Show Notes

The machine learning industry continues to grow by leaps and bounds, creating a whole new space in the technology sphere. That’s where Tigran Petrosyan and his startup SuperAnnotate enter the picture. Tigran is the co-founder and CEO of SuperAnnotate, which provides the training data infrastructure that’s needed for machine learning to work.

Tigran recently joined startup coach Roland Siebelink on the Midstage Startup Momentum Podcast to talk about this new space and SuperAnnotate’s role in it. They discussed SuperAnnotate’s unique origins as well as Tigran’s key learnings along the way:

• What it was like for Tigran to found a company with his brother.
• The way that co-founder roles and responsibilities have changed in the first four years of SuperAnnotate.
• How Tigran decides whether he should take someone’s advice or not.
• How to set up a go-to-market that can sell to a variety of industries.
• How SuperAnnotate went about attacking a space that already had big players.
• Why SuperAnnotate abandoned its self-service platform despite early traction.

Transcript

Roland Siebelink:Hello and welcome to the Midstage Startup Momentum Podcast. My name is Roland Siebelink. I'm the co-founder and CEO of the Midstage Institute, and we help fast-growing startups get even better at maintaining full momentum. One of those fast-growing startups is with us in the studio, as every week, but I'm particularly honored this week to have with us Tigran Petrosyan, the co-founder and CEO of SuperAnnotate. Hello Tigran. Welcome.

Tigran Petrosyan:Hey, Roland, excited to be here with you.

Roland Siebelink:That's excellent. You're dialing in from Miami today, which is fast gaining a reputation of the best new startup hub in the Americas. How do you like it there so far?

Tigran Petrosyan:It's amazing. The weather is amazing. The atmosphere is a little bit different than the San Francisco Bay Area that I was before. The interesting part here is also you see a lot of more relaxed people, relatively relaxed people than in the San Francisco Bay Area. It gives you a feeling that there is another life in the world.

Roland Siebelink:Very good. Living in San Francisco for about 12 years now I think it has a lot of advantages, but the percentage of relaxed people is definitely not something to write home about here. That is absolutely true. Amazing observation. Let's move to SuperAnnotate the company. As always, the first question is what do you guys do? Whom do you serve? And what difference do you make in the world?

Tigran Petrosyan:SuperAnnotate, we are in the space of machine learning. I'm trying to put it in a simple way that maybe it's easier to understand because it's such a tech-heavy world we're operating in. If you think about machine learning, it looks magic like autonomous vehicles identify subjects or Siri recognizes my voice or any other application you can think about like face recognition. In order for that to work, usually those applications need to collect and label tons of data. Images or audio, for example, where is the car, where's the human, where's the tree in the street - just as an example for autonomous vehicles. And data is being inputted into the machine learning pipeline algorithms to build models. And these models ultimately recognize those objects or send payments or being used in speech or texts. What we do is we cover the part which is very hardly talked about much. Everything is magic, things work out somehow. But of course, behind that magic are all that annotations, training data infrastructure that is built in the number of millions and hundreds of millions in certain applications in order to make it work. And we're providing that training data infrastructure, and people who actually build that training data and manage in order for machine learning to work, whether it's for images, videos, textual files, audio, et cetera.

Roland Siebelink:Excellent. It's really like the people who got rich in the gold rush were not the people digging for gold but the people who sold the shovels and the tools.

Tigran Petrosyan:That's true. It's a very good parallel, yes.

Roland Siebelink:Okay. You're helping all those machine learning companies get the raw material of their data and the annotations and the labeling in place so that the models can actually start learning. Can you tell us a little bit how you got into this space? What is the founding history behind SuperAnnotate?

Tigran Petrosyan:Absolutely. About three and a half years ago, I was doing my PhD in biomedical imaging side of physics, in Switzerland. And my brother was doing his PhD in machine learning in Sweden. I could see the applications of computer vision and how it can help radiologists, for example, to make better diagnosis from medical images. And I could see countless applications; it was still quite early in that field but this was fascinating to me at that time. But the main drive came from my brother's technology that he was working on during his PhD, which was something called super pixel segmentation on images. And he applied that to image labeling and realized that we were much more accurate and faster than any other technology that's tasked to do labeling. Since you're doing image labeling for millions of images, the speed and accuracy become very critical because the costs are crazy. Somehow, we realized this opportunity and seeing some traction and some competitors wanting to buy the tech at the time - just the PhD work about that time - we realized what if we can do it ourselves, me and my brother. We dropped out of our PhD programs and decided to start a company. In fact, we took the company to one startup event in our home country, which is in Armenia. We got in and we won that whole competition that basically triggers even further to actually do it for real. That's the starting story, how we got started. And then some early funding that helped us to grow some team in our home country, Armenia, then US and Europe and Bangladesh and other parts of the world.

Roland Siebelink:I immediately want to go to the co-founder question. What it's like to found a company with your brother.

