From 0-50,000 Customers: How to Build a GenAI Company for the Enterprise with Synthesia's CEO

Please welcome to the stage Philippe Butteri, partner at asel, and Victor Riperbelli, co founder and CEO at Synthesia.

Hi everyone. It's my pleasure to be here with you today with Victor. I think we've been properly introduced. So I'm curious, like, how would you rate the quality of the avatar? Like, if the avatar didn't say it was an avatar, how many of you would have thought that it was an actor speaking a real human versus an avatar? All right, one person. Well, that's not. Well, it says a lot about the progress that AI still has. Still has to do. All right, that's a good start. So I'm Philippe Boteri, I'm a partner with Accel. So we're a global venture fund and we do a lot in software, B2B and AI. And we can do anything from million dollar seat to 100 million gold stage investment. And I have the pleasure to be here today with Victor, who is the founder of CEO of Synthesia. So you kind of got a preview of what Synthesia does. But to get started, I'd love to start with your entrepreneurship journey and how you went from skateboarding and punk music in Copenhagen to start one of the great generative AI company out of London.

Yeah, sure. And great to be here. So I think my journey towards founding Synthesia started in my childhood years. Always loved computers, loved sci fi, and kind of in my late teens, I guess figured out that I could turn it into a career on a personal level. I've always been more interested in like fringe ideas. And I think the Internet is a great place to find a lot of fringe ideas, interesting perspectives that wasn't around me physically in Copenhagen where I grew up. So I think that's just always been like a kind of a personal interest in mine. And maybe in my earlier years that was around like skateboards and punk music and weird music in general, which is still a hobby of mine today. But then in my kind of late teens, after figuring out that my interest in computers could become a career, I started building websites for local businesses, small things like that. I graduated into working in the Danish style startup ecosystem, which I did for four years or so. Mostly like product and growth roles. I'm kind of like technical, but I'm a terrible developer. So I always found the sweet spot for me was thinking about the product, but also kind of like how that connects with the customers and the business side and the kind of technology side. In those four years, I definitely learned that I wanted to build my own company. I Love building products, but I also learned that I just wasn't super passionate about building accounting tools or business process software. I wanted to kind of combine it with my interest in sci fi and kind of like the fringes of technology. So to make a very long story short, moved to London to kind of pursue that because Copenhagen didn't feel like the right place to build frontier technology. Spent a year on just kind of finding my feet, meeting lots of interesting people, did a lot of VR and AR and eventually met my co founder, Professor Matthias Niestner, or one of my co founders, Professor Matthias Niestner, who done a research paper called Face2Face, which was like, I guess the first research paper showing a neural network generating highly realistic video. And when I kind of saw that, I felt like I saw magic for the first time. And I had sort of two thoughts in my head. The first order effect of this is going to be that producing video content is going to be possible to do from behind your desk. No cameras, actors, studios, microphones and all that stuff. Just like you can make music from behind your desk, you can write or publish a book from behind your desk. Like most other types of media you can already do entirely in a digital workflow. And then the second order effect, which I think is what we're slowly getting to now, is that once that's possible and video is generated via code, it's going to change video as a format. New technology always breeds new type of media formats. And for video, what that probably is going to mean is that it's going to be less linear and it's going to be personalized, it's going to be interactive and intellectually, I think that's very exciting. And it's also a great business opportunity. So that's the short version of how I got to Fountain Synthesia.

And how do you go from. Well, AI is a fantastic tool to create video and we can get rid of all the hardware, all the camera to produce great original content, to focus more on avatar and business productivity.

