#2278 How to build a $15M/year AI company.

0:28Hey there, Freedom Fighters. instantaneously would be dramatically different for a business. And so that's what today's guest set out to do. He created a company that does that. I met him last Friday and I said, oh, these stories that you're telling me are killer. We got to do an interview about how you built it. His name is Pavel. Oh, wait, let me make sure I'm pronouncing it right. Dolezal? Dolezal? Perfect. Pavel Dolezal. Company is Kabula. They use AI to make data more accessible for more use cases. is Pavel, you were telling me about that stationary store that's a client of yours. Tell the audience.

1:04Hi, Andrew. By the way, thanks. Thanks for inviting me. You know, Mixer is OG of podcasts. You know, like, so thanks. Well, the stationary company is actually called MacPen. And the guy who actually runs it, he was working for the large, you know, enterprise and they kind of like were shutting down the business. And he bought, I think, one or two shops out of them. And then he started to grow them. And then he knew that he needs to borrow money to grow them and that he needs data to run them. And he saw that in the enterprise, right? They had a huge team of dozens and dozens of data people to run the data.

1:41He didn't want that, right? He didn't want a huge team of data people. And his vision was that he's going to teach every clerk, every seller on the floor, and this is stationary business, that they're selling pens, rubber gums, you know, back to school, you know, items. And that every single person is going to be a data analyst and is going to contribute to actually, you know, making business better. Kind of like when you read the stories of Walmart, St. Walton, they would meet every Saturday and they would benchmark who is selling what. And he was like, no, everybody's going to do that daily.

2:19So he reached out to us. We built a project for him. We used Cabot to integrate the data. We clean up the data. We set up automation. So it would be always fresh. And then we set up dashboards in the software called Good Data. And then it was kind of like one o 'clock in the morning. I remember that very vividly because like we gave access to every single employee to dashboards. And he calls me like 20 minutes later. And he's like, you fucking, sorry. No, we do it. I'm like, fucking crazy? This is not how you do it. Immediately turn it off. And I'm like, no, no, no. You wanted data democratization.

2:57He says, yes, but this is anarchy. If you want to democratize something, this is not how you do it. You need to have a change management plan. So he actually forced us to lock down the system. And for the first three months, we worked only with him. After he started to understand, get hands on, you know, then we expanded to his leadership. after his whole leadership started to understand it, we expanded to everybody in the company. Long story short, the company grew from five locations to over 60 in like three years. And he didn't have any VC capital, anything. He borrowed money from bank, he paid it back.

3:40They went through Corona, they went through Russian invasion of Ukraine and the hike in prices, everything profitable and what they do this is like sorry this is like you know today we have a lot of ai clients you know everybody's crazy about ai but a lot of cool businesses but this is kind of like the most old school business in the world they sell you know pens and papers and every day you know it's mostly older ladies like 60 plus who actually sell there they come to the office They turn on their computer, which is their BOS system as well. They log into the good data and they see the data that they sold yesterday and they benchmark with everybody else.

4:25So have a leaderboard and what the questions he told them, questions they should ask. So they're selling something. What would be a good upsell? So like when back to school is coming, is it the best to actually upsell the rubber gum or is it best to upsell candy or whatever? and they compete with each other. And this is just awesome. And that's what this up-to-the-minute data that's accessible to everyone can do. It means not only that they get to see how these ladies' stores are comparing to others and how they're doing as salespeople compared to others, they're also getting insights into at this point in the year, today, if somebody buys this one thing, they're more likely to buy the other.

5:08So when you're selling the one thing, make sure to sell the other. The example you've given me was back to school. When someone comes in to buy a pencil, ask them if they want to buy an eraser. And it's an overly simplistic example, but at scale, you're talking about real impact to their business. That's what you're making available. And this is so huge impact, you know, because like you can literally double the profit margin just by these simple things. I always, you know, compare it. Remember, it's like digging for gold, right? Either you go and you find a big, big nuggets. There's just a couple of them.