Tigran Petrosyan:Yeah, that's a very good question. There's definitely something to say about it. We were super close from young ages with my brother. I was slightly older, maybe one and a half years older than him. My dad did his best to make sure that me and my brother are studying the same year at school. We were sitting next to each other in the same classroom for several years, doing the same homework, being basically best friends all the time, doing things together. And before we went in different directions - I went to study in Switzerland, my brother went to the US and back to Sweden - but we were very close. And then when an opportunity came to get together again, we got very, very excited to do that. Of course, once you start a company, there are so many challenges you have to face and discuss, and it's not easy. With time and the right effort, I think we have found the best way to first split the work and then at the same time to make sure that we complement each other. For example, my brother is a much more visionary technology guy, and I'm more trying to push things into some reality and trying to understand what will work and what not because sometimes you need that creative genius behind things that seem impossible at first. But eventually, once you learn and dive more, you can understand that it actually makes sense. It's a very good combination for us.

Roland Siebelink:Yeah. It's what some companies call the rocket fuel behind the founding team. The one visionary and then the other one being the integrator or how it all fits together and how you can actually make it understandable to a team.

Tigran Petrosyan:That's true. It's interesting that our role - we are almost four years old - how much our roles have changed, how much our responsibilities have changed during this time, and what we're doing. But of course the core of what we do and what we focus on stays the same. He's more on the visionary side. I'm more on the business.

Roland Siebelink:Okay, very good. But since I know many of the listeners will want to understand a little bit more on this division of roles between co-founders - not always brothers or family members, of course - but as you say, it's changed over the years. How did you have those discussions? How did you decide on changes happening? Was it always driven just by clear needs from the outside world or more frustrations or just energy, like let's do something else. Tell us a little bit about that.

Tigran Petrosyan:Yeah, it's interesting. He was always the product product guy. Because he built this technology first during his PhD, he knew about everything about this space, about different platforms, what are the advantages, disadvantages, what we could do better, what are our strengths, weaknesses? Of course, as I was learning that, I was focused more on the business side. We clarified the roles early on, which made things a little bit easier. One thing that also very much helped us when we brought in our first business co-founder later on from the US - it's not only me and my brother, but also we have someone else in the equation. Then any questions if there is any sensitivity of the conversations or any biases that can come from our childhood as being brothers, we know that there are folks that can come into the conversation and we need to listen to these folks because they are more experienced than us. We brought in more investors, of course. We got very close to our investors in a way that they also contributed a lot into the conversations and decision-making.

Roland Siebelink:How do you know when to take someone's advice or to disregard it?

Tigran Petrosyan:Good question. I can get different perspectives from different people on the same question. The best thing I'm looking at is how much this person is better or experienced on the topic they're talking about than me. And second, when those folks talk, how much extra knowledge or extra perspective they may have that I don't have. And really putting together from different perspectives and really understanding what would feel best for me, for the company, for that point of decision-making and ultimately coming to a certain decision. In the beginning, it can be very frustrating because you really want to make sure you're doing the best possible decision all the time. But one big learning I got is that you're just realizing what I did was the best decision I could make based on the information I got at that time. Even if I would feel stupid several months later looking back, which I do very often, that basically means that I gained some more wisdom, more knowledge, more understanding that helps me to make better decisions in that specific area. That's a very interesting dynamic I'm keeping in mind.

Roland Siebelink:Exactly. And you learn every day and there's a lot of reinventing that needs to take place as part of this startup journey.

Tigran Petrosyan:Exactly. There isn't a road somewhere where it’s written what exactly it should look like because it's a new space. The way it works is different. It's all about doing and iterating, setting something and iterating. Ultimately, when we realize after several months or years how much value this became, it's not just technology but all of those processes, all those things that you've learned and iterated, it's so much value in there. Even if someone I would tell all the details, I would dare to say that it will be crazy hard to replicate because all the learnings are very different than when you're seeing it rather than listening and trying to replicate. It's incredible.

Roland Siebelink:That's what they say in the beginning, inexperienced founders are always worried that people will copy their ideas. And in reality, it's just the kind of energy and the kind of persistence you need to have, plus all the learnings it's just by nature not replicable. I think we all learn that over time. Let's move a little bit back to the business. Can you talk a little bit about the traction you've been getting with your product? Whatever number you want to share with us that gives us an impression of how big you are becoming and how fast you're growing.

Tigran Petrosyan:We're really excited that since launching the product in early 2020, we have been working with over 200 businesses that are using our product, not only for their internal labeling purposes and setting up their training data infrastructure - training data is another magic word for annotation. Then there are other groups of customers that not only use the product but also teams that we have integrated into our system to label their data as well.

Roland Siebelink:Okay. Who would you say are your core customers, your ideal customer profile? The people that you know upfront if they have this and this characteristic, they're going to be very satisfied with what you offer?

