Hakob Astabatsyan, Co-Founder and CEO at SynthFlow: Driving Double-Digit ROI with AI Workflows

0:00Hello and welcome back to the Scaling Europe show presented by Deal. I'm Seb Johnson. I'm joined today by Hakob of Synthflow. Hello, welcome. Thank you for joining me. Hi Seb. Thanks for having me. How are you doing today? Great, great. We have a bit of very wintry weather here in Berlin, but other than that, all great. What's it like to be building in Germany at the moment? um i mean it's it's great right and uh i think we have very thriving ecosystem in berlin yeah i think just ai brought some revival i would say berlin used to be like i would say maybe after london and paris the third top in in europe but i think like during covet there was a little bit of a slowdown maybe it felt some subjectively like that i haven't looked into numbers but it seems like since this ai era like since a couple of years it's again thriving some some really good companies coming up yeah amazing the big thing in berlin you know five ten years ago is of course rocket internet which i know you spent time at um but before we get into all of that you know before before yeah before we get into the nitty-gritty can you tell me a bit about yourself and the company and kind of what you've built and who for?

1:14Yeah, I'm Hakob, one of the co-founders here and CEO at Sint Flow. I'm originally from Armenia, but I moved to Germany around 17 years ago. So actually, finally this year, it turns out to be more years here in Europe than in Armenia. And I've worked, as you mentioned, at Rocket Internet, at Boston Consulting before, but right around the COVID time I actually started my company with my two co-founders Albert and Sasson and first we had another company before actually starting SynthFlow which turned out to be an agency but then around 2023 we actually started SynthFlow just right when the first models of GPT were released.

2:11Like I would say you had an API and it was intelligent enough to do like any task. And this is where we started actually SynthFlow. And the original version of SynthFlow was to bring a no-code layer on top of the LLMs so that business owners and I would say non-developer community could actually experiment and play with AI agents, especially on the text-based agents. and around summer 2023 we thought okay what's the next modality of of text and either it's video or or voice and we started looking into voice and what really excited us about voice is that it was really really complicated no comp not even open ai no company was actually had anything to offer in voice AI space.

3:05And they were like in developer communities, they were the first ideas of the orchestration, basically converting speech to text, right? Passing it to LLM, getting the answer back and then converting the text back to the speech. But it was very hard to make that work. I'll deep dive into details later. But this is what really excited us, that it was really, really hard problem to solve. And since then we started working on voice AI, and invested all our resources. And we became the voice AI company. And since then, a lot of things have happened. But happy to deep dive on those down the road. Amazing.

3:45You're building voice AI for enterprise. Can you touch a bit on some real-life use cases? So how would an enterprise use your technology? Yeah, very good question. Look, when we started with building the first version of the product, just making AI to have a proper, meaningful conversation on the phone for like 30 seconds plus was already a huge challenge because the latency was very high. So just imagine when you and me, we just talk now, we have roughly 400 milliseconds of latency as humans. And that's why our conversation is so smooth. The moment you interrupt me or I interrupt you, we immediately grasp that and then make the flow, keep the flow going.

4:33But AI initially had like two or three seconds of latency. It means humans were not able to understand, is AI post? Is it broken? Is it going to answer to me? And the conversation would never happen. And then there were interruptions. AI would talk over, right? And all the calls basically would end up in a disaster, like in the original versions. So there was first this challenge to solve, right, of the conversational component. And, but once we solved that, it was still, I would say like early 2024, this is where we actually, sorry, we released the first version early 2022, and you were able to do proper conversations back in the day, right?

5:19But the cases were very, very simple. AI was not intelligent enough, you know, like we have now GPT 5.2, we have Gemini, we have cloud. But back in the day, the LLM versions were really basic. So LLM was not even smart enough to process complicated tasks, yet alone to do any complicated follow-ups, right? Or parse documents or reference complicated knowledge base. And the use cases it was able to do was appointment scheduling, FAQs, or AI receptionist, as many people love to call it. But ultimately, it's all about having a very quick 30 seconds to two minutes conversations to save time for humans, or I would say contained conversations, deflection rate, other metric to reduce basically the workflow.