So I think we did what conventional visual will tell you not to do, which we kind of fell in love more with the technology and a vision of where that, what's going to go in the future. So kind of once we decided to start a company, the question was, okay, great, we think this is going to be possible in 10 years time. How do we sequence a company along the way? Right. How do we build a product we can sell to people so that we can raise money and have customers and all that stuff? And the first thing we did was we sat down and we thought, okay, what do we think we can realistically build and sell within 12 months? Because that's kind of what the Runway determined that we had to. And we ended up on doing this AI dubbing idea, which was pretty simple. You give me a real video and we'll dob it into a different language by changing the lip. Synchronization. Synchronization. And of course, the voiceover. That has had a huge resurgence now because technology has matured. But back then it was two PhDs spending 10 days and making just like a 30 second clip, right? So we went to video production professionals agencies, we tried Hollywood with this technology and it didn't fail completely. Like us, there was a decent business taken into consideration. We were like eight or nine people at the time. But it was very obvious to us that we were selling a vitamin, not a painkiller. If we disappeared off the surface of the earth the day after, people would be like, ah, we'll just put subtitles on, right? And we also realized that because of that, we're not in control of the go to market at all, right? Like, we did a famous campaign with Beckham, for example, where we translated him into nine different languages. And obviously the bulk of the work around that is like convincing Beckham to do this campaign with this company over here and all the stuff in between. It's not just the AI part of it. So after doing that for a couple of years, we sort of sat down and said, what have we learned and do we think is the right path for the business? We didn't think it was the right path for the business because the path would be we become a visual effects studio with a proprietary technology. That's not what we set out to build. And then what did we learn? Well, we'd actually learned that the people we were trying to sell this to, their house wasn't on fire. They're making lots of awesome video content every single year, right? They are working with Beckham and Lionel Messi and Snoop Dogg and they make awesome videos. But what we learned was that there are actually millions, maybe billions of people in the world who are not making video today. And they're desperate to make video. Their house is actually on fire. And this ranged from just any kind of random person off the street who wanted to start a YouTube channel but didn't want to show their face to the world's biggest companies where someone really wanted to make video content but just couldn't get the budget because it wasn't prioritized and so the thesis was if we focus on all these people who are not making video today, we give them a much, much faster, much, much more affordable solution to make video. They'll probably be okay with the quality of the videos being slightly lower than what you get off a camera. And that led us down this path of developing the avatar technology, which we're famous for today. And we released that kind of three years into the company. And that was the sort of inflection point where we clearly really hit a nerve. Despite the products like almost barely working, being very low quality, there was just such a big pull for the market. People really wanted a better way of making videos without having to use cameras. And we looked at all that traffic, we figured out who's doing this because it's cool and fun and wow, I want to show my friend and who's doing this because they actually have a real business need for us. And that sort of took us down the path of building a B2B SaaS platform for video creation. And today we're the biggest one in the world doing this. We help our customers both create the content using our AI models and our video editor, collaborate on it and share it with the end user. And what we learned was that the business opportunity here, which I think is very common for startups, was more sort of a market expansion. It was like enabling all these people to make videos. So if you're a big fast food company and you need to train a lot of people to work in your restaurants, instead of doing it with a 40 page PDF handbook, you can do the videos, which is much more effective if you have 2,000 salespeople and you're zoom all around the world and you want to make sure they understand what's going on in the competitive landscape that they're in with pricing changes or whatever, then instead of sending them long emails that no one reads, you can make these bite sized videos which has much higher information retention, engagement. And so that was kind of the unlock, right? It was enabling people who used to create slides and text enable them to make video. And that was the big unlock. And I think what we did well in that journey was being very realistic and very intentional about what other technology in its current stage actually good enough for and not trying to get ahead of ourselves. And I think that's a lot of AI companies, I think that's a struggle people have today with where can it go in the future, what can it do today?

But did you have a sense when you started the company in 2017 of how fast the technology would progress in the coming year. Like, did you. Did you already have the vision saying, well, yes, the avatar from 2018, 2019, the first version, it's very gimmicky, it doesn't look like a human, but soon we're going to get there. Or do you say, well, there is a journey, but there is already a market for something that's not perfect.

There's a combination, I think. I mean, so we set back in 17. I think in the first investor pitch, investor deck we had, we said that in 10 years it's going to be possible to create a Hollywood film from your laptop, not needing anything else. That's 2017. And at current rate of progress, that wouldn't really surprise me. So I think we got the timeline, sort of like ballpark, right? But of course, in between, what's going to happen the next six months and next year is harder to predict. So I think it was more case of the latter in terms of, like, running the business of being like, yes, these avatars made to the average person on the street not look great. But what is it we're trying to replace here, right? What are we trying to better? And the realization we had, especially in the early years, was that actually the market for some DC videos is like the world's 90% worst videos. Right? Which are not even videos. It's like a PowerPoint being screen recorded or something. And I think that that's the kind of thing which was not obvious from the outside.

So to get the company started, you obviously had to raise funds. So I don't want to spend too much time on the fundraising. But there's one anecdote which I found very interesting. So can you tell us how you got Mark Cuban to invest as an angel in the company?