Tigran Petrosyan:We have been working across many different industries. As you know, machine learning is integrating in every single industry now. For example, autonomous vehicles are a huge use case that drove our industry. We have had customers there, but I wouldn't say this is the main driver of the industries we're working on. A lot of good applications on aerial imagery analytics, medical imaging, manufacturing automation, retail automation space like Amazon Go type of companies, we're building such systems. Overall, it's pretty broad. But I'm happy to say that these are some industries where we found a lot of good traction. And one thing to mention - it's interesting to see the dynamics in this space. Basically, what we're building is for data scientists, for building models and applications. And these teams can be groups in large enterprises or they can be a specific startup at different stages from seed, Series A, series B, et cetera, they're growing. And even one startup building one specific application on machine learning can become a billion-dollar company or more. This is where this infrastructure becomes really critical for them. It comes across the board from companies who are just raising money to groups in enterprises when it comes to building such machine learning applications.

Roland Siebelink:How did you start setting up a go-to market organization to target all those industries and regions and profiles? How did you get your sales going and how did you scale it?

Tigran Petrosyan:Yeah, good question. When we were very early, there were already some known companies in this space. We really needed to put our name into the market. Now it sounds crazy, but when we were just starting, there was a huge computer vision conference and we spent more than 30% of our money at the time to just showcase ourselves at the conference to show that we exist and we're big and we're really cool, come and work for us. This was one way we set our foot into this market and space and get some visibility.

Roland Siebelink:Did it work for you?

Tigran Petrosyan:I believe so. A lot of folks that we're not thinking of us seriously that we met at that time, the conversation started to get more serious, whether it's for advisory opportunities or customer related opportunities.

Roland Siebelink:And is it very much a hard sale with a long sales cycle in your experience? Or have you got such a clear no-brainer product that people just buy it right away?

Tigran Petrosyan:It's a very good question. Actually, about two years back, we were like, "Shit, we're going to make a self-service platform. People come in and put their credit card and then that's it." Actually, that brought in some traction early on, but then we realized that as the platform gets more and more complicated, it gets very difficult for the people to understand all the functionality, all the value. And we really need to engage with them to make sure that all the setups are right. And then there's another big part where they need to integrate their pipelines to our data pipelines, and this needs quite extensive coaching and support. Of course, if they need services people who do the labeling, we offer data support as well. We abandoned that self service. And with that, the sales cycle gets longer, maybe a couple of months or a little bit more. With that, you also realize that you're giving the maximum value to the client. And of course, the average deal size increased and the whole partner engagement is getting a different game. It's a fully different go-to-market business model.

Roland Siebelink:You mentioned already that your service has been used by over 200 businesses in founding adjuster two years ago. That's amazing traction. How big do you see a SuperAnnotate become in the long run?

Tigran Petrosyan:Good question. At this stage, it's so hard to think about where to go, whether it's an IPO or an acquisition. We're just thinking about bringing more value to our clients and spending more and more to bring value. We've already seen a company in our space becoming a several-billion-dollar company. There are already some large unicorns emerging in this space. Hopefully, we're getting there. This is certainly something that we imagine going on. The biggest value - what I'm seeing - is to make sure that I'm building the team that is very excited to work with together with us and bring that value to the company, to the client, to the infrastructure - clients are happy, investors are happy - that drives us much more than imagining what's the valuation going to be in a year or two or three.

Roland Siebelink:Okay. I like it. European pragmatism compared to the Silicon Valley mindset. If people want to hear more about SuperAnnotate or even work with you guys as a client or an employee, where should they go? And is there something specific that they should look to download or to investigate on your website?

Tigran Petrosyan:Yeah. If you're just getting started as an enthusiast, if you need to build some training data infrastructure, we have a free desktop app you can download directly from our website. If you're a little bit more mature, you have done some work internally and built your early models and you want to scale, the best way to learn more is come to our website, superannotate.com. Set a meeting with us and then our team will immediately set up a meeting with you, learn more about you, and share more about how we uniquely can be positioned to help solve your problem. And then we can take it from there.

Roland Siebelink:Excellent. Very good. Super annotate.com and you download the desktop app or set up a meeting with the team in order to get a demo and a custom offer for what the team can do for you? That's very good. What else did you want to say, Tigran?

Tigran Petrosyan:I just want to say you can write to me directly also [email protected] and I'll make sure to direct you to the right person, personally. Since we're in this setup, I am also happy to directly get involved with folks who want to get engaged and direct to the right people.

Roland Siebelink:Perfect. And if anyone listening knows me and not Tigran and would prefer an introduction, I'm also happy to provide the introduction, of course. Thank you so much, Tigran Petrosyan, the co-founder and CEO of SuperAnnotate. It was an honor having you on the podcast today.

Tigran Petrosyan:Great talking to you, Roland, and very exciting. Thanks so much.

Roland Siebelink talks all things tech startup and bring you interviews with tech cofounders across the world.