6:15It was all about reducing workload for contact centers with this type of use cases. This is how we started SynthFlow. It is, I would say, pretty different right now, but this is how it all got started. And also, this is how most companies in the market actually do. They cover this very simple voice AI use cases and connect that with chatbots and etc. And people call it conversational AI. But in my opinion, this is not where it's really interesting. And I'm happy to deep dive on this. there is like i would say we're entering into the second wave of conversational ai but but that's how everything started for us amazing and so far things have been going well i read that you've done is it 45 million calls you've handled or something like that yeah we're handling at this very moment we're handling many million calls per month uh and um so and and what we also so So in our first year in 2024, we actually grew our revenue 15x.

7:22Last year, we grew 3x. And we expect to grow our ARR another 3x this year. And we process several million calls per month. And there are also like other metrics we measure, like the action items executed, right? The success rate. I think that's a very important metric for the customers. but that's where we are at this moment. Amazing and that's incredible growth. Was that one of the reasons why you were able to raise I think it was a 20 million Series A led by Excel last year? Exactly, yeah. So I think there were two reasons why we actually did the round. One, we were growing very fast but I think what we also do at SynthFlow is pretty relatively unique.

8:11It's very hard to discern that uniqueness from the outside at this very moment because AI is generally a very noisy space. And I mean, look, just the fact that we call everything AI, it already tells us like so high level, right? Like, what do you mean exactly, right? Everyone's AI. So I think it's a noisy space. But ultimately, what really matters is who are your customers and what exactly do you do for these customers? And what is the deep value that you are delivering to these customers? And I think these two points were critical. And the reason we actually did the round is because we reached the moment where we said, okay, it's time we grow the organization.

9:01There is only as much as you can do with limited resources, right? And don't get me wrong, there are companies, right? There was this narrative, you can build a billion dollar company with three co-founders. And that is true for some businesses, especially like very PLG, consumer oriented businesses. however because we moved serving larger companies enterprises it meant we also need to scale our workforce so that we can actually serve them and we grew from roughly in early 2024 we were like six people and now we are 70 uh employees that's in flow yeah amazing and when you look at the enterprises that you're selling to what kind of customers are coming on board yeah so originally Finally, we started our SynthFlow offering was about basically AI contact center for enterprises, right?

9:58All companies that had in-house contact centers or outsourced their phone operations to BPO and other contact centers, basically humans talking on the phone. We said, look, why don't you connect AI with basically your human workforce? and gain efficiency there. So I call it the single digit ROI space, right? So for example, look, I'll just give a specific example. I think that's easier. Imagine you're getting like 10 ,000 calls per month, and it's all about, let's say, requesting information or balance on your account, or maybe canceling something. Probably like four or five thousand of that calls could be erroneous in the sense like someone forgot to check something, right?

10:53So it's very simple to handle cases. And if you can do that part with AI, sort of tell, hey, what's your number? No, actually, you're wrong. Your credit is full. You don't have to upload, like this kind of very simple things, you can say 5 ,000 calls for human agents. And whenever it's complicated, there is a real issue, right? AI would hand it over to the human and the human would handle that. It was all about that, right, in the beginning. And I think many companies are still about that. Most conversational AI, voice AI companies, it doesn't matter what they say, is they're handling this type of cases.

11:31And I call them single digit ROI because, okay, you saved, let's say, 8 % efficiency, right? So what it's all about right now for SynthFlow is that because we have been very specialized on this contact center cases and on the phone, right? So our wedge is very specific, like we're a phone operating AI. We also build our own telephony system. And if you look, very few companies actually do that because telephony is a very specific business. But this is where we started going very deep. And what we're handling right now with SynthFlow is what I call the double-digit ROI cases. To give you a specific example, what double-digit ROI means is, let's say you call and you want to extend your credit line, or you want to, let's say, qualify for a loan.

12:25Right. This is very complicated because it necessarily involves multiple parties, let's say two, three plus parties with different functions. Right. Someone has to approve. Someone has to check. Someone has to audit. It also has to happen in real time. So we call it time sensitive cases. And these are very complicated to do. Plus, there is like context specificity. It means what is the context of that company, right? If it's a bank, right, there are probably hundreds of systems that need to be updated for that type of thing. So this is where we actually focus a lot at this moment. We're very specialized.

