Season 4, episode 4 of the DataTalks.Club podcast with Elena Samuylova
Alexey: Today we'll talk about building machine learning startups. We have a special guest today, Elena. Elena is the CEO and co-founder of Evidently AI, which is a startup that develops open source tools for machine learning model monitoring. She has been in the machine learning space since 2014. Evidently is not her first startup. Before Evidently, Elena co-founded an industrial AI startup, so Elena has a lot of experience with startups. (2:01)
Alexey: It's not the first time we speak with Elena, actually. Elena is a returning star. In February, she gave a presentation about “How your machine learning project will fail” which is a great talk and I recommend it to everyone. Check it out – it's on our YouTube channel. Before this week, it was the most popular talk on our channel. Now it's second because of Santiago’s talk on Monday. But this is a great talk nonetheless, and I do recommend checking it out. Thanks for joining us for the second time.
Elena: Thanks. I'm super glad to join. (3:19)
Alexey: Before we go into our main topic of building machine learning startups, let's talk a bit about your background. Can you tell us about your career journey so far? (3:22)
Elena: In short, it has been either about startups, machine learning, or both, pretty much for the last nine years. Originally, I started in the startup scene in Berlin when I joined a startup that was doing some ecommerce search aggregation. This is how I was first introduced to both topics, because they also were using some machine learning. It was pretty basic things back then, but still. Then I joined Yandex, which is the largest search engine in Russia and I was running a startup accelerator there. That is one of the things that I'm still super proud of, because it was a very rewarding experience to help other people launch their companies. You meet so many people who are active and doing things. It's actually very, very inspiring. (3:33)
Elena: So I got inspired too and I joined an internal startup with Yandex, which was called Yandex Data Factory. The idea was, since Yandex did so much machine learning internally, we could apply it elsewhere. So we were working with different industries, and I was doing business development mainly.
Elena: Then I left to co-found the startup where we worked with manufacturing. Companies like steelmaking, aluminum production, oil and gas – we helped them apply machine learning. I was holding a product role, basically as a founder that does many other things, because that's what you do when you do a startup. Right now we're building a new startup that works with open source model monitoring. You can check it out and try it, it's completely free and open source, which is probably another topic that we will discuss – why we applied this strategy to it.
Elena: That said, I want to give a bit of a disclaimer – if you try to start a startup, you'll get a lot of advice, whether it’s solicited or not, so please take everything I say with a grain of salt. No one is an expert in your business, but you can still learn from someone else’s mistakes and thoughts. So just adjust to the experience when you listen to someone.
Alexey: So including this podcast? (5:09)
Elena: Including this podcast. Don't take it at face value. (5:10)
Alexey: Do you know why people start their own business? What motivates them? Why not just work for a company? (5:18)
Elena: Well, I think it really depends – in different countries, there are different circumstances. It's also important to distinguish between starting a traditional business and building a startup. For example, you can open a restaurant or teach a course – something that has a known business model. Or you can go into startups, where you launch some tech company and try to solve something really new – it's a very different approach to business. When I talk about traditional business – for example, for immigrants, it could be like a survival strategy. You come to the country and start a business because it’s the only way for you to build up wealth. Some people want to get rich – I would actually advise against that. If you want to get rich, there may be other schemes than building a startup. (5:28)
Elena: Some people want to genuinely solve a problem that they see. Some people just enjoy creating things from scratch – that's probably more about me, because I'm a type of person who likes to start up things from the very beginning. It can be inside a company or it can be elsewhere, but it's all about doing something from scratch.
Alexey: So getting rich is not what startups are about? (6:22)
Elena: Well, you can succeed, but the chances are pretty low. There are more guaranteed paths to do that, I would say. Working for big tech is probably much more lucrative. (6:25)
Alexey: I think I read an article somewhere where it said “If you take the expected value – how much on average you would earn working in a big tech company versus creating a startup – startups would give you smaller expected returns.” (6:32)
Elena: But many people enjoy the journey. (6:54)
Alexey: Yeah. I can imagine that when you make a startup, like you said in your previous startup and I guess in your current one, you took a product role. This meant you ended up doing many, many different things. (6:56)
Elena: It's not just the product part, but basically everything that you need to get yourself to the next level, right? In different stages in the company that might mean different things. (7:12)
Alexey: Let's say I am motivated to start a startup. I know some machine learning, so I want to build a machine learning startup. Where can I get ideas for that? (7:23)
Elena: Well, first of all – don't. Don't start a machine learning startup. What I mean by that is – when you frame it like this, you actually decide on a solution before the problem. You already know the technology that you're going to use instead of looking for the best way to solve a problem. Here's a trick. You might know some machine learning, but ideally, if you can avoid it – don't try to build a machine learning startup. Try to find a problem that you can genuinely solve. A random example – say you want to help grocery stores get rid of ‘out of stock’ situations. You think, “Oh, I'm going to build better forecasting models for them.” But then when you come and talk to them, you might learn that it's not the forecasting that’s the problem. Instead they maybe don't even have the data about what they have in the store. (7:34)
Elena: The solution for them would be to actually build some sort of data collection maybe through something else. If you solve this problem, you don't even need machine learning. You can create value with just this. But if you come into it with the mindset, “No, I only want to do modeling.” You might pass this opportunity by, right? So it's better if you come with an open mind, if you can. That said, I know many people who will get stuck on a technology and it becomes a “hammer in search of a nail” situation. Some people succeed with this approach, but if you can, it would probably be best to try and avoid it.
