How Superlinked is building the Next Generation of AI Search with Ben Gutkovich

Foreign. This is Rochel Nares from Mastery. And so I'm excited to have Ben Kutkovich, who's the co founder of superlinked, which helps AI engineers build information retrieval systems using vector embeddings. And he's got over 20 years experience with companies like McKinsey Strategian and operation roles like head of Business Development at Easy Car Club. Ben holds an MBA from London Business School. Welcome to show Ben.

Thank you so much. It's a pleasure to be here.

Excited about our awesome. Yeah, no, absolutely. You know, thank you so much for your time and you've had an interesting career but you know, before we start, about your experience in, you know, in the consulting world. But you know, how do you get your start in this, in this world of technology and startups?

Absolutely. So look, I did study my bachelor's degree in computer science, so started my career in software engineering, but straight out of uni, I actually went to work for a defense contractor. So as you can imagine, innovation wasn't top of mind for them. So it was quite a rigid, strict environment and, and it was more around the code being reliable rather than doing really exciting things or driving impact. And you get this feel of being a cog in a big machine. And for me, from the get go it was important to deliver impact. And so during that role I met my co founder. He did my first startup and that was more than 20 years ago now. It was an asset management platform back in the day, actually using some AIs and neural networks already back in those days. But you know, as we were all meeting together for coffee in that company and we're discussing how all rigid and kind of restrictive the environment is and dreaming about our own business, we decided both to leave and actually start that business back in 2004. Ever since, you know, got infected by the entrepreneurial bug as you. As you might say, and was looking to go back to those entrepreneurial opportunities and had a couple of other corporate roles in my career through which I learned a lot. But it was always my dream to go and again start my own company.

Very interesting because you know, you, you've done your computer science engineering and you've also done your MBA and you work with likes of McKinsey, PwC and now you founded an AI startup. So what's the biggest difference that you see between advising on transformation and actually building that sort of technology yourself and now scaling it up?

Yeah, I mean that's really interesting, right? I think advising is more of an intellectual exercise. You don't really Own the outcomes, you put some nice slides together, you can do a model that predicts what the outcomes are, but you don't actually own the implementation, the execution. And I'll tell you a little secret. Sometimes even some of the bigger consulting companies I worked for know that the client doesn't have the capability or even the political capital to execute our suggestions. Right? And for me that was actually quite disheartening because you work long hours in consulting as well. You work long hours in startups, but also in consulting it could be 15, 16 hours, days, easily works in weekend, etc. And then your work doesn't actually get implemented, it get executed and you don't have control over that. Right. And that was the thing that was missing for me in consulting. Learned a lot, got great exposure to different companies, different industries, different types of problems, which was fun, but always wanted to do things with my hands. And in startups, the difference is that you cannot just come up with any dreams and ideas and put on slides and hope someone implements it. You actually need to go and build those things, which is much harder, right? You basically leave the consequences of those suggestions and initiatives that you propose. So I think, yeah, for me it's better, but I can also, I know that that's also harder on a day to day basis, right, because you leave the consequences of your ideas, of your initiatives. And hey, you're not always right, you're actually often wrong. But it's all about learning from that. And I do enjoy that more than just giving advice, hey, hoping that they take it and do something.

Got it. Super interesting. And you know, especially in the last two years we've seen there's a boom in, in the world of AI where there's a AI talent war heating up with, you know, especially Meta. And few of the companies giving, I don't know, hundreds of millions of dollars. I don't know how much of that is the truth, but as someone, you know, who's recruiting in an AI startup, how do you go about competing when you know you, maybe you won't be able to match those sort of salaries, but you'd want really high quality talent to, to join your team.

Yeah, that's, that's a great question. And one of the bigger challenges is AI startup these days. I totally agree. But look, we can't outpay meta, right? And so we take the approach of, like I also talked about before, of getting people to feel that they can deliver impact, right? Or give the meaning to their work, you know, excite them about the dream and the vision of what we are building and we believe, you know, it's also a good recruiting tool if people just start motivated by money. Those are not the kind of people you want to have in your startup because startups is all about ups and downs and roller coaster. Right. So we want those engineers who want to deliver impact, who want to own their results and outcomes and don't want the bureaucracy that comes with a large enterprise, even, you know, an innovative enterprise like Meta. And at Superlink, we make sure that they deliver input. Basically from day one. Our ML engineers work closely with the clients, help them get the results and celebrate those results within the team and also with the client. So this is the kind of people we want and this is the kind of people we are still able to attract even with this competition on salaries with Meta and other AI companies.