5:41or you have a river and everybody sits alongside the river and they just you know like do this and everybody finds one small gold flake and a day but if you have a thousand people and they do it every day it's more than a big nugget a day right and it's kind of like what you can compare the company to you know the river is your processes people are everybody who works there and they're using data to get insights and if they find one flake every day all right let's go back a little bit you ran a huge portal back in the days when portals actually meant something portal was like supposed to be the place where a user would get onto the internet and then from there figure out where to go it's a portal called atlas you had a problem there that led you to launch this company um atlas was what i i don't know atlas i've never been on the site before well you would not be yeah well of course it was a clone of yahoo if you would say it was but yeah of course uh how big and famous did you get from it oh it was it was for eastern europe so we were you know like like uh very large in czech slowback number one in ukraine and so it was like 15 million people using it every single day we actually had you know like online maps in 99 you know like google maps was 2004 we had the online maps in 99 that it was awesome but honestly we didn't know how big it can get you know how how big a business it can be and we sold it to warwick and pinkus a couple of years later and the investment company yeah did you get rich from that oh i didn't i had just a couple of percentage but my my my friends who actually started that with me they got yeah because you were the chief product manager at the time not the founder yes got it Yes.

7:29Okay. And then right after that. I joined two weeks late. Seriously? Yes, seriously. Okay. All right. And the problem you had there, and I know you've launched a few companies after that too. The problem though that you had there was what? Well, it was always, you know, I launched a couple of companies and they were all using data and machine learning to actually automate processes. And the problem with Atlas was like 15 million people are doing some searches like in Google today. and I needed to understand what do they search for? And it took me like half a year to get engineers to build this view for me and to get all the data together so I would not bring down the systems.

8:12And so when I saw this, the same problem, exactly the same problem in three different companies I built, I was like, it's time to change it. And when the cloud started and the whole SaaS industry started, The proliferation of SaaS data sources, it was just a bigger and bigger problem. So I was like, this needs to get changed. I'm surprised. I get it. You're talking about with Atlas, 1998 is when the company was running. You were there until 2002. Okay, so back then, getting any kind of data was hard, right? But you've had other companies since then. We're talking about, what was the company that you had just before this one?

8:52You were at NetMail to 2019. I think so, NetMail. Yeah, so data was still too hard for you to get. Oh, yeah. It started to be even harder. I thought that when the cloud started, it would be like, okay. But remember, 2012, nobody knew Snowflake. It didn't exist. And so the de facto standard of data was Hadoop technology. It was built by engineers for engineers. it was originally started uh off of the white paper from google then you know like yahoo and facebook started that and it was apache project it was like so freakishly hard to do anything so it started to be even harder and so when i got together with my co-founders we were like they they saw the same problem from the consultancy you know angle they had a small consultancy you know doing it projects and then more and more people wanted this data you know and so they started to build it together and they saw it and so we started to actually get together we're like hey what's gonna change in next years we're like well there's gonna be you know somebody is gonna figure out the the database problem it's not gonna be hadoop it's we're gonna go back to seql enough and it's not gonna be one back and it's gonna be multiple backends so yes now we have redshift snowflake uh bigquery dcdb iceberg you name it you know there is there is like endless pletaphora of backends right and then we were like well there's gonna be we see this sas businesses you know taking off and there's gonna be you know like proliferation of sas businesses within the enterprise guess what gardner says that there's up to 300 different sas tools in the enterprise business 300.

10:44you know how hard it is to get data from two sources just like 300 and third one You're saying, look, the problem, you're saying, just to catch up on what you're saying right now, you're saying, look, the problem is that there's that, first of all, the people who are stored, the software that's holding on to data is not meant to be user accessible. So most people don't have access to it. And that's something that's always bothered you. Everybody should have access to data, number one. And the second issue that you said is there's also now, in addition to these data warehousing tools, there's also hundreds of other SaaS apps that people are using.

11:21And the data is stored within there. So, yes, it should be getting easier, but it's actually harder because of all those. And then you had a third reason. What's the third one? Yeah, and the third reason was exactly it should not be a magic. You know, you should not be a part of the voodoo clan that knows how to SQL or Python. You know, like this is a business problem. So you as a person, tech savvy, person in the business, should be able to handle all of your questions, you know, and automations yourself, right? So that was our North Star. So that's why we built Kabula as an API first, you know, so we can abstract from the technologies and change them, you know, under the hood as they evolve.