Alexey: So if all I want to do is machine learning, then maybe a startup is not the best way of doing this? There's Kaggle and other websites, right? (8:42)
Elena: Yeah. I think in general, if you come to a startup, your role will change so much. Therefore you shouldn’t come into it with the mindset of “Hey, I want to do this and this is the only thing I know.” You can take this approach if you're an employee in a company with a predefined role and understand what exists in the market. With a startup, you'll be constantly doing very different things. (8:51)
Elena: That said, I think there are some ways you can search for ideas, whether it might be related to machine learning or not. One would be actually to look at your own problems or even better, the problems of your company. Because when you are inside some business, you can observe some things like inefficiencies, come up with an idea, and say “Hey, let's test it. Is it big enough of a problem? Are people actually experiencing pain with this problem? Do they want to pay to solve it?” and these sorts of things.
Elena: Specifically for machine learning, I would definitely advise trying to team up with someone whose main expertise is in an area that you're interested in. Just as an example, this can be people who are into insurance, finance, healthcare, or whatever it is. I think the best way to create value in machine learning right now is to apply it to other vertical domains. But for that, you would typically need someone who can explain the fundamentals to you. You don't want to be reinventing the wheel and coming up with algorithms and making things that are obvious, like “Hey, I can do this great stuff!” But it turns out to have been done a million times before.
Elena: I would also suggest maybe to follow other startups. Because if you don't yet understand principles like “What is a startup?” “What is valuable on the market?” “What can become a company?” You first need to just develop this sense of what's going on by following what others are doing by following investment news, what type of companies have been started, and so on. This can give you an idea of what’s going on in the market. But don't try to search for an original idea. I think this is a caveat that many people fall into. It’s like a trap. “Hey, someone else is already doing this. No. I should come up with something that has absolutely never existed before.” Typically, these become things that are not really needed. More often than not, they just try to come up with some shiny thing that no one will ever use.
Alexey: So basically, to come up with an idea, there are some ways to go about it that are better than others. You can try to find some problems in your own life and see if there is something easy that you can do to fix them. Or you team up with somebody who has a background in insurance or some other domain that you’re interested in and they can tell you about the problems they have. Then, together, you can think about the solution to these problems. So you would need a non-technical co-founder who is a domain expert in some particular area. (10:40)
Elena: I wouldn't say it's necessarily these are the best way, rather just some of the ways how you can find better problems than just by searching for it yourself. Ideally, if this person is as involved as they should be, they should become a co-founder. But you can also start a startup with people you know, and you don't have to be business people. Actually, I think a lot of companies are started by technical founders. If you're ready to learn and pick up some things and do some sales, probably – that is not comfortable for many technical people. This doesn’t mean that it will necessarily be this way. You can still learn everything you need to succeed. (11:12)
Alexey: Yeah, let's talk a bit about finding a co-founder. Let's say I am a technical person. I know some machine learning. I know how to code. I'm not good at sales – I don't enjoy selling things. I want to find somebody who might enjoy doing these things and maybe also be the main expert in some area. How do I go about doing this? Who do I need to start with and how do I find these people? (11:44)
Elena: I think there is no silver bullet for this. This question is almost like “How do I find my best friends?” There’s really no predefined recipe to do it. One thing I can say is that, ideally, you should have worked with this person before. In teams that have just met, it's very difficult to check how compatible you are. Being co-founders is a lot more than just being colleagues. When you’re colleagues, you can maybe be okay with most things – you don't have to agree about everything, especially on the values level. But when it comes to being co-founders, if you're successful, you have to be together for 5-10 years, working day and night. Therefore you should really click with this person and understand each other. It's a very personal thing. (12:16)
Elena: There are people who can work together successfully and people who can’t, and you should really check for this kind of personal compatibility. Also consider your motivations. Sometimes people say, “Hey, I'm gonna join you in this startup, but only if you pay me this money.” Or something like “Only if it's successful in the first year.” But a startup is a really hard journey, so you should check each other’s expectations beforehand.
Elena: In terms of how exactly to find these people – I think it’s worth looking at your past colleagues and people that you've worked with – maybe as partners, maybe as clients, maybe you went to school with them – that is usually the better way. However, I've also heard how people just sometimes meet each other at hackathons or other events. Usually, they still work together a bit, right? Maybe not at a startup, but for some projects or some events. This way you can get a feeling of how well you communicate with person and in general, how well you get along with them.
Alexey: So maybe you can do a hackathon project together and then you see how well you get along? (13:45)
Elena: Maybe. Yeah, but a hackathon is a bit of an artificial thing. I think ideally you should have at least one situation when you had some adversity – where something went wrong. (13:50)
Alexey: Like a conflict? (14:04)
Elena: A conflict, right. Or just a difficult situation. It will be helpful to see how you deal with that, not just some one-time project where no one really cares about what's going on. (14:05)
Alexey: I also heard about this, especially in Berlin, there are meetups like that. I think they're called ‘co-founder dating meetups’. What do you think about these events? (14:13)
Elena: I think it can work. You’re still expanding your network, right? Naturally. This is another way to expand your network with the people who are active and interested. I think joining different communities is a great way to meet people. But it often just happens organically, from your first work or something else. But whether you need these meetups? It's a cool community conference, anything can happen. (14:23)
Alexey: But I think you said that it’s best if you work with somebody. (14:44)
Elena: Work as in work on something – not necessarily in the same company. One thing I can add to this is that I think there is a thing called ‘Founder Market Fit’. When you understand that you are the right founders to tackle a specific problem. This is something that you should consider, but it’s just one example. Sometimes, there are companies that go after the enterprise market, which is very sales heavy. You go on, you try to sell a very expensive thing, and you sign a very long contract. Their sales is one of their core capabilities. (14:54)
Elena: Then there are startups where you can be creating a tool for developers and your main capability is actually being able to understand these developers – communicating in their language, so to speak. There might be equally great ideas and equally great market opportunities, but you as a founder of a team might be better suited to tackle this or that idea. This is something that people sometimes don't take into account. But I would definitely recommend checking for this.