13:11And the hard part here is, to sum it up, it's not the voice part. Voice part is, in my opinion, to a large extent, a solved issue, at least for us, but I'm also sure for a few more companies. latency is very low right conversations go very smooth voices are amazing so if you talk to a very good AI voice AI company agents basically they are really good the hard part is the workflow right what happens after that this is really hard and this is I think what SynthFlow is all about going forward. Interesting when you're pitching to an enterprise and let's say you lose a deal, are you losing it to a competitor or are you losing it because the enterprise maybe is not quite ready to take the leap into using AI in this way yet?

14:00Yeah. So originally losing a deal to enterprise was, so the number one reason that we identified originally that we were not able to serve was the wrong ICP, right? But what we understood is that very quickly, you cannot serve everything at the same time, right? So what are you really good at and for which customers, right? So for us, the customers that are basically looking for AI contact center with for very, I would say, double digit ROI cases, we have been winning this actually cases very reliable, I would say continuously. And that also explains our growth, right? The reason is that we do not position ourselves as a generic voice AI company, right, that just does very good voices or voice.

14:56I think that part is great. But in my opinion, what the companies enterprises look for is they look for something deeper, something that actually can do a work end-to-end and extract very deep value. And the single digit ROI cases, in my opinion, they are going to get commoditized, but doing a deeper work, right? And more end-to-end workflows is where the key for us, right? Also, like, for example, we don't do cases, like if someone wants to do, let's say, an HR recruiter, right? On their website, that's not what's in flow for us so initially everyone would come to us because we were one of the very first companies out there and and of course if you search for keywords in the inevitably sin flow is one of the solutions you would stumble across and when we started having those type of conversations we would very quickly start to understand okay we need to have a better qualification mechanism of this company and then we started implementing those etc and there were way more tailored and i would say our last qbr we had we looked at our conversion rate basically right the success rate in our pipeline and it's above industry average it's actually pretty high what you would expect and i would trace it back to a very good basically rev ops gtm ops that we did at synthflow in the past year nice and there's it feels like there's something interesting happening across europe within uh voice ai you know like yourselves uh parlo voice in fact all three of you are german but then you've also got like poly ai even 11 labs to an extent you know there seems to be like this real strength within europe on vocal ai voice ai in a way that we don't we absolutely don't have the strength on the llms why do you think europe is is kind of taking the lead in voice ai over other regions you know it's an excellent question i think if we were not doing podcast, but who went for a beer, we could have very long conversation about this.

17:01But it's very interesting. So I don't have a definitive answer. But in my opinion, so there are two things top of my mind when I think about Europe. So one is, I think Europe has an incredible talent pool. But Europe is a sleeping giant, right? The problem is not the amount of talent. The problem is the talent density. So, right, if you're going to like geographical places, like if you go like to Bay Area, right, or New York area, these regions, they have very high talent density. So a lot of talent at the same place. and there are also like very strong network effects there, which allow companies to very quickly kind of scale the work talent, right?

17:55In Europe, it's a bit harder. So based out of Berlin, you just have limited talent in Berlin and then you have to go beyond and then it becomes a bit harder, basically, right? You have to start combining Europe and then you have a more distributed workforce. So that's kind of one challenge I see here. The second is the mindset. I think there is incredible talent here that's not working on innovation. In my opinion, of course, I'm a founder and I'm biased. In my opinion, that's such a waste of talent if you go and work for a corporate. And maybe there are some incentives not in the place, like economic, I don't know.

18:40You can deep dive into these options and all this stuff. But in the US, I think the mindset is way more favorable for startups and innovation than in Europe. That said, right, you're absolutely right. Like there are winners like across the board coming from Europe. And I just think that's our European founders, right? European founders just woke up, right, one day and decided, okay, we're going to go global. And look, I think we always had that. But the problem before AI was that it was very hard to scale globally. Let's say if you would build an e-commerce company, and I remember from my time at the Rocket Internet, right, you are, let's say, let's say you do some e-commerce and you start with Berlin, right?