Yeah. So in 17, when we tried to race the first round, we went through, you know, we kind of. When we got the idea and we made the first pitch deck, we were like, this is amazing. It's gonna be so easy to raise money, right? We clearly have an insanely cool idea. We have a great team. People will love this. And of course, no one loved it. People thought we were batshit crazy, I think. So we got turned down by, like, 100 investors, something like that. There was not a hot space. The technology was very, very early. I'm not technical myself. I think back then, I think when. When. When VCs look for founders, a lot of VCs look for, like, the researcher to be the CEO, because it feels like they all then built the Tech and I think that can be valuable. But I think back then they looked at me, they're like this business guy is trying to create like a deep research lab to make AI video. Doesn't really, not really going to fly. So we almost, classic story. We almost gave up my co founder Stefan. We had this list of interesting business angels and Mark Cuban sat well in the kind of Venn diagram of media and technology. And Stefan sent him a cold email. He responded back within five minutes and then we had this 14 hour exchange with mobile email. He doesn't do calls. And 14 hours after maybe 3:30 emails going back and forth, he agreed to invest a million dollars at like 4am in the morning UK time, which was very, very surreal. But that's kind of what kickstarted the company.

Cold email can lead to a million dollar financing. I think that's such an amazing story. Show kind of the grit that you need to have as an entrepreneur and think that nothing is impossible.

And also I actually think what my biggest takeaway from that is, especially if you're doing something which is not in the sort of current Overton window of what people like to fund, you really need to find someone who already shares your thesis. Like it's very difficult to convince someone cold about your thesis. Mark Cuban had actually implemented this paper himself at home. He already believed that the same things that we believed. He was more looking for a team to fund it. Right. And I think that's what made the process a lot easier. Whereas a lot of the other conversations we had, we're almost starting at like what's AI and watch a video and you know, and then it's very, very difficult.

Nice. So shifting gear a bit. Let's talk about why an AI software company is different from a traditional SaaS company. So in a software company I would say there are two things that matters. You need to build a product and sell the product. In an AI company it's different because you need to build a product, but before doing building the product, you actually need to do the research. So it's not only an engineering team, but it's a research team that you need to have, which is a different budget, different type of people to manage. So I'm just curious, like how did you apprehend the setup of the company at the beginning to make sure that you could address this shift the best way possible?

So back when we started we were very heavily focused on like purely the AI. And the reason we're purely focused on the AI was because if we couldn't make that work. We probably couldn't really make, make anything else work. That was like, that was very much the core of the company. So in the beginning it was basically me and Stefan on the kind of business and product side and the rest of the team which was, I think eight other people, were all researchers. Once we kind of built the first product we hit the inflection point actually only then we hired like a team to build like the whole SaaS platform and what traditionally was kind of start off with. I'm not sure I would do that again. If I, if we had raised more money back then I probably would have started the product engineering side earlier. But what is very unique to, to, to building an AI business and I think you have to be very intentional about. Do you actually want to do the science? Especially in the market where now with these technologies are much more mature and much more accessible, do you actually want to do the science? If you want to do science, you should do so because it unlocks a fundamental new market would be my take. The issue with research is that there's so much uncertainty, right? It's so difficult to put timelines on. You're kind of per definition trying to do something which is really hard and may not succeed. So you have this division in your company where the DNA of a researcher is so different from the DNA of like your traditional SaaS hacker, early stage engineer just like wants to get stuff done. Researchers are scientists, right? And they have to be scientists because otherwise they can't solve, they can't solve the problems. But it can be quite frustrating. I think a lot of AI companies are still trying to figure out what's the right interface between those two teams. How do you run the research team. And I think what we have found to work well and what other founders that have built big AI companies from my experience have found works well is that you almost need to take a VC style approach to your research. There is some problem you're trying to solve and there is many different ways people try to solve that problem, right? That could be like different architectures, could be different data sets, could be a whole bunch of different things. And because it's research, you don't know which one's going to work, right? There's a lot of people who try to do things similar to LLMs that just didn't work. But LLMs actually did work. And so the way we've approached it is almost like a spread betting portfolio where you have a few teams working to solve the same problem, but they approach it from different angles. And that works well because you get the kind of wide coverage to kind of hedge against you not picking the right direction, which is a very big chance of you not doing. But you have to be very good at then killing your darlings, right? So when you see one direction take off, shut down the other projects, pull them into that project, and that's just very different than how you would do, like a traditional SaaS app, right? Like, you wouldn't have three different teams trying to build a billing system, and then the best team gets to keep the billing system because you're dealing with much more uncertainty in these kind of teams.