Elena: Also you need to choose your market. Because, again, if you're successful, you want to be in this market, following these trends, talking to these people – for years to come. Even if the technology might look similar, you're doing machine learning. In my previous startup, where we were working with manufacturing – I was talking to heads of manufacturing companies. Now we are working with data scientists. It's a very, very different market in how you communicate with them. You should be comfortable with what area you choose.
Alexey: How do I choose? How do I know if I like manufacturing, for example, if I’ve never worked with it? (16:18)
Elena: Ideally, you shouldn’t go to a market that you don't know anything about. (16:22)
Alexey: Or we should probably have a co-founder who has experience in the market, right? (16:25)
Elena: Yes, exactly. Or find someone who can help you get that. (16:29)
Alexey: So what are the things that I should consider? Let's say there is an idea. There is a co-founder. We seem to like each other – we go along well with each other. (16:35)
Elena: At least you try to. (16:51)
Alexey: Yes. Which things should we consider before we actually start a startup? (16:55)
Elena: I suggest that you should agree at least on a few things between you. It makes sense to really spell it out for each other. Sometimes people expect different things, but you don't understand it until you start talking to one another. So be honest with yourself – why are you doing what you’re doing and how long are you planning to be in this game? I've heard about this problem multiple times. (17:04)
Elena: In general, co-founder breakups are probably one of the most popular things for why early stage startups fail – when people really have different expectations. They think, “Well, I'm gonna be here for years!” and then go to work for Google. Another person says, “Hey, I'm going to really put everything into this startup. I’ll work days and nights for it.” So you should be honest in this communication. Why you're doing what you’re doing and how committed you are. It's better to know these things early on, if you have very different views on what's going on.
Elena: Then I think you should decide what type of company you want to build. Here, if you're new to this whole startup thing, a very important thing to understand is that startups are a very special breed of companies. If you want to raise venture capital, you should build a fundable venture startup – something that goes after a really big market. Something that can become a unicorn (1 billion dollar valuation). Something that can grow really fast. At the same time, it doesn't necessarily need to be profitable in the early days. It can bring you money later on. The core idea is that you want to build this company that grows really fast and has a chance of becoming something really, really big.
Elena: At the same time, you can take a different route. For example, you can just build a service company where you create projects, according to clients’ needs – whatever they want, like custom projects – you build them, you get revenue, you get profits, and spend them on your own things. This is not the venture way. You cannot raise venture capital with that. Sometimes people just don't figure these things out in the early days. Especially if they are not accustomed to the way venture capital works. Many things there are really counterintuitive. Like not having profits – profit is a bad thing. It means that you're not spending money on growth, right?
Elena: You should really understand and learn a little bit about this and decide if this is for you, because it really is a very special thing. You will have to go raise money, talk to investors, pitch yourself, and so on. Maybe this type of work is not necessarily for you. Maybe all you want is to have a relatively small growing business. That’s okay, too. You just need to really understand which one you want.
Alexey: How do I understand that? (19:20)
Elena: Well, you should really think about this venture capital startup concept. Understand the benefits and the problems and then decide whether you want to try it or not. If you don't, you’re probably going after a more classic business approach, right? Then you can maybe take out a loan, you can invest in it yourself, and you can build it up. That's a great thing. Actually, you can be better off financially by doing this type of business – compared to a failed startup that tried to go to the Moon and didn’t. (19:26)
Alexey: So it's about how ambitious you are? If you're very ambitious, if you want to conquer the world, then you go into venture capitals. If you want to just have one client and start developing a solution for them silently, then you just bootstrap yourself. (19:51)
Elena: Well, bootstrapping can still be done for venture capital startups in the early days. Bootstrap is more about not raising capital in the beginning and trying to maybe reinvest the money that you get from some clients or just trying to live off your savings, or do some consulting on the side. It's just the way of how you basically live to a point when you're going to raise money. Sometimes you want to do this because you want to delay it or maybe you want to just keep your options open. Because once you get investors’ money, you get on this train and you have to continue raising money and growing the business. If you’re maybe still undecided, you can keep both options open. (20:09)
Alexey: So the caveat here is if investors give you money, your startup no longer belongs to only you. Is that right? There's investor money at stake and you basically need to do what they tell you sometimes, right? (20:40)
Elena: In the early days, they still cannot tell you exactly what to do. It's more of an expectation of how it’s going to happen. In the later stages, they're probably going to have more than half of your company. Generally the idea is that these investors put money into companies that have this moonshot potential, with the expectation that they're going to get 10x return at least. They put money in multiple companies and they know that 9 out of 10 will fail. They want to find the one that will not. You have to continue trying to give them the belief that this shot is viable and continue growing as fast as you can and pouring money into growth. It's a very peculiar thing. It's counter-intuitive if you don't see how this actually happens. (20:55)
Alexey: Anything else we should consider before starting? (21:34)
Elena: Particularly for the machine learning startups, I think you have a choice of whether you want to do the vertical startup approach, like I mentioned – when you're solving some specific need of some specific market. Or whether you want to build tools and infrastructure, which for technical people is often the first thing they think about doing, “Hey, I want to build this library.” Or “I want to make this tool better.” So there is this choice of going vertical or going after the infrastructure market, the latter of which is getting really crowded. We are both in this community. You see how many startups are popping up every day? (24:37)
Alexey: If you think about the ML ops space, there are many companies that are trying to solve machine learning operations problems. Yeah, it does feel crowded sometimes. So this would be an example – if we take one of these companies – that would be an example of a tool startup? A company that develops a tool for engineers. While in the other case, I think you said vertical solutions? (22:09)