Got it, got it. So super interesting. And, and you know, you work with, I mean, you're also in the BD of EZ Car Club, right? So especially, you know, I've been in the, in the world of business and partnerships. But how do you go about, you know, you mentioned that you went about doubling your owner recruitment rates in EasyCar Club. So what was, you know, some of the incentive structures that you're able to give across in your partnership team? And were there any pricing innovation that allowed you to double that recruitment rates when you're working for EasyCarClub?

Absolutely. So it actually came through talking to our car owners who are renting the cars on the platform and understanding their mindset. And what we understood is, is that for someone who already paid for the car or is already paying a certain amount, whether to rent out that car to other people is another decision they need to make. So we decided to reverse it. Right. So that not renting out the car would be the decision they need to make. And we did a partnership with one of the largest leasing companies in the UK to offer part of their inventory to EasyCarClub car owners at a reduced price with the condition that they rent their car out a certain number of days per month. Right. And in that case they got a really good deal. They got a car for, let's say half of the price or 75% of the price that the car is leased out in the market, but now they are committed to rent it out and actually if they don't rent out the car and they will be paid more. So by changing the mindset of the car owner and actually allowing them to take maybe a bigger car or maybe a second car that they couldn't actually afford for a lower price. We managed to recruit many more car owners and this was one of the first types of this kind of proposition in the world and the first one in the uk. And so that really helped us grow, grow our car owner base, which in the end resulted in acquisition by Turo. Right, so Turo, one of the bigger companies in the US, operating in the same model, they came into the UK in 2017 and they failed to actually recruit car owners because, well, they didn't have the base to learn from and, and it was very hard for them to go and build this car owner base. And Indiana acquired Easy car Club in 2018 to actually start, start their business in the UK in the peer to peer space.

Got it, got it. So super interesting. And you know, I've worked in the UK as well as in India and I've seen like remote work has become like a, like a default, especially after covert, you know, since 2020. So what's, what's your, you know, take on, you know, do you think remote work would hold AI starters back or do you think, you know, like the 9 to 9 to 6 sort of culture should work in Europe? What are your thoughts on that?

So, you know, we started during the pandemic, so we started in 2021. During the pandemic, you know, best talent everywhere worked remotely and pretty quickly discovered that it doesn't work fully. Right. So if the tasks are well defined and people know what they're doing. Yes, working remotely, working in your comfortable, convenient environment without the noise is really good. But when you're building something revolutionary and innovative, which requires a lot of this back and forth and collaboration and thinking together and whiteboarding. Right. It's very hard to do over zoom. And so after the pandemic was over, we went towards a hub strategy where we have, we co locate people in hubs and there they go to the office at least once a week, in some locations, twice a week to actually collaborate. So for example, in the UK we meet in the office on Tuesday and that day is mostly collaboration within the team. Team. It's learning, it's brainstorming, it's whiteboarding and then everyone goes back to their home offices and during the week they actually are more productive and deliver results. So yes, I think, well, I again, I think remote work, only remote work can hold AI startups back. Especially if the challenge is not very clear, well structured and clear where the company is going.

Got it, got it. Makes sense. And, and you've been part of a lot of Strategy companies and M and A deal. So when you're looking at potential for M and A deals, what's one financial operational metric that lets you to decide whether this deal will create that sort of value or not?

I think the gold started of MA is operational cash flow, right. You want to look at whether there is actually an increase in the operational cash flow as a result of the synergies, because it's very easy to say, oh yeah, we go to the same customers and we expand or kind of sell new opportunities for the same customers, but, you know, it doesn't necessarily happen, right. So making sure that there is actually an increase in operational cash flow flow, either from revenue strategies or from cost strategies, you can actually reduce the number of people in the company or, you know, cut costs as a result of the M and A deal. That's. That's what you want to look for. So, yeah, you, a lot of people look at like, oh, revenue synergies or cost cutting, but I think it's really looking at the cash in the end. Cash is king. You want to see increase there as a result of the enemy.