And what does this mean in terms of culture and performance for a research team? Like, what are. I mean, you said something which is very interesting, which is, you know, people are searching, you don't know if they're going to find, right? So it's very hard to say, well, you need this discovered, you know, in six months, and they're like, oh, you haven't found to mean you're good, you're bad. Like, so how do you, how do you evaluate the team?

That is really hard. It's really hard. I think you, you, you can still look at things like Velocity, for example, right? You can look at are they smart about how they do things. And that can be on an IC level, but I think also very much on a manager level, you need to make sure that, like, a lot of researchers will naturally be drawn to the hardest problem. And the hardest problem is not always the things that's the most valuable for your business. Right. Let's say with Synthesia, we're building avatars, and the research team has built something that kind of works, but they figured out that for people with a big red beard, sunglasses on, and a pirate hat, the technology doesn't work. A lot of researchers will be. That's very interesting. Why does it not work with a pirate hat? But it works with a normal hat, Right. And they might get drawn into that and be like, oh, let's try and solve this problem and how do we do this? But obviously, for your business, it doesn't matter because you could just tell people, look, you can't make an avatar with a big red beard and a pirate hat and sunglasses on. The technology doesn't work. Right. That's such an edge case. So I think measuring, like, how well are they at thinking about the product in the choices they make in their research is one vector and the other one is, I would say, is velocity. Right. Even though you're doing science how many experiments are you running per week or per month? Right. How good are you evaluating those experiments? Because otherwise you run the risk of running like a high school science lab with no oversight. But, but it is, it is very difficult. I don't think anyone has like the perfect answer to this yet.

And how do you think about strategically about where to focus the research? I mean, there are a lot of things that are being developed in AI. There are a lot of things that could be off the shelf that you can use and there are things that, where you really want to push the boundaries of the science. So, and that's obviously costly. So how do you think about this, this balance? And I think what advice would you give to the entrepreneur in the room who are kind of dealing with the same question?

So for many years for us there was no choice but to build it ourselves, right? There simply just wasn't anything else on the market. And the open source community was, was, was working on like very different things than what we needed. I think today you want to be very intentional about what narrow slice of AI you can be uniquely the best in the world at. And that's a pretty high bar given how much investment and how many great companies already trying to do things right. You're probably not going to be the best in the world at building a fully generalizable LLM model today, unless you're like anthropic or OpenAI of Google or whatever. For us, we think, well, we're not going to be the best in the world at general AI video like Sora style from OpenAI, but we think we can be the best in the world at are people presenting to the camera. And that is a huge category. And so if we focus all of our efforts on just being the best in the world at creating videos of humans talking to the camera, I think we can win that. And the rest we can partner with. If our customers want to create a video of a dog running on the beach with a plane flying in the background, let's integrate that from someone else. So for us that means we build device technology, we build the video technology, build what we call our director module, and all these different components of the system are tuned for that specific use case. And I think if you have three different systems that are all tuned towards a specific use case, that does build a lot more competitive differentiation. But I think in the world we're in right now, right, you want to be very careful about making that slice as thin as possible.

And how do you think this is? Where do you think this is going over time. Do you think that over time all the OpenAI and the hyperscalers and entropic, etc. Will develop enough technology so that you don't really need research anymore and you're just going to use things off the shelf, or do you think that there's always going to be some area where you can continue to do something specific that's better and giving you a competitive edge on the market?