Elena: I can give you a specific example. There's this company called Tractable that I think is based in the UK. Recently, they became a unicorn. What they do is help evaluate the damage for insurance purposes based on images. For example, you broke a car, you take a picture, and the machine learning based algorithm automatically judges how much the insurance payout should be. This is an example of a vertical startup, but it's also a somewhat boring problem – insurance and damage estimate. This is a vertical thing where you probably need someone who understands how insurance works. (22:36)
Alexey: I heard this term called “AI First Startup.” What does it mean? (23:10)
Elena: It pretty much means “machine learning should be at the core of the technology”. But the thing is that sometimes people misuse it, just because it's a cool thing to say. So even if you have just a little bit of a linear regression in your startup – in one feature – you may say, “Hey, we're AI First.” You should really look what comes first. (23:16)
Alexey: Okay, so it's a way to attract more money from investors, because investors want to invest into AI? Is that right? (23:37)
Elena: Well, I think good investors look into their investments, but it's also a sort of ‘hook’ for the media. Media write more when it sounds cool – “AI First!” Also, it’s sometimes for the buyers, because some companies would be interested in buying something novel and innovative. When you say that you're AI First, you’re probably more novel than others. But there is also another angle to that. Many startups start from the machine learning part, and they eventually grow into much more than just machine learning algorithms or some system that was there initially. They start solving a growing diversity of problems, start collecting data, actually add the interface, and all the workflows. They become more of an analytical startup that does many things, AI included. So it’ll be called “AI First” but at that point, I don't know how ‘first’ it is. It's just a kind of marketing choice whether you decide to advertise yourself that way. (23:43)
Alexey: Speaking of tools, let’s go in the other direction. I heard an expression from somebody that goes something like “Building tools for engineers is not always the best idea because engineers like building things themselves.” So they don't always like going and buying something – they'd rather build it themselves. So if you create something for engineers, how do you even overcome this? Is it something that you also need to think about? (24:33)
Elena: You should think about this regardless of the market you choose. How are you going to sell to your buyers? Is there any purchasing intent there? But you can’t really say, “Hey, you shouldn’t do developer tools.” Look at all these companies that are doing just that, like Datadog or New Relic, which are doing monitoring. People are paying for this – GitHub and GitLab, they're doing tools for developers and everyone is using them. Like Jenkins. There's a whole plethora of things that people are still paying for, right? (25:06)
Alexey: Yeah, actually, Datadog is a good example. This was actually part of the context when I heard this phrase. I was talking to a couple of people about Datadog specifically and one person said that engineers like open source more – they would rather have Prometheus, Grafana, and things like that – their own infrastructure – rather than just paying some external company like Datadog for collecting the metrics. So what they’re saying is that it's a picky market. (25:35)
Elena: But there's a piece for everyone on this market. No? All those companies still exist. There's space for both approaches? Both still exist and are actually growing. (26:02)
Alexey: Yes. Datadog is doing pretty well, I believe. (26:08)
Elena: Grafana too, even though it’s open source. It's actually one that generates money. I happen to have Grafana Enterprise. Some people are certainly paying for it. (26:11)
Alexey: What kind of skills do I need to start a startup? (26:21)
Elena: I think you need to be a ‘self-starter’ and to actually want to start a startup – everything else can be learned. You should just be comfortable with learning things constantly and be comfortable with uncertainty. Be ready for a lot of rejections as well – be more of an optimist, you know? If you are risk-averse, you want to be sure of everything, and have everything guaranteed, startups are probably not for you. Skillswise, I think everything can be learned. There's no perfect startup founder that knows everything. If you're ready to learn and put some time into it, anyone can become a good founder. (26:25)
Alexey: So a ‘self-starter’ is somebody who starts themselves? Or what does this term mean? (26:59)
Elena: By ‘self-starter’ I mean that you're ready to take responsibility, initiate things, and motivate yourself. You don't wait for someone to tell you what to do or explain how to do something. You do it your own way. You make a hypothesis, find the right people, and talk to them. So you’re doing things with an internal motivation as opposed to waiting for someone to explain to you what to do. (27:05)
Alexey: So you just count on your own intuition that the thing you're doing is the right one. Is that right? (27:25)
Elena: Well, I don't think that you need to always count on your intuition. But you can say something like, “Hey, I have this idea. Now let me check how I can verify it. So I'm gonna do this, I'm going to do that. I'm going to talk to these people. I'm going to talk to these potential customers.” But you basically need to decide what to do on your own. You come up with a hypothesis and how to check it. (27:32)
Alexey: So let's say I work at a company and the main manager gives me well specified tasks. I just take these tasks and do them. Then maybe doing a startup is not the best thing for me. Is that right? (27:52)
Elena: Maybe not yet. I think that people are not necessarily static like that. At some point in your career, you might feel differently about the whole thing. (28:08)
Alexey: What are the risks? I think we discussed a bit about how working in a big tech company is better than doing a startup in terms of how much money you will earn or the ‘expected value’. What are some other risks of building your own startup? You probably need to live on your own savings for some time and there is no guarantee that this money will come back. Are there any other risks? (28:17)
Elena: Well, actually, I don’t think you necessarily have to live on your money. There are other ways to go about it. Some people start alongside their day job. Then at some point, they become ready, roll over, and try to raise capital. Generally, the financial risks are huge just because you're most likely going to lose money with a startup. It's almost guaranteed, even if you pick the business up later on. So it's not the best financial decision in most cases. Generally speaking, you’re highly likely to fail. It's almost guaranteed statistically that you're going to fail as a startup, so you should prepare for this possible outcome. (28:49)
Elena: I think the interesting part here is also the cultural acceptance. Depending on where you are in the world, people can treat startup failure very differently. For example, if you're in the US, having a failed startup is something to celebrate. I may be overstating it a little bit, but it's perceived as ‘okay’ – it's a learning opportunity for you. In other places in the world, it can be seen as something that’s damaging to your future career. If you later want to try to go and continue your professional career after doing a startup, this is something that you should take into account.