19:28Then going outside of Berlin to Germany, okay. But then if you want to expand to US, to UK, it's very hard because you have to sort of physically scale the business, right? You have to open the office. It becomes very expensive and it's very slow. And I think this is why European founders were not winning globally, because it was very restrictive. With AI, you don't need that. Let's say if you look at amazing products like Syntesia or other products, it's like you can scale way faster. Of course, you still need a team in the U.S., etc. But if you have an amazing product and amazing distribution and positioning, you can start monetizing in the U.S.

20:06and other countries from day one, actually. And I think that's the key why suddenly this hidden champion guys or girls from Europe are winning on a global scale, right? Look at NA10, Syntesia, right? All these companies. And that's probably, in my opinion, the reason. and yeah amazing and looking at this technology and i guess what i want to ask is if everything goes right you know if you continue to develop your technology if you continue to to make the sales and to scale how how big a difference will this make to enterprise clients like what's the you know 10 years time if the technology keeps getting better if things keep going the way that we think they will how different will the world look like because of a product like yours look I think in terms of, at least speaking from Synthflow's perspective, right, I don't think the voice component, so it will get better, right?

21:01There are like voice-to-voice models that are going to come and the moment they become commercially viable, also amazing conversational flow, etc. But I don't think this is where the real value for companies is. The real value is, are the workflows, right? Like how much work can I actually automate end to end, right? Okay, there are chatbots on the website, etc. But talk to every leader at any enterprise. They will tell you single digit ROI at most, right? Really going deeper, right? Integrating into the context, like the context specificity is a huge topic. I think this is where we will be seeing the real value coming in the next couple of years.

21:48This is at least where we're working and putting most of our effort, is how much and how complex of workflows or work can we actually automate. And if everything goes well, right, everyone keeps innovating, like LLMs actually deliver, right, everything starts with, like down the, like early in the value chain, right? First, the GPUs have to deliver, then the LLMs have to deliver. And if all the layers start really improving, I think we will be seeing ultimately from our side, basically the companies that actually delivered the value directly to the customer, this is what we'll be seeing. More and more sophisticated workflows actually get automated.

22:37And this is my vision and this is my dream, to be honest, to see that actually this tangible end-to-end work automation. Because when I started, everyone was so excited. Everyone was thinking about this auto-GPT. You know, there was this baby AGI that you just hit the button and AI books your entire holiday and then whatever, right? So this kind of things. But actually, we start hitting now this value of enlightenment, right, where all of us are just saying, okay, wait a minute, the truth is somewhere in between. This is still amazing technology, but there is still a lot of work that needs to be done to actually see that ROI.

23:17And let's keep on working, right? Magic doesn't happen overnight. Let's keep working on this and see what is actually feasible. and this is where we'll be spending most of our time. Amazing. Well, thank you so much for joining me today. I've got one more question for you. What do you need to achieve next year or the year after to get you from your 20 million led by Excel to your next big round, your next big success? I think one thing that we are extremely proud of as SynthFlow is our retention with our enterprise customers. We actually just closed the year and we had 100%. logo retention, logo retention with customers, which actually, you know, retrospectively, it's nice and expected and kind of that's the playbook expectation, I would say, right?

24:10But if you go back into 2024, that was like probably the biggest fear all founders had. Like, is this actually technology that's going to stay? Are you going to able to renew your customers. And of course, customers aren't stupid. They will only renew it if they see a significant ROI or value out of this, right? And this was a big question. But seeing already like early signs, or I would say like good signs that this technology is delivering was kind of very positive for us. And we also scaled our organization. So we were actually able to serve our customers. And I do still think that you need very good forward deployed engineers and et cetera to actually alongside with your customers, right, to help them basically kind of adopt this AI technology.

25:05I think one thing we need to do this year is first of all, we need to deliver on the expectations that we have, right, on our targets that we've set for us. and we also need to be a bit more specific this year about SynthFlow's value, basically what I just shared, but we have to be more explicit with the market. I think we are entering into this 2026 where each company has to be extremely specific and deliberate about their value, right? Go away a bit from this AI for everything and be very deliberate, okay, what I'm really good at and what I'm delivering to you and this is the homework for us for this year and we're going to work on that.