I believe the latter. More. I think that the world is messy and complex and when you build systems that work in production, right, there's. There's a million different things that they need to be able to do. So I do think that there is going to be value in research for a long time. I think you want to be also intentional about. Do you want to be an infrastructure layer where you truly have to be the best in the world at something, or do you see yourself more operating in the applied layer? And I think if you operate in the applied layer, there's sort of one extreme version of the world where you say all the research we're doing, we're doing because it's a go to market advantage, because we can take our products to the world first, but maybe over time it matters less if you can replace it with an API call from OpenAI or something like that, right? But you had the advantage of taking that product to market 6, 9, 12 months before your competitors, and that in itself is hugely valuable and it's probably worth spending millions of dollars in research on, right? But you're building your product in a way where the sole value of what you're building is not just the AI model. And to give an explanation for this, like if you take ChatGPT, right, there's like, ChatGPT is like an interface to a model and obviously OpenAI needs to be the best in the world at that specific thing. But if you're building a customer support chatbot that needs to be able to respond to customers inquiries, that piece of technology is very important. But you probably still have some kind of an agent flow. You have a knowledge base, you have some tech that helps customer support team make sure that model doesn't lose. There's so much scaffolding that sits around that, and that is actually more your product than the AI model themselves, right? And I think if that situation you're in, I think it's actually pretty good, right? Maybe someone comes out with a better LLM in three years and you can just replace it. But for your customers, they're not just buying the interface to an AI model. Right. They're buying the full solution, the full platform that sits around it. And I think that's when you look at a lot of the AI companies, people have been built last 12 to 18 months. That's where you really run the risk of being steamrolled by an upgrade to GPT technology. Right, because you build like some little thin layer on top of GPT4 and it's very cool at the time, but it eventually gets eaten up. Right. So I think the traditional rules of building a business doesn't really change. Like you want to build, you want to build a platform that solves a problem for your customer, doesn't just give them a piece of technology, unless you're an infrastructure company, in which case you are selling just the technology. Right?

Yeah, no, that, that makes sense. So to develop AI, you need research team, you need algorithm, and then you need data. To what extent do you think that data can be a competitive mode in today's world? And what did you do to go after sources of data that you know are more unique or proprietary?

So data is definitely going to be very important and it's probably going to be more important than the model layer, in my opinion, over time. We definitely seeing this play out right now where people train these like massive models. They're very powerful, they're very capable, and now people are trying to figure out how to make them really good at specific tasks. And most of that work that's going into that is saying, okay, you need to curate the data that you train, maybe even the base model on, and you probably want to fine tune it on data that's very specific to whatever task that you're trying to accomplish. And it is just though that like the world's publicly available data and information which these models are trained on, there's a lot of stuff that's not in there, right? Like there's probably not like a million MRI scans, there's not like a bunch of confidential spreadsheets from like a really big business. So that, that's all the data that people are now figuring out, okay, these, this data is very valuable. It's not available on the public Internet. And obviously if the model hasn't seen the data, it's not going to be able to, to learn whatever it needs to learn from that data. So I think that's going to be very, very important moving, moving forward. And that's, I think there's two parts of this. There is like existing data that you can take and prepare for training and then there is like acquiring new data, which is also something that's growing. So for us, we're quite focused on acquiring new data. We have a big studio here in London. We captured thousands of people following a specific set of instructions, which serves as the kind of the base for some of the foundation models that we're training right now. So I think that is going to be very important. Of course it matters if you're like a medicine, like a pharmaceutical company where probably there's a lot of private data, you could argue in the customer support kind of world, maybe it's less important because everyone's knowledge base is already public. So that's easily like scrapable you can put into the model. But, but it definitely seems like curated data is, is how we're going to see the most progress in, in all modalities for the next 12 months.

Yeah, that's. So that's what it takes to build, you know, kind of your core AI engine and then you need to sell it, you know. So it's been very interesting to see that Synthesia managed to develop a large footprint with the Fortune 2000. So to some extent it was one of the early AI products that started to get broad adoption. So how do you make that happen?

So obviously distribution is as important, if almost not more important than actually building the product. And I think it feels like I've been in tech now for more than 10 years, right? And there's always like one sort of channel that's like significantly better than all the others that changes all the time. That's kind of a window that moves, right? So when I started out in tech, SEO, for example, was huge and there was many, many billion dollar companies that was built with SEO as like just the primary way of driving traffic. SEO is still great, but it's very competitive. People search less and so on. In AI, it's very clear that that channel is word of mouth. AI has this sort of magical element to it, right? It has a huge wow factor. People feel a bit threatened by it, so they want to try it and understand how to use it. They want to look cool on LinkedIn by posting the latest and greatest AI technology. And so this is one of the things we've seen the last 12 to 18 months, right? And so when you see something like that, I think it's about, okay, how do you utilize that for your business? And I think, I think we've been pretty good at that. You want to make your product super accessible, you want to make it fun, you want to make it shareable. And if you're good at those things, you can actually build this sort of crazy word of mouth groundswell. And I think most of the AI companies that are very successful today have used an element of this because that, that distribution channel has been incredibly Strong the last 12 to 18 months.