Elena: Work-life balance can also be tough. I must say that doing startups is often a bit of a luxury. Not everyone can afford it in terms of their personal commitments and capacity to take risks. It's not always possible in many places of the world. You might have a family to take care of, so you cannot risk financially by doing a startup. So there are these things related to failure. I would not mention any career prospects or losing your knowledge, because I think you learn so much that you can later maybe find new career opportunities if you decide to go back to a day job.
Alexey: That’s interesting. That's one of the things I always think about. Let's say I'm an engineer. I'm a data scientist and I start a startup. Most likely, I will end up doing other things in addition to that. Or maybe I will stop doing data science altogether and focus on other things. One way could be that in five years, I will forget how to train a logistic regression or something like that. What you're saying is that it's totally fine because the skills I would pick up while building a startup would allow me to change my career later. Is that right? (30:22)
Elena: Well, I think that maybe for your technical skill set at least, you might become a bit rusty, unless you really invest time to support it. But you will also have more new career paths opening up to you. Maybe it will be more of a managerial role or maybe it will be a more product-focused role. Because after you have tried to build your own startup, you look at things not from just a viewpoint of how a little technical detail is implemented, but rather you look at the bigger picture. What you're doing, for whom, and why it is needed. (30:56)
Elena: The sense of ownership and understanding the big picture is very valuable, especially when your join and work in another startup. Maybe one that is a bit more successful than yours was -- if you left your startup. To lead some divisions to grow faster. I know many founders who were struggling to hire people that have a startup mindset. Usually someone who tried to do one is the best candidate for this sort of hire.
Alexey: You mentioned that when starting a startup, we should be prepared to fail. How can you prepare yourself for something like that? Is it like in Buddhism, you just see that the thing as already failed? Then if it actually does fail, you don't worry about it? How do you go about doing this? (31:50)
Elena: Oh, I don't know about the specific mental techniques. But I think you should understand that this possibility exists and kind of normalize it for yourself. You shouldn’t associate yourself personally with your startup and say, “If this startup fails, I'm a waste. My life ends.” This is a very dangerous mindset. Instead you should think, “Hey, this is a thing that I'm trying. I know that it might not work. Even if it doesn't work, these are the things that I’m going to learn and this is how it’s going to help me.” Of course, you need to stay optimistic. Because why else are you doing this? I think just normalizing it is very important, so that it doesn't come off as a huge shock if it happens. Hopefully it doesn’t. (32:12)
Alexey: Another thing you mentioned is that the work-life balance suffers when you do a startup. Is it possible to make a successful startup while working, let's say, 40 hours per week? (32:47)
Elena: I don't know of such an example. Maybe they exist. Maybe I’m the one who’s bad at it and the people that I meet are not so good at it either. But I think when you're passionate about something, it doesn't always feel like it’s work, right? Sometimes my partner even jokes, “There's weekend work!” which is the work you like to do. You choose to do it on your weekend, because you enjoy doing it. But you’re still doing something for your startup. (32:58)
Alexey: Okay, so you said you haven't seen an example? (33:24)
Elena: I personally don't know of people who are living a very relaxed work-life balance and doing a startup at the same time. Out of those who have traditional business, I know a lot of people that are. Maybe also those who are at later stages of a startup, when the company is a little bit more mature and they already figured things out. But if I look at everyone who is at the early stages, whom I personally know – maybe these are just the people in my circle – it's always erring on the side of overworking a little bit. But I think it’s by choice. You know that every incremental hour is gonna bring you value, right? It's not like in a big company when nothing will happen even if you don't touch something for a month. (33:28)
Alexey: We already started getting questions. You already mentioned that when you start a startup, you don't necessarily have to bootstrap yourself – meaning that you don’t have to live off of your savings. You can do a startup while still working. So this would be like a part-time startup. What do you think about this? Is it a good approach? Because you need to work 40 hours and then you have this part-time startup that eats up the rest of your week. Is it a good idea to do this? (34:06)
Elena: It's always very individual. It might be that your workplace is more calm – things are more relaxed and sorted out. If you're in a high growth startup yourself, probably you cannot do this. In general, of course, it's better if you can dedicate yourself full time to the startup. But it's not always feasible. And especially if you're a technical person – you're mostly likely the one who can create the minimal viable product. Meaning that you provide the code yourself and then test the demand. So this is something that you can do or try to do. It can be something like a ‘weekend project’. (34:39)
Elena: There are surprising examples of startups where it's not the code part [that is the most important]. It's not a huge, complex system. It's more about figuring out the demand, what is needed, and then building maybe a small thing that solves this demand. Then you can see a huge list of people who sign up for it, “Hey! Looks like something. Maybe I should go and talk to investors and try to raise money for it.” I think actually in Germany, you have some really nice ways to support younger startups, especially for those coming out of university. There are some programs where you can apply and you will get some money as a grant – for free – so it's not taking equity. There are also some other opportunities like this, especially in Europe.
