year, everyone talked about "AI tourists." Many content creation tools had massive growth and then suffered this year as they had to readjust expectations.
At ElevenLabs, we didn't have the AI tourist problem, but we planned for it. We also realised last year that if we want to build a $100 billion company that's here for the next 100 years, we need a pure B2B motion in place as soon as possible. The longer it takes to build that motion, the more you're stuck with a mentality that won't help on the B2B side. Needs and expectations on the B2C/prosumer market are very different from the B2B ones. So, we aligned the entire company last year to dedicate teams, resources and products to B2B.
We're doing well on the consumer side and will continue investing in it, but we're transitioning toward enterprise. Building specific products, focusing on higher quality, more research allocation, etc. We are spending more time talking to customers and helping them along the way. This year was the full transition, and it's worked really well. We're building sustainable, long-term engines for both B2C/prosumer and enterprise.
There's a lot of discussion about the transition to outcome-based pricing, and having full pay as you go models. What's your view on balancing this with buyers wanting predictability and not having costs run out of control?
Last year, many investors suggested we remove all barriers and move to a pure pay-as-you-go model, like OpenAI. At ElevenLabs, we have preset plans on the website, and for enterprise, we have tier levels based on credit usage with monthly pricing that decreases as you generate more content.
I pushed back hard against pure pay-as-you-go because without any barriers, it becomes a race to the bottom where nobody can predict anything. I'd rather give better discounts but have certainty about the revenues I'm generating so I can allocate the right resources for accounts. When you explain this to customers, it's a no-brainer.
The conversation today is moving towards unlimited plans or something that can generate or render on-prem or on-device. We don't have unlimited plans - no one really does - because we all use the same GPUs, which are extremely expensive. Could there be a world where we don't need such high-end GPUs to run the content? Maybe in 3-4 years.
For smaller GPUs or on-device, it’s a trade-off: do you need something very quick and extremely high quality? Do you need something where latency doesn't matter but quality is crucial? Or do you not care about quality or latency and can render cheaper on CPUs?
Moving towards a credit system with pay-as-you-go and minimum spending was natural - it was created by the big cloud providers. They support PAYG and commitments to give discounts; this has been going on for the past 10 years. We do the same.
Community has been key to ElevenLabs' early growth and continues to be a pillar of strength. Can you share more about how you’ve cultivated such a strong bottoms-up foundation?
We approach it in three different ways.
First is working on events and helping companies build things - hackathons, webinars, in-person sessions, and plenty of community activities.
Second is the ElevenLabs grants program, where we give 11 million credits per month for free for three months to any startup with less than 25 employees, though we've made some exceptions when needed. We created this program last year because, as a startup ourselves, we know the pains of getting started and how expensive it can be. We're lucky to be in a fantastic situation with community support and investor backing, so we wanted to give back. We've given almost 2,000 grants across the world in less than 12 months.
The third element is recognising companies building with our foundation models. If companies or startups are building amazing things, we want to write about them. We proactively reach out or they reach out to us about writing blogs, case studies, partnerships, press coverage, or webinars. We want to help them grow their business because we love what they're doing. We now have a big network and if we can help some startups grow, we should do it.
For example, we partnered with the Scott-Morgan Foundation in the US, which helps people with ALS/MND. If someone lost their voice due to ALS or cancer, they could get their voice created for free with past data; we had been doing it for months but at a small scale. We partnered with the Scott-Morgan Foundation to make it more official and scalable. It's been a fantastic partnership. For us, that's working with the community to build things we're passionate about, giving back to the world and the ecosystem without focusing on revenues. Personally, this is the most rewarding thing we’ve done. A lot more things will come but, so far, this wins.
As you alluded to, open source models are increasingly making specific base model capabilities cheap or free. How do AI startups build pricing power and longer-term retention or lock-in ability?
There's the ecosystem element we touched on earlier - that's fundamental. Investors tend to focus on the negative side because they're more risk-averse. However, when you think about what's going to happen with open source models, how many companies will actually be able to deploy an open source model in their infrastructure in three years, even if that model is extremely high quality? Not that many. Deploying state-of-the-art models is expensive and requires a know-how that most companies don’t have.
Yes, you'll have more competition because there will be more applications. However, companies aren't keen on changing providers; it also depends on the vertical and segment you’re targeting. If you're building your ecosystem and delivering good ROI, when companies face the option of implementing their own open source model, most won't have the resources, capacity or will to do so. And out of the 90% who won’t have resources to deploy open source models, can another provider deliver everything that we’re already providing? Unlikely.
You end up with this network effect from years of upfront investment in building your community and products, ensuring the best value and ROI. Some customers may switch, but the vast majority will continue because it's more expensive to switch and start from scratch, even with another end-to-end provider. As long as we continue to deliver the expected value.
Look at examples like Facebook, Instagram, or TikTok. Instagram didn’t have a moat in their early years - anyone could build a social media platform. Look at Snapchat. Why are Instagram or Snapchat still there when hundreds of competitors popped up? Because they moved extremely quickly, owned the ecosystem, and showed value for users and companies building on top of them.
Salesforce is another great example. Companies spend hundreds of thousands a year with Salesforce and never move away. Salesforce knows this. There are products with better UI and better experiences that are newer, like Attio in the UK - really good companies. But they don't have the Salesforce level of integrations and depth. Hence, people continue using Salesforce - the same applies to Hubspot.
What advice would you give to AI founders? What do founders often overlook at the pre-seed and seed stages?
I always believe that even if you're going pure B2B, you should think like a prosumer company. Consider if you're going to open up your platform or products. This forces you to automate payments, simplify messaging, and service customers in a leaner way.
The other aspect is going extremely deep into end-to-end applications for specific use cases or verticals. If you want to win in the healthcare space, own healthcare. Forget about distractions across other areas. There's always the appetite to enter new markets, but own your vertical really well, nail it, win there. Once you're winning, then you can think about moving to another region.
Each region has different challenges you may not be ready for today. If you spread too quickly, too fast, you'll lack fundamental elements. Still, it's worth doing some "ice-breaking" tests - how could I win a few customers in Japan, North America, Europe, LatAm, Middle East, or Africa? I consider myself to be an “ice-breaker”; I like testing messaging, new segments, markets or ICPs without distracting the team. Only when I am certain a test was successful and it can be scaled, I bring the team.
Testing hypotheses constantly drives momentum. If you test hypotheses, discard what doesn't work, and know what works, you'll gain momentum to open new markets quickly or scale upstream faster.
Momentum is the most fundamental element - it defines pretty much everything.
Thank you for reading. If you liked this piece, please share it with your friends, colleagues, and anyone that wants to get smarter on AI, startups and strategy. You can find me on LinkedIn or Twitter.