12:03And B, we were hoping for something like LLMs to come, you know, to come out very early on. It took us seven plus years. So we were ready for LLM back in 2016. And we're like, there's going to be something that's going to help us to write these SQL Python queries so the user can just work in natural language. Yes, that happened seven years later. By the way, you said not just get data accessible, but also do something. One of the discoveries that I've had over the last week when I've talked to different AI companies is they're much less excited about the agent thing than they are about the make data more accessible so every time i talk to them i want them to tell me about how ai will will magically do a thing for me and they say no no andrew uh like this company hyo he goes real estate brokers have data in all these different places they can't access it so what they do is they text someone on the team and they go can you tell me and that's a waste of time and it takes forever to get the answer so he goes what we created was a text that you can that you you know like a almost like a person on text you text this number you ask your question you get an answer instantly and we pull the data faster than anyone else he goes that's the exciting part it's not what we could do with the data it's can we actually make it accessible and you're smiling the same way well because that's the first problem if you can like uh you know for us this has been known problem for years so it's a hurdle you know like you see all these nice demos with agents and everybody, you know, but like if you look under the hood, what they are actually doing, they're, you know, using Excel or Google spreadsheet and maybe Google calendar, right?

13:42It's easy. But, you know, in the reality is these guys, you know, most of the interesting data is locked in the proprietary systems or, you know, specific SaaS systems, or they might have Salesforce with a different implementation, right? So that's hard. And that's not sexy. It's kind of like plumbing. But once you actually unlock it, and what we do, automate the data pipelines, they would run for eight years, 10 years. There is a company, DXC, like$16 billion a year company, the original IT company. And they use us to run all their sales and marketing data pipelines around the globe for last eight years.

14:31It just runs, right? So you automatically get the data accessible. Once you have that, you can start building those interesting use cases, and you can then start automated processes with it. But unless you have data accessible, no magic AI. The other thing that's interesting is, first of all, that you're saying first we make data accessible, then we act on that data. there's like an earlier step that i'm noticing come up over and over which is consulting that as a consultant not a software vendor not a sas maker those things are sexier but as a consultant you get to go in you have deep understanding of what the customer needs you do it a lot by hand and then eventually you create software that systemizes it you're smiling take me through the early part of the business when kabula did that well this is how we started so my co-founders actually had the consulting business, right?

15:25They would be deploying their engineers, you know, before we all knew forward deploy engineering was a thing. And just like consulting with clients, building the data accessibility, automations, insights for them. And so I came from the second angle. I was the owner of the business, you know, I had the issue. And so together, we started to go around different clients and we did implementations ourselves. And by that, we actually learned what are the hard problems, right? And that's how we started to build people. And that's the principle we keep up to today. We have a couple of clients where we actually do implementations ourselves and we actually learn and we co-design with them.

16:11One of the examples is one of the fastest growing unicorns in Europe, which is called Rohli Group. They do grocery delivery within one hour. Kind of like imagine Amazon Fresh actually working. That's running in Europe. Everybody's so used to it. And so we are designing things together because it makes sense. Give me an example of the early days. When you were just consulting or your partners were just consulting, give me an example of a project that was done that then led to software understanding that that led to something that's usable by other clients yeah yeah so so actually i will use the wrongly group because like the guy who started that is this is his third company and the first one was a groupon clone for you know six countries in europe which was actually very successful and profitable by the way interesting side note groupon as a original groupon is now run by checks as well you know some of our friends and client and it's our client actually have bought in as a private equity yeah yeah i have redesigned the whole coupon and data using kibula it's it's a great story you know and they are they are awesome team how they are actually executing but going back to tomas trooper is his name when he started his first company it was called slow and he like we found him on twitter honestly one late night you know he was tweeting was like again 2 a.m it's a magical number, you know, like between midnight and 2am, things happen with founders.