How do you make the two work? Because usually when you have this bottom up engine, it's mostly suited for kind of prosumers, small businesses. But how do you get from that kind of individual individually driven traffic to get to a Fortune 2000 accounts? Right, yeah.

So obviously every company is different I think for us. But what we realized early on was that all these people who work in the enterprise and who have the problem of wanting to make videos, which is rarely the CEO, right. That's often someone in the organization who has a job to do and they wish they could make more video. All these people are also on TikTok and Instagram and whatever social channels that we're on. So actually if we focused on those and drove just a very, very kind of big top of the funnel, then there would be a lot of people in that top funnel who will not be relevant to us because we are looking for that like one person in a Fortune 2000, but that person would also be in there. And then it's our job to get them into the funnel, get them to the website, help them make a video. We of course identify this is someone who's like interesting and then basically building more kind of a filtering machine of take this top funnel, figure out which ones are the most interesting people who came to us and made a video as opposed to the other way around. Right. Of like cold outbounding would probably be the opposite of this where you kind of filter at the top. You only talk to the people that are super directly relevant to your business and you do everything you can to find those. So we kind of flip that model around. Let's just get it into as many people's hands as possible and then from that we'll kind of sift out which ones are the interesting ones.

So that worked very well at the beginning. I mean this bottom of motion I think is what kind of made the company very well known because I think one of the benefit of all the prosumer behavior that you get to be on social media and that kind of drive also awareness which then kind of feed the enterprise business. But then at some point you need to also do the outbound. Right. So can you walk us through this journey from inbound to outbound and how it has worked so far.

So for us we went through like sort of different phase, right? We hit this kind of first moment where it just took off. It was just this force of nature, people just coming to us using the product, they wanted to pay for it. It was very, very exciting for us, of course. But we're also very aware that in that batch of people, right, there's a lot of what we call AI tourists, which are people who, you know, are very willing, which thinks very unique to AI is that people are actually very willing to pay to try something out, both in enterprise and personally. Much more so than the kind of previous generations of SaaS. And that's great, that's revenue, it's dollars in your bank account. But it doesn't necessarily mean that these people have a long term use case, right? They might just be trying out. So like first thing for us was like, okay, let's filter into who actually have a use case for this. Then we started on the product side. Let's build for them. Let's make sure that when people come to the platform, this is what we tell them. They should do these things. We do the right templates for them, we talk to them, we kind of build a business around that. I think that's all the, the second phase, right? We got all this viral moment figured out who's actually the interesting people in all this sort of viral stuff. And then the more we learned about who are these people that are really using the product and love the products, the more it also made sense for us to say, okay, let's begin to go a bit more outbound. Let's do some paid campaigns where we try and target our specific icp, right? And then you kind of go deeper and deeper down. Up until today where we're doing like actual cold outbounding because we know who the right user for the product is, we know what value they get out of it. It means that we actually can do this, right? But that would also have been more or less impossible to do in the beginning because we had to learn what they were doing with the product and what was valuable. I think that's the sort of interesting thing with these products where they're very open ended for us. I mean, you can use video for a million different things, right? You can use it in sales or marketing or training, or send a birthday message to your friend. So very general technology is a bit the same with LLMs, right? Like they can do a million different things. They can teach you about history, they can help you write an email. And so I Think and I think that is a bit unique to AI right in that these things are so open ended that you need to sort of curb that before you can build that outbound engine.

So as you move from, you know, one motion to the next, I mean, what was there a particular aha moment that you had that kind of drove kind of the next wave of growth?

I would say the two big aha moments was the first one was like, people are not replacing video production, they're replacing text. Was very fundamental to our understanding of like why this product is valuable. And the second aha moment was when people, if everyone in this room kind of close your eyes and think of video, probably the videos you think of is like the Netflix show you watched yesterday, a really well produced YouTube video, or like all the great content that floats to the top of YouTube or TikTok wherever we consume content, right. And so if that's what's in your mind and you look at a Synthesia video, you might say, well, that's not a great video. Right? And we had to overcome that bias by saying that's true. But that's not the videos we're targeting. There's most like 99.9% of videos in the world are not beautifully made YouTube videos. They're like some really crappy learning video in a big American bank that everybody hates watching. Right? And that's our opportunity. It's making that video slightly less worse. Even if the final result is not like, wow, this video is amazing. And that took a lot of kind of like adjustment in the team to figure out like people love replacing video, they're replacing kind of text. And when we then go into the enterprise, the sort of videos that people are trying to create with us, other tools are already crappy. And it's our opportunity to just make them better, which has real impact for them. But that was so non obvious from the outside. I think for many years it was kind of our little secret that this was actually such a good business because a lot of people would look at it and be like, oh, it's kind of a fun, cool idea, but it's a toy. No one is actively using this for anything.