Alexey: But there are also startup accelerators that would take a part of your company. (35:47)
Elena: Well, they're usually already investors. So you often need to show something to get into these accelerators. But yeah, this is like an early investor. It can also be an angel investor. (35:51)
Alexey: I know a couple of people from such startup accelerators. I think the accelerator, the company, gives them some sort of stipend – some sort of money, a monthly allowance – so they do not have to starve to death while doing the startup. So they pay them some money while they're looking for an idea, looking for co-founders, etc.. But then, I think they take like 30% of the company. (36:01)
Elena: I think this is not a classical model. The classic model would be to just act as an early investor – to take equity or have other ways of how they take a portion of your company. In this case, you can spend this money however you like – as a startup. But you need to pass certain criteria to do so. There are these other/newer ways, such as Entrepreneur First. There are some other organizations like this that try to invite individual people and then form teams inside. But yes, they usually take a bit more in equity. I think all this is possible. In the end, when you have a successful startup, everything that has happened in the early days is not that important. Most startups make mistakes, do something wrong with their equity, and so on. It's more about whether they take 1% of a billion or 1% of nothing. (36:24)
Alexey: Okay, so maybe it's not a bad thing if the startup accelerator owns 30% of the company? (37:10)
Elena: 30% is a bit high to be honest. I think it's usually less. Often a startup accelerator will take like 5-7% – something more reasonable. (37:16)
Alexey: I was actually talking about this Entrepreneur First company, I just remembered the name. They did reach out to me also a couple of times. This is how they work – they just write on LinkedIn to different people saying, “Hey, you seem to have great skills and your technical expertise will be beneficial. So join us. These are the conditions.” Then probably some people agree and that’ll be the startup. (37:28)
Elena: Entrepreneur First is definitely a good example, simply because they still exist, right? Also, they don't do it for free. They eventually make money because some of the companies become successful. So it should work theoretically. (37:56)
Alexey: We talked about a situation where “I'm a technical person, and I want to start a startup, this is what I do.” What if I'm not super-technical? I also don't have any resources to develop something, what do I do? (38:08)
Elena: In this case, building a machine learning startup is probably going to be hard, but building other startups is still possible. There are currently so many no-code tools and you can actually hack an MVP using something like an AirTable and Bubble. There are Webflow and all these ways how you can create a website or some simple mobile app to test some idea. Then if you see that there’s demand for it, you can already attract the technical co-founder or even some investments. It doesn’t work for all types of startups, obviously. But if you want to build a biotech startup, for example, you also probably shouldn't go there without having biotech expertise. So your options are a little bit limited, but they still exist. There are startups, especially in SaaS, (software as a service) that being technical is not the most important thing. It's not some specific engine inside that makes it different from others, right? It's more the idea. Maybe you automate accounting for some companies. What you do in the first days – you are the one who is doing the accounting for them through the same interface that you later want to automate. You just need to prove that it can work and that people are ready to pay for it. (38:22)
Alexey: I've heard this term “prodictionizing your services”. I think this is what you're referring to, or “productizing” – making your services a product. You can do accounting, for example. So you start doing accounting for clients. Then you start wrapping it up in a nice interface. You then let your customers or clients actually use this interface. Maybe at the beginning, you do the accounting yourself under the hood, but then you can automate it. Is that the approach? (39:25)
Elena: If it can be fully automated, yeah, that's an awesome startup. The challenge is that sometimes you can’t. So you have a very custom thing and then it's very difficult to scale. Because you still need someone who understands accounting to deal with each individual client's issues. Unless you're really standardized, it’s hard to make it work. (39:57)
Alexey: Another question we have is, “I have found a problem in my domain. I have learning statistical modeling and machine learning. Do I need to find a tech expert or a domain expert to help me?” (40:13)
Elena: It depends. I don't know the details about the exact case, but I would suggest that if you already have someone who is ready to work with you, just try and learn from this first customer. You should always favor action when you're doing a startup, because you're going to learn from it. And you will learn something new when talking to the client and possibly discover “What exactly is needed?” Very often, in a startup, you change your idea or it evolves – it's almost normal that it evolves. If you look at many successful startups, they often started as something else. I would not overthink it, though. I would not overplan stuff. I would always just try to go and talk to these potential users and learn something from them. If you cannot get to these users easily, probably that's not your startup, because you will need to continue generating ways of how to reach them. (40:33)
Alexey: Yeah, so basically, you must favor action. When you favor action over inaction and overthinking… (41:23)
Elena: Like a plan, like, a super amazing strategy, and [hope] it will be then fulfilled. [Instead] Go talk to people, hear what they say and try to do this. (41:29)
Alexey: So you just sometimes need to ‘just do it,’ right? Ok, you found the problems. If you found the problems, there is somebody who has these problems routinely – go to that person and talk to them. (41:39)
Elena: Yes, but just do the talking first. Don't do the coding without talking to people. Don't misunderstand me in this way. Because this is often a thing – especially with technical founders – you want to do this perfect service, do the perfect code without talking to your users. That sounds awesome, right? Doing a very shiny thing. But you might be doing the wrong thing, which is why you should talk to people first. (41:52)
Alexey: So how many people should I talk to before starting to code? (42:15)
Elena: I think until you stop hearing new stuff. If you try to find people who are similar in some ways, you can say, “Hey, this or this. These roles in this industry, those might be my potential buyers – accountants (for example). So I'm going to talk to five accountants and ask them questions about their problems.” And if you still keep hearing new things – each of them is saying different things – you probably should continue talking to people. If you talk to maybe ten and you understand that six of them were saying exactly the same thing, it's probably a reasonable enough number. (42:19)
Alexey: I think there is an article from Atlassian or some other company that says “You don't need to interview many people.” If you interview three or five of them and they have the same problem that you're solving, then that should be enough. (42:49)
Elena: I think the problem is how you define whether people are the same. Because they might come from companies of different sizes, or they might have a little bit different roles. A parts of the problem is actually figuring this out. Are you talking to data scientists, machine learning engineers, data science managers? In small companies? In big companies? In tech companies? In non-tech companies? In specific industries? These are all sub-segments, which might have different needs. If we’re talking about the data science world as an example. (43:03)
Alexey: How many people did you talk to before starting Evidently? (43:26)
Elena: Well, we did go really overboard with that. In total, I think before starting to work, we talked to probably around 50 people. And then over a hundred during the early days of development. (43:30)
Alexey: So you met your co-founder even before talking to these people? Is that right? (43:41)
Elena: Well, Emeli and I have known each other for over seven years. We worked together in Yandex and then we did the previous startup together. So we are a team that has known each other for a long time. (43:46)
Alexey: So your previous startup was also with Emeli? (43:57)
Elena: Yes. (43:58)
Alexey: Okay. I'm just interested to hear the story. You worked together and then at some point, you decided to create a startup? How did this happen? (43:59)
Elena: Well, talking about Evidently – after working with so many applications that get created for different companies, and most recently in the manufacturing world, we saw that there are a lot of problems that are not related to the core technology, but rather to its adoption. Also the problems with what happens after we put models in production. I know if you are hanging out in this community, that may sound trivial – that many people are already solving these types of problems. But in the beginning of last year, it was not that obvious for many people. If you talk to more traditional companies, you will also learn that they actually have these same problems – how your monitor your models work, how you understand what's going on with them, are they working or not, and other caveats. (44:12)
Elena: Based on our past experience we thought that it's an interesting idea. We actually had several different ideas. We went to talk to people in order to validate them and monitoring was the one that we saw a lot of response for. People were quoting us, “The model’s broken and people don't notice!” “It's annoying and no one wants to do this!” “The data scientists leave the project and monitoring doesn't happen - the models are left uncontrolled.” So we heard this and said, “Hey, we validated this idea out of something we have. Let’s work on this.” So we went on to build Evidently and we’re super happy with how we're doing. We have a great team. This is an important part of working together – having a relationship.