17:43And he was like, I've started this company. I have my database. I have a real problem. It's a MySQL database. And what I need, you know, I need to connect my sales reports. I need to connect this and this. I don't know how to do it. It's like so hard. So we immediately, my co-founder reached out to him. Next morning, we were on site. We started to discuss and he helped us to co-design, because he's very tech-savvy. He's a very business magician. This guy's incredible. And he's forward technologically. He likes technologies. So together, we designed that Kibla needs to be API first, right? So that abstraction layer I was talking about, that started with Tomasz Jupar.

18:24Meaning he told you, just do API. Meaning just suck data in from other API. He didn't tell us, just do API. But he started like, how do you want to use it? well have these you know sources that always change and then we took it back to our team right and we're like hey we need to have an abstraction there that's that's obvious right and so and then he was like well i can i know how to run the database i don't want to do it because like you know like it's just too much effort i want just insights so we started to run database underneath kebula right and so like these design principles we literally started with him as a client and then Then he used us in three other companies because we designed together.

19:08Right.

19:11All right. I get that. I see it. Now let's talk about some of the challenges. So you built the business. Then Corona hits. Right. Yeah. COVID-19. What happens to the business as you're building it? Yeah. So before Corona, we were, this is my, I think, third or fourth company. And we wanted to be bootstrapped, you know, in the beginning. so we could do what we wanted to do we had this vision we knew it's kind of a couple of years and we wanted so uh we started business we start to grow business pretty successful i moved to us in 2019 and then you know we start growing in chicago and corona hits so i need to stay in europe uh honestly we we need to you know nobody knew what's gonna go going on we need to lay off people in the US because we didn't have money to pay them.

20:04We need to refocus. I don't know if you remember, but in the beginning of Corona, the original or the first death in Italy, there was a mortality rate around 5%. That essentially means, that's why there was such a panic in the beginning. That means that it would wipe out the population within a year or whatever. right and so we're like well okay business is good but we should help now now it's kind of like time how to help use technology to to actually to actually help the governments and people and so we put out on twitter again hey check government do you want help and before you know it you know six hours later we are actually you know with the prime minister and the whole cabinet And then, like, we don't know what to do.

20:56You know, this is kind of like a technological play and, you know, contact tracing and everything, how to connect the data. And so we're like, yeah, we're going to help you. We can integrate the data. And so we did. But then it was like more and more. The governments were like totally not ready for anything. So we actually built a group of enthusiasts, you know, technological enthusiasts, 5 ,000 people, 5 ,000 software engineers, and they started to help. Well, that sounds great. the only two issues was that we started to help with one thing but then we started to see that they need more and more and then we started to run you know with couple of friends part of that government program for the government and it we couldn't get off because if we would get out it would just collapse and so we almost lost the business uh because like in the times where every like that company was selling and selling and selling we were focusing on you know like pro bono work and it was great if you if you look at the stats in the first wave of corona czech republic was best in the world but then the government people you know they are not really yeah it's a strange i i would not want to work with government ever more you know just like it sucked us in so much we're trying to change that system and after almost half year we had to say stop you know like our company is like almost going down and we are helping someone who doesn't want to help so we went back we actually it was it was it helped us in a way that we actually started the blg motion which we didn't have before uh and since then over 21 000 companies actually sign up for us elg motion we grew it we started to grow it you know and now like last 12 months we grew it three times so like there's you can always get something from crisis i think but yeah there's so many crisis building the company that you need to make sure that they don't kill you right and you need to have at least plan b and c and everybody says go in you know for I think that's great advice if it works out.

23:13Was the government grateful for all that you'd given up for them? No. No. It was like we knew that like half year later there would be great articles how we fucked up everything and, you know, go back. And did you get those articles about how you screwed everything up? Yeah, they're online somewhere. where you can, it's just like, you know, like there was 5 ,000 people involved, engineers. So. You said that it helped you get to PLG, product-led growth. What do you mean? It seems like you were completely in software, I mean, it seems like you were completely in the other direction then. We have two legs, you know, like a human being, two legs, they need to walk in the same rhythm, the same direction.