That makes sense. So we talk about inbound, outbound, outbound, bottom up motion. What about channels? What do you think are the right channel for any I business and your business and Synthesia in particular? And when do you think is the right time in the company journey to kind of start looking at this as a new vector for growth?

I think as soon as you Feel like you have like actual product market fit, you need to start thinking about how do you build a scalable distribution engine, right? And the reason you want to do that pretty early on is you don't want to be in a situation where you hit that moment, right, where people just coming to you. It's crazy. Your website is getting flooded. There's so many people that want to talk to you about using your product and it's just amazing. It's one big party, right? But that's gonna end at some point. Maybe a month, it may be six months, maybe a year, it may be two years. But at some point, most likely that moment is gonna end, right? And so what do you do when that moment's gonna end? You don't want to be left with no inbound for your sales team to pick up. So you want to start thinking about what is a distribution strategy for a company before that, right? Because then you're in control of your own destiny. For AI, I think in many ways, like, I don't think the rules change that much from like other companies, if you're building something that people are searching for, you probably should start doing SEO. If you're building a new category, you should probably also begin your SEO even if it'd be less or even be less impactful in the short term. But like what we found for example is like owning AI video was less valuable three years ago. It's very valuable today, right? So I think you look at all the different channels that we all aware of already. I think the big thing is this word of mouth which is, I think is kind of very unique to AI, right? There's so much public interest in these technologies and it's your job to really kind of harness that. It is kind of a double edged sword because you, it can also be easy to kind of be pulled in too much by the hype and end up focusing on like short term experimental revenue versus short term valuable revenue. Right. Short term revenue is a lot of like consumer driven self service. People come in, they have big ideas of what they can do with the tech. Like people come to us and be like, I'm going to make ads for Facebook and we can already tell them like, look, it's not good enough yet. Like the quality of the advertiser is just not there. It will be there at some point, but I know you're going to churn, right? But you get a thousand people in there and then you get like some enterprises who are interested in like for us, training content, for example, that Tire kicking a little bit. What should you focus on? I think if you focus too much on the short term thing, you might be left in the dark when those people turn because they didn't find a use case. So balancing the kind of short term pumped revenue versus the long term enterprise revenue is very important. And the long term revenue is just, it's more painful. Right. It's like much longer sales cycles. You have to talk to the customers. But I think it's very important to invest in those and will not get carried away by the hype.

And what about channel partners? Like you're doing a lot of, I mean one of the use cases, internal training. Are you working with internal training company, educating them on the technology so that they can distribute it to their customers or use it produce content? How are thinking about alternative channels?

I think partnerships for us has, I think, I mean I think every business is very different. Right. But I think partnerships in general are quite, for us at least been quite difficult in the sense that outsourcing your sales and your customer success to someone else. And for us the most obvious thing that a partner could do is like help our customers make the videos. It's like a new discipline that they need to learn before they can sell it. Right. And they need to have a good incentive to partner with you. And that's not like making $5,000 off a referral on a 50,000 enterprise deal. For us, what has worked well is actually going much higher up the partner stack. So we work with a lot of the world's big consultancy agencies like EY and Accenture for example. We've gone in with them and really spend a lot of time on building practices around Synthesia. So how can we help their customers communicate better and have a good value proposition where we of course have the technology and our partners have all the know how in terms of how do you communicate to a 10,000 person strong organization, right, who's gonna make and maintain these 400 videos as part of like your new training library? So that has like worked pretty well. But, but I wouldn't say that partnership for us has been like a kind of a channel. We've invested like that much in so far because it's generally just been hard to in such a new category find partners who can take that to the market.

Yeah, no, that, that makes sense, Claudia. We need a bit more maturation before you get there. So we're reaching the tail end of the discussion. So let's, let's look into the future where you think what, what's in store for Synthesia. Where do you think the technology can go? I mean, do you think that next year if we sit here, like half of the room will raise their hand when we ask if it is a human or an avatar?