Alexey: So it’s just the two of you, right? (45:28)
Elena: Yes and one person that is part time. (45:30)
Alexey: How did you do it? You talked to these 50 people to actually start doing this startup? What did you do after this? (45:35)
Elena: Well, you know, talking to people is something that’s always happening. You talk to a few people, you process the feedback. In the process of talking to them, we already made some mock-ups, for example. We tried to show “Hey, is this what you expect? Would you like to look at this or that?” It kind of happens throughout the process. When we began working with Evidently, it was the beginning of the lockdown. The world felt a bit crazy. At the same time, it was very easy to talk to everyone on Zoom, because everyone was just sitting at home. We were just talking to a lot to people and then switching over to coding and things like that. In our case, we are complementary in our skillset. For example, I was focusing on initiating contact with these people – calling them up, inviting people to chat – and Emeli was doing the prototyping at first, and then the actual coding when we proceeded to development. (45:45)
Alexey: So what do you do at Evidently? (46:32)
Elena: Everything. [laughs] (46:34)
Alexey: I think you're the non-technical co-founder, right? (46:35)
Elena: Yeah, I’m non-technical. Well, I'm our designer for blogs, too. So the role really evolves over time in your company. When we were just starting out, it was talking to users and trying to process the feedback, organizing it, coming up with ideas. Then we started working on it, and now do a lot of content-related work. I'm writing all these blogs that help us talk to users and generate interest in what we're doing. I’m talking to investors and potential investors, and building up all the parts beyond the code, pretty much. How you set up the company, where you incorporate, how you organize it all – CEO is not just a title. It isn't a certain role that you do, right? It's “chief everything officer” which is not code in our specific example. But I will say it really depends on what startup you are doing. It can all be different for you. If the startup was focused on direct enterprise sales, for example, I would be doing the sales. This is what I did in my past startup. Now we are dealing more with evangelizing and talking to users about open source. Because it's free to use, we are more interested in people trying it out and giving us feedback on it. That's what I'm trying to do in my most recent days. (46:39)
Alexey: So you talk to people and write content. You're also a designer to some extent, right? (47:49)
Elena: The designer aspect is part of writing content. But the job is that you have to do whatever is needed and whatever you can. In our case, we can create this high quality content, so we’re focusing on this. But it could have been something else, just as easily. (47:58)
Alexey: We have a question from Anonymous. And I also have this question. “Why do you choose the open source model for evidently?” (48:11)
Elena: It can actually be very successful to go to market with this strategy for many infrastructure startups nowadays. It can also just be a rational choice if you do infrastructure. You can see companies like Grafana – they actually monetize their open source business, but they create a lot more value by creating this open source tool as well. For us, it also resonated a lot because it seems like a very natural and good thing to do, because it creates a lot of value for the users. So what you do is basically monetize only some parts of them, which would be for the large enterprises that are ready to pay for what you do. (48:19)
Elena: Another aspect of that is that you actually get very fast iterations and feedback loops because of it. For example, if you just create a huge, monolithic, closed source thing, you first need to create it and deploy it on some customer and then see if they like it or not. With open source, you can release small features. It can even be not yet fully functional, but people are already trying it and they say something to you about it. This helps you to iterate a lot faster. So I think it's a very nice way to build up your product, when it’s public like this. It’s also a good market strategy for later on, because open source doesn't mean that you don't make money out of it. It's just a different strategy.
Alexey: There is a concern I often hear from some people. They say, “Why would I open source my code? People will just go and copy my code and deploy it. So this strategy doesn't work.” This is the concern they have. Is it a valid concern? (49:29)
Elena: Well, if we cannot create anything else beyond this code and we don't plan to, then probably. But if you will continue building the product and innovating and building the team, you will be the one who owns all the knowledge as well as the future vision of the roadmap. Then you can create and maintain the open core products. That's how it's often done. You have the open source part and you have some functionality that is part of a paid enterprise offering. You, as a creator of the project, know how it actually, truly works, right? So you can be successful in maintaining it. You also own the community. Companies who want to pay for the product usually want to pay the original developer not someone else who just copied it. (49:50)
Elena: That said, there is one thing related to cloud providers that has recently been a big thing in the open source world. Another way you can monetize machine learning that's open source is by creating a cloud version of your product – literally the same product, but you do all the hosting and scaling and provisioning. So basically, all the DevOps parts behind it. And people literally pay just for you to run this project and product for them. For example, if someone like Amazon comes and takes the same product, and hosts it...