24:02so uh one uh originally we started just like like uh we didn't have sales originally well so we started just word of mouth and people recommending us and you couldn't actually start kebola online you had to talk to someone we had to start for you uh and then uh when when after corona like well it's great that we are getting these bigger clients how are we gonna work with us you know like we don't have money now to invest to go back to us physically so how are we gonna work with it so we're like well there is this plg motion that people seem to do well and so like like we were like lowered our principle was lowering down the barriers for people to actually start using curula right so we started like free account we don't have very very very generous freemium where people can actually run their businesses on that.

25:03And we're like, well, this is our way to go back to US because like, I don't need to be there to target people who are there. And so, and then it started to take off and it's actually, I think it's a, the world has changed, you know, after Corona. You know, I don't think that people are buying now just, you know, like by phone only or introductions or reference. Always, especially in our industry, there will be one technical who will want to try the software. So that's kind of like, even when we sell, and our tickets are, mid tickets are 75 to 150K as a starting ticket, we still have hundreds and hundreds of companies that pay us by credit card just like$20,$50.

25:53And they grow. or that there's thousands of people every quarter joining who want to try it. And then you can trace them to the companies that we actually do outreach to. I see. Let's talk about a bigger customer that you got. You had this shocking story about how you got a bank to buy from you because you couldn't get them to work with you. Why couldn't you get the bank to work with you? Well, imagine, right? banks are one of the most regulated industries, especially in Europe, right? All the GDPR and things, we invented that as Europeans. You know, bureaucracies are a thing, right? And so I don't think there is a harder customer to work with than a bank.

26:40But as well, you know, banks are great customers because everybody knows them, right? And they have great use cases, you know, especially retail banks. You know, so right now we are working with the Ersta bank and we are in over 85 departments. There's hundreds of people who use us in 85 departments but that took us almost 10 years to get there. So when we first started, I was like, I want to have this bank as a customer, right? It's a very well-known brand in Europe and etc. But like you don't get them as a customer like this, right? So we're like, okay, they don't typically buy from startups.

27:21They are very risk aware. You know, they have so much regulation, so much to lose that their first principle is not to innovate as much as possible. Their first principle is kind of like, hey, like do it securely. We are a bank. We actually work with people money. So we need to be secure. Yeah, so, but there was a trend of cloudification. so I knew this bank from the company I invested in Netmail and I was like this is a great customer but they will not buy from us for a couple of years so what we did we actually approached them we started to do community hackathons because we wanted to get people excited about data we wanted to show our platform and we were hoping to get customers out of them so we approached the bank with the prospect of doing a hackathon and so we used our data we anonymized the data and together we did the hackathon for over 500 people aws was sponsoring ibm was there google was there everybody took like four days three new companies were started out of that hackathon and it was it was just amazing so we established the relationship and then you know we worked with them for a couple of more years two three years to actually really work with them, do proof of value, to show them that we did a lot of security.

28:46And finally, in 2019, they signed the first deal. So exactly when Corona hit, we were starting the first implementation there. When you got that many people to participate in a hackathon, how did you get them? Well, just really being active in the community. so you know but first be active in the community so we did a lot of blog posts twitter back in the day and then we started actually to to work with other people like before corona the meetup.com was actually a very active website now it's not as much right so there would be people with different groups and we found out that they would have one thing they would have some audience but they would not have enough content right so what we decided to do like and before corner we actually ran a london data enthusiast group as well in london with several thousand people there and like every two weeks i think we would have we would have we would have an event you know there would be like 150 people like looker.com when they launched in in uk the guys actually came to launch at our you know like at our meetup it was awesome but so that's we would we would collaborate cooperate with people who would have their small groups but they would not have content and we found out that it's very hard for people to get content and to organize so we're like okay everybody has small audience let's combine together we'll organize we will do the content we will prepare the data we would do use cases we'll moderate and it turned out pretty well we've done this several times the huge hackathons and then we were like well this takes half a year to prepare right great data hackathon takes a lot of time so we were looking for a small concept which would be repeatable we got together with the back in the day 2015-16 there was a big it was a big women in tech movement.

30:49And so we got to go with one of the groups, Czech IT girls, and we started, hey, you don't do data. They're like, yeah, we would love to do data content. So we designed a specific workshop with them. And since then, it's been taught to over 30 ,000 ladies and it actually evolved into an academy which tried to be sponsored by Google, three months program, and stuff like that. I think it's great to work with community. I heard in your bootstrapping days, these types of community things and hackathons ended up being your Salesforce. How did you get other customers? Definitely. How?