Got it. And especially in the AI space, a lot of people are thinking, is there an AI bubble? But with the kind of investments which are happening, looks like, I don't know, looks like 20, 20, 21, where there's a lot of investment going on. But what are some of the biggest predictions for the AI market in the next year, year and a half, that, you know, you think most founders are either ignoring or they're not getting it right at all?

I mean, it's hard to say if most founders are ignoring. There's so much noise in the space. I mean, the space is moving so fast. There's so many voices and people are saying lots of contradictory things. One thing I think is probably a bit premature is agents. Right? I mean, there is a huge boom this year around agents and building agents. And of course OpenAI has just launched their agent building studio, which I think, you know, they're all great developments and the future is probably agentic, but I think not as fast. Right. So I think we'll still reach a plateau of this whole agentic system deployment. And until the framework around trust and control of those agents developed properly and tested properly. And I think, you know, we work mostly with enterprise clients. The promise sounds really great, but their mindset is more reducing risk, more about reducing risk and letting those agents go free in the enterprise. Doing whatever they want is not something that's going to happen anytime soon. No, you know, as in next 18 months. So I think in that regard, while there is a lot of money pouring into companies that are building agents, we will still hit a plateau before we get to real productivity trajectory and adoption by enterprise, big enterprise clients of frameworks.

Got it. And just to double on that, do you think a lot of this specific AI agents coming, coming on board and you know, you, you have people like Jason Lemkin talking about, you know, 20, 25, 26 to be the year of AI agents, especially a lot of, you know, SDR roles going away and all that. Do you find, you know, companies like Klarna and FIO and Salesforce saying that will have lesser number of stuff sales reps or customer success reps in the next coming years or do you think there'll be more number of, you know, revenue rules coming up in next couple of years? If you look at Klarna, you know, they had, you know, removed lot of the customer success reps and they, and they filled out, you know, now I think they have gone back and they took them back.

Yeah, I think it was a nice publicity. Distinct. But o. But in, in the end they brought them back.

Yeah, yeah, yeah.

And that's what I mean. Right. So I think, look, in the longer term that probably makes total sense, but I think there is still quite a lot of guard rail to develop around those agentic systems to get the trust of the enterprise. Right. I mean especially the bigger clients and that everyone kind of aligns to the bigger, to the bigger enterprise in the end. Right. Those, those bigger clients will take time to really adopt it. And, and I think still needs to be a lot of building a lot of testing, especially in Europe. They'll want to see that, you know, there aren't too many failures. I mean there's a story about the agentic system for Air Canada. I think that promised super cheap, like 100 super cheap tickets and they had to actually go and deliver it. And then of course they removed that system from production. Right. So, so I, I think there will be still. It will still take time for those agentic systems to really take hold. So I don't think next 18 months, but hey, let's see. But it's good to see to say it. Right. It creates headlines and everyone of course is worried about AI taking our jobs. And so talking about this, this is good pr, right. It definitely attracts attention. What happens in reality? Yeah, like you know, Clarna. Oh yeah, yeah. No, no more customer success people. All AI and then suddenly quietly bringing them back. Right.

Right. And, and and obviously, you know, with, with Lovables and Curses, you know, and a lot of companies like Replay are also coming on board. But you, you, you started computer science, but would you advise somebody who's 18 years old to do computer science now or what are your thoughts on that?

I mean, absolutely. Look, my co founder who's an ex Googler, he's now using those tools like Lovable to build demos really, really fast. But actually for production systems, they're still not good enough. And even for the demos you need to guide them quite a bit. And so if you understand computer science really well, if you understand software engineer, this is the tool that makes you 10x more productive or maybe 3x more productive at this stage. Right. But it doesn't completely replace the person I haven't yet seen, you know, production grade applications really built on lovable and useful by enterprise clients. It's great for demos, great for prototyping for sure stuff still not there in terms of production grade applications.

Got it. And you know, in the last couple of years I've seen, you know, with SaaS companies, they always had a seed based pricing, but now you seeing like a, like a hybrid pricing where there's, especially with usage based pricing, adoption has doubled in the last couple of years. So where do you think that adoption is coming on board and do you think that sort of adoption is going to increase from now?