Yeah, for sure. I think so. We've always looked at these. One of the kind of key things we've always had these rounds in these years is that ultimately our TAM is rate limited by the quality and the capabilities of the avatars. Right. And what we've been good at along the way is figuring out the right market for that particular point in time of the technology. Right now we're training our foundation models. In the next probably three to four, five months, we'll see that hit the market. And that's going to be like a 10, 10 x 15 x better product than what we have today. I'm really excited to see what people are going to start creating with that. We've already seen just the last six months as the avatar has gotten better and better. Huge increase in the types of use cases people do, much more external facing content marketing, sales for example. But I think it's pretty obvious that if you with AI can generate something that looks and feels like an Apple Keynote in just by just typing it out, that's going to be have a very big impact, right? A bigger impact than the sort of stuff we're seeing today.

And what about the latency? Because with better quality means more data, more processing, is the latency going to go up or do you think that over time we'll be able to have a zoom dialogue with an avatar who is basically interacting in real time?

Definitely. I think it's only going to go down. And I think we're also getting to the chapter of Synthesia now where we really begin to say, okay, up until now we've built this great platform for making videos. But ultimately what we're doing today is we're helping you make MP4 files by taking the camera out of the equation. And that's very powerful. Right? But where this gets very interesting is once video evolves into something that's different. Today video is a broadcasting medium. It's one to many, right? You upload your video on YouTube and that video, the same video is watched by 10 million people or whatever, how many people watch it. And in a few months we're going to be making some big announcements around trying to evolve video into a type of media format that's much more interactive, it's personalized, it's eventually going to be something you can conversate with kind of like you would have a zoom call. And I think that's actually where most of the potential lies over the next five years. And I think this is almost going to evolve into becoming a new type of interface. That's of course very different from text, but also very different from video as we know it today.

You think five years from now we could do an audio zoom that looks like a video?

Yeah, definitely.

That would be quite interesting. It'll definitely be interesting and scary at the same time.

Good.

Well, it's about time to wrap up. And the last question I would like to ask you is what are the three key pieces of advice that you would give to the AI founders and entrepreneurs in the room? But just before you answer, let's have Alex come back on video and see if she can do a better job than the first time. I think that's the cue to project the video.

That was a great session. Here are some of my key takeaways. But it would be great to hear from the humans on stage too. Anything I missed, gentlemen?

I think that was great. I mean, when you see a short message like that, come on. If anybody says, no, no, I know it was an avatar, I think you're not being fair. But Victor, do you want to elaborate more on these three key points?

Yeah, I mean, I think we kind of covered them. Right. But I think the first one is like focus on driving utility over novelty, as we say internally in synthesia. Short term revenue is great and it can be very positive for your business, but make sure you don't lose sight of what's actually valuable for your customers long term. I think a lot of AI companies have been struggling with that the last 12 to 18 months. The research versus implement versus buy off the shelf touches on just being very intentional about when do you want to do the science yourself? Because you need to do the science yourself. When are you going to implement something that may be available open source or it's kind of a solved problem so you can put more of a timeline around it. And when do you want to buy off the shelf from another provider? Right. Getting the balance between those things right is very important and especially with the way that the landscape looks today, where now there is actually a ton of AI companies doing all sorts of different things. It's very different from 12 months ago and it's certainly very different from three years ago when there was like very little infrastructure available. Right. So sit down also and think about what, what do you think is going to happen in the market, which kind of technologies is going to have specific players just doing that, that you may be able to tap into not today, but in three months time. And then the last one is just like every new market, I think has these growth hacks. And the growth hack right now is making people talk about your product, which means make it accessible, make it fun, make it shareable. And if you nail those things, then you can build this kind of force of nature which is. Is just incredibly powerful. I think the biggest growth tactic right now in AI, right, is just announce a bunch of stuff that may not actually be ready or work in the products. I think we'll see how that pans out over time. That's the sort of extreme version of this. I wouldn't necessarily advocate for that, but I think it's an interesting data point. Right. Like almost every company right now just announcing things that don't actually work because that builds that groundswell around, you know, that particular company, which if you use it right, can be very valuable.

Well, you did a nice hack with GPT4.0 real time dialogue.

We did actually.

That was, that was fun. One great. Thank you very much, Victor.

Thanks for having me.

It's a fun discussion.

Thank you.

Thank you everyone.