31:32Honestly, people would invite other people. So the participants, our North Star for Good Hackathon was not a number of new clients. I remember our CFO back in the day, she was like, I need to understand what is the ROI on these activities. And we're like, we don't know. How can you do them? We have a theory or hypothesis, and we are ready to put all of our effort in it and just either prove it or disprove it. But it will take us half a year to understand. And it's okay with us. And so the guiding principle was to provide the value for people. to provide the value for a community, people would be inviting people, other people to actually build their use cases at the hackathons, right?

32:25So it was kind of like proof of value done on site. And that's just how people started to invite other people. So the idea was, I'm going to make a useful event when the people want to come. If they come, they might bring their friends and then all those people will get to know we're doing a Kabula, but not necessarily sign up right away. Yes. That's it. yeah and uh our also our guiding principle was we're not gonna do prizes because what happens with hackathons when you do prizes you have a professional hackathon hunters and they go and just swoop in you know they take the prize there and they go off that's no fun right so like no prizes well actually would have a price like ham right like spanish ham you know like big big chunk of hand uh or or something like that and uh that that was that's still working pretty well what's the revenue for the business now we are approaching 15 million ar one five 15 yeah one five wow you went bootstrapping for so long with all these different ideas why did you decide to take some money a few years ago uh for scale uh it's like we saw with corona that one big, you know, black swan event, what it can do to us, right?

33:50And when we saw in 2022 that what Transformers did with the GPT, right? And how it's kind of like our vision that finally there will be something that can help us with the data. and that these two things, we unlock the data potential, make it accessible. And there is the second part which can actually make it easy to talk to that data, right? We were like, wow, this is now, right? And so like, well, we had a huge discussion, honestly. It took almost a year internally. So are we gonna lose our freedom? You know, like, or do we want to, you know, like as many, like our mission is actually to automate every single business process with data and AI.

34:47That's been for almost... So it's not even to make it accessible. It's to get to that automation point. Yes, because the accessibility is the prerequisite for it. That's what we identified as the hardest, you know, like thing. And that's where people fail. but only coupled with that AI, it can actually, you can make it accessible. People can get insights, right? And by getting insights, they can run their businesses better. But then what they want to do, they want to use the same data, the same data, right? The same systems to actually automate their businesses. And so that's what we saw with AI, that it can actually have that potential and we analyze that we need more capital for that.

35:34right and so uh and you know a year and a half later that vision is actually you know coming together because like right now uh with our systems you can start with the question you know like then you know the systems the lm systems actually help you with actually defining what data you need like oh i have a i have my you know like let's say it will be mac pen running on Shopify, right? I have my stationary business. I want to see which customers have not bought from me in the last 90 days, right? You input it in. It says, well, you need the data from your BOS system. Oh, I'm using Shopify. Then Kebula helps you in one click to integrate the data.

36:16And it says, where do you have your store, you know, like warehouse, you know, in this system, like ShipMonk, Kebula helps you in one click to integrate them. But then LLMs help you to join it together. You know, the auto magic, right? And then you can get the answer, but that doesn't end there. Then people are, okay, interesting. Now I want to send it, you know, to something like Brace for audience, you know, like automation. And then people are like, okay, now I want to run this every day. So that's when you build the agent, right? That's when you build the agent. Not before. or let's kind of like putting the horse in front of the cart in front of the horse.

37:00Because first, you need to understand what are you solving? What is the process? And only then you can build the agent. And you need to have data. You need to have quality data. You need to do QA on the data. And you need to run the automations. So that's why most businesses are failing. Before you run the automations, you need human beings to do it to teach you what needs to get done? Or is that not necessary? yeah before you run the automations you know like like with systems like us you actually like the human beings like yeah like think about how do you get to describing the process either uh you have the you have some advisor or mackenzie guy coming in analyzing the system and analyzing the process and then saying this is it this is this you know goes with huge project The hard thing is business people don't usually understand how their processes work.