I would say so. I think there are, there are two aspects to it. First of all, with those AI workloads, you can actually align pricing to value, right? You run those workloads, you use GPUs, you use inference that actually should be delivering value. So usage based pricing makes sense in that regard. And the other thing is that, you know, it's more about having data and about replacing guesswork with what actually, what you can actually see. Right. So I think that makes it easier to use usage based usage based pricing. Right. Because before you maybe it's just very difficult to see how the software is used, how much is it used. Right. So having good telemetry, good solutions for actually tracking usage, plus the AI, the nature of AI workloads, I think that makes sense. It's probably something that will grow in the future. So usage based pricing over seats. And you know, it's funny, when I was at Beckett Orange, we implemented this platform for running campaigns and of course it has a seat based pricing. But then, you know, and we had like three marketing people who had seats on the, on the system and they just didn't use it right. And so we were paying but not getting the usage and not getting the value. And, and of course you know, it didn't took, well it took some time because it's still an enterprise but it didn't take that long to actually just drop the system and stop using it because you know, we don't see value out of it. And I think if there was incentives better. Right. So if there was a usage based pricing and the company that provided the system could come and help us use it more and not just say hey great, we sold three seeds and this company has paid us a thousand a month and that's done. But rather actually help the company use it and align the incentives then I think everyone would be in the batch of place you would have a better campaign management system that's actually used and they would retain the customer. So totally makes sense.

Yeah, absolutely. And with user based pricing, you know I've also seen they spend more on R and D but than on you know, sales and marketing. But would that, do you think that would also translate into better unit economics, you know with, especially with outcome based pricing also coming into picture?

Yeah, I mean I, I think it kind of depends on what, what your products sell on. Right. Whether the product sells on capability and actually delivering value or you know, brand. Right. If you just sell on brand and you're not really focused on delivering value, I mean that's probably short term strategy anyway. But if you sell on building those new capabilities that's actually useful for the customer. Again as you say, usage based pricing shows in real time. Right. Almost in real time that those features are actually useful because people are using them and then paying for them. Right. If your R D can deliver that. And of course that improves in your economics if you build a bunch of stuff that nobody actually uses. Just expensive tinkering. Right. Not really improving the economics in that.

Got it, got it.

Got. Got to separate this in short term, medium term, long term. Right. In short term maybe, but in the longer term it won't work, won't hurt.

Right, right. And, and you know you, you've raised a, a large seed round and, and, and you know you, you present across uk, US and you know on other countries as well. But how did you position your market opportunity to Index Ventures, which is one of the best VC firms and any learnings for founders who are looking to place that sort of money around this market.

Absolutely. So I think we had a combination of a big vision, this whole vector compute layer which I could talk about but also some proof. Right. So we started from Actually building a full stack recommender system. And we had clients, we had clients in production. So we proved that we can execute and that our thesis of vector compute works, but we did for ourselves, not for the clients, but the clients used what was built on top of our infrastructure. And then coming to the big vision, it was like, okay, look, we, we built this vector compute layer that is now used by clients who are paying six figure contracts right to us. And it works, right? It works in production, it works at scale. Enterprises who want to use AI applications, build AI applications could also use the same type of infrastructure and actually make their AI development more predictable, higher quality and you know, 60, 70% faster. Right. So going to production from six months, which is your typical AI project, getting to production to just three, four weeks. Right. And so that big vision with the proof of us having production clients paying significant amounts for the value we deliver. So the ability to execute was the thing that I believe convinced Index Ventures. Plus obviously great founders.

Got it. And, and especially that, you know, looking at the shift from, from LLMs to custom models and now opens these alternatives, how do you see in our economics of a infrastructure heading and, and where do you see that, you know, Superlink is going to position itself there?

Yeah. So I think, you know, what we see now with clients that we work with and it's very typical. Right. It's the same as happened with the data stack. Right. It's all about modularity. So everyone is blending different models. Some custom, some open models, some proprietary models, some train models in house. Train models, Right. So they want this flexibility because different models perform better in different use cases. Of course you want to optimize costs as well. Some models are more costly, some less costly. Right. You don't, you know, one of the things that Superlink for example allows is to use some open source, obviously much cheaper, smaller models which you can host in house, but by combining it with other types of data, you make those models perform as well or sometimes even better than the proposal. Proprietary, expensive models. Right. So it's all about really blending, combining modularity of different components and different models. And we believe that, well, our goal is to enable it in the enterprise. It's really kind of stitching all of those models together. It's what the vector compute and our infrastructure is about.

Got it, got it. So, super interesting. And Benny, I quickly want to do the top three. What's your favorite business part? Yeah.

So I don't know if you're familiar. The hard thing about hard things.

Yeah.

Ben Horowitz.

Yeah.

I think a bible of entrepreneurship. I really like it. I mean, I read it several times, but I also really like how it's organized, you know, by chapters and by specific topics. So it's very easy. So I'm an audiobook listener, so I listen to it on audio. But I also have the, the ebook copy where I can go back and just jump to the chapter and just remind myself some of the techniques and strategies that he employed in his businesses, which, again, I think are very good, very useful. It helps me think about those problems as well. And so, yeah, I think you don't need to remember what exactly in which chapter it was because it's really organized by topics. So I really like that. And of course, was a very inspiring job. Me?

Yeah, absolutely. I think, I think it's, you know, Ben Horowitz is a great operator and I think this book has been the most recommended among. Among all the guests. Yeah.

Not very original. Sorry about that.

No, no, it's a great book, by the way, but we'll put that in the show notes. And, you know, if you could go back in time when you started Superlink, what is the one thing you would have focused on or done? Thinking differently.

Yeah, so I think so. So we had this transition from a full stack system which we very early went and built with the clients, right. And, and, you know, got on those contracts and delivered value and we had those great results. When we raised the seed route, we went on a journey of unbundling. So it's not necessarily just when we started. It's actually the kind of midway in the transition from, from the full stack end to end superlink system, recommender system to the infrastructure on which the enterprise clients can build. We took way too long to work with clients, right. We took, well, let's say probably six months longer than we should have to get to actually work with clients because we wanted to design the perfect system and, you know, probably should have known better, but I think you learn much more by building with the clients. And now we actually have several really great design partners. I mean, I can mention some names already that we have a new website coming, so it's all approved. But guys like Trivago and Thredup, some really great companies that we've been building with in the past year and their feedback just been invaluable. Right. So I think we probably lost quite a bit of time by trying to design a system which is great for everyone, right. But without actually the rubber hitting the road. So I think as soon as you can, I Mean again, I'm not being original here, but getting to work with customers as soon as you can, even for free, even subsidizing the deployments, I think this huge value to it because in the end one of the main, I mean, money is of course king, right? Or cash is king. But time is super valuable for startups, right? If you don't move fast enough, you lose your main advantage because the big guys, they have the brand, they have the money. The one thing where the startup can win is just being fast and nimble and iterating fast and getting fast to the market that the big companies cannot do. And so, so yeah, get early to work with the customers.

And do you have any favorite online tool, for example Gmail, Slack, Zoom, ChatGPT?

Well, I mean ChatGPT, but you know, I won't be original saying that as well. I am a big fan and I'm using it a lot. But I was actually thinking that whiteboarding is probably one of my favorite tools. And again, it goes back to the discussion we had around working remotely versus working in the office. You know, every Tuesday there is at least couple of hours in front of a whiteboard. So I, I run operations and also go to markets on sales and marketing. And so we always like come in and we think, okay, this is the data we have. Let's start brainstorming, putting some ideas, connecting the dots. I don't know, maybe it's my. I still do also written to the do list by the way, but.

Maybe.

Just through handwriting I come up with the best ideas that I have. So for me, just brainstorming on the whiteboard, taking a screenshot of that and that afterwards become our plan for the next month or two or three. Right. So for me this is really useful too. And I did during the pandemic use those online whiteboard tools and it just didn't work as well for me.

Absolutely. We'll put down the show notes and, and you know, Ben, what's, what's the best way people can reach out to you and know more about Superlinked?

I would say LinkedIn would be best. I'll send you my URL. I am trying to be responsive there and also share some of the content both about the company and about my journey and so happy for people to follow me and connect and yeah, happy to get some questions if they have.

Awesome. We'll put down the show notes. Ben, thank you so much for taking your time speaking to us. I really enjoyed my conversation with you.

Thank you, Rohit. Same here. And I hope it was helpful for the people listening. Thank you so much.

Thank you. Thanks for listening to the Life Self Mastery Podcast where we teach you how to start and grow your online business. For more information, visit Rohit's blog at www.lifeselfmastery.