Alexey: Like ElasticSearch, right? (50:57)
Elena: Exactly. So they have to introduce a license that explicitly forbids just this part. But still, now it's more of a collaborative relationship with the big cloud providers, because they also don't want to be seen as the bad guys killing someone else's open source. If you do the open core approach, you're not hindered by this. (50:58)
Alexey: Okay, so it's not a bad idea to develop open source? (51:18)
Elena: I don't think so. But I feel like many people are a bit too protective over everything they do. Sometimes they’re even like, “No, I'm not going to tell you my idea.” It doesn't really work like this. It's still a lot of work behind just what's out there to make it turn into a company. There are so many great examples like Mongo, Elastic, Grafana, GitLab – they're all open source, right? And they're all huge. (51:22)
Alexey: I guess in case of open source, what can happen is that the engineers and data scientists find your library, start using it, and then it reaches the management. The management sees it and then you sell the company your enterprise offer, right? So you don't need to go through the management, and kind of force the engineers to use your solution. Because the engineers are already using it, right? (51:48)
Elena: Ideally, yes. This concept even has a name – it’s called “bottom-up adoption”. You come in from the bottom and go up in the company. The truth is, big enterprises want to pay for a good product, because if you're running something in production, you rely on it. So you want to make sure that someone is responsible for how well it performs. If you're an enterprise, you’re ready to pay for security, for safety, for the peace of mind that this thing performs really well. (52:14)
Elena: It's a very natural way – adopting it from the bottom up – compared to when a CTO just signs a big deal with someone like Oracle, and they bring it to the developers and say “Here, you better use this thing.” And the developers don’t like it. They don't want to learn the thing that is maybe not transferable in terms of skills to their next place of work. Whereas open source is something that's going to continue to be out there. So you are essentially increasing your own personal value when you work with open source. It's a very efficient model. Now that enterprises are used to running open source products in production, this business model has become possible.
Alexey: We have an interesting question. Let's say you decide to create a startup or consultancy – how do you deal with clients or customers that do not want to share their data with you? (53:09)
Elena: Persuade them. [laughs] In all seriousness, you have to constantly overcome so many barriers when you're selling something, right? It's just the initial part of offering something. You must provide enough value for people to do something or to work with you. This ‘doing something’ might even be just installing the product or running it – it's still work that is needed from their side – or to adopt it into their workflow, or learning how to use it. You should just present the value as so high that people will overcome this barrier. In the end, it's all about finding this value. This is kind of what makes a startup a startup. (53:27)
Alexey: Have you seen any difference or do you know if there is any difference when you build a startup in different countries? For example, Germany, Russia, or the United States? (53:57)
Elena: Yeah, absolutely. It's really, really different, especially if you build a startup for local markets. When it comes to venture funded startups, it's all about your exit opportunity. Are you going to do an IPO and be traded on NASDAQ? Or are you going to be sold to a large company that's going to pay hundreds of millions of dollars for your startup? These opportunities are different in different markets. The reality of the startup scene is a very different one. The attitudes of investors also vary. For example, in Europe people tend to be a bit more risk averse – the valuations for startups are a little bit lower. In the US, it’s a little bit different, especially in California, where there are so many startups popping up and there's a lot of money available to them. The way you build them and how much money you can get really depends specifically on the market that you’re addressing. You can be based in Germany, but you can sell worldwide. But if you're doing some accounting software that’s specifically focused on the German market, you might eventually have 100% of this market, but it's going to be very different. This probably deserves a whole separate discussion, but yeah, there is a difference depending on where you're based. If you're raising capital from investors, it matters how it looks in terms of startup scene and the capital available. (54:06)
Alexey: Going to the previous question about sharing data, I think in Europe, people are more hesitant to share their data. While maybe in Russia, people say, “Okay, here's my data. Take it. Just make sure it doesn't leave the country.” (55:17)
Elena: You know, I think it's actually a bit easier in the US. Also because people are more used to using cloud providers in general, so the data on your sales will be in Salesforce, right? In many other places, people still want to keep everything on-premises. But I must say that almost any startup that works with B2B, it will come to data sharing. Even if it's just sales software where you put your sales numbers – you're still putting in your sales numbers, it's a big deal to overcome this aversion. But you can see that many startups made it, so it's possible when you provide enough value. Open source is actually another workaround to get around this. To use an open source tool in the early days, you don't have to send data. It makes it so much easier to try the tool, as compared to sending the data somewhere in the cloud. (55:34)
Alexey: I think when Emely did a presentation a while ago at DataTalks.Club, one of the questions was “Hey, I'm a bit concerned about my data going to Russia.” What you answered was “Hey, it's open source. You don't have to be concerned. You just take this thing and run on your hardware – on your machine.” Is that a correct assessment? (56:17)
Elena: Yeah, it’s open source. You can run it anywhere. I mean, you can still organize it in a nice way. If you send it to the cloud, you can host it in the right location. All the cloud providers allow this. It still requires you to pass through some safety checks in many, many cases. It's doable, although it may be hard. If you're selling something, you will have to object against many things like this. Just paying money is also a hard thing for companies to do. You have to convince them that you're worth it. (56:38)
Alexey: At what point do you start thinking about hiring engineers? Let’s say, we can maybe think about Evidently, or maybe there's some other startups. There are two co-founders – at what point should they hire somebody? (57:06)
Elena: Well, I think A) when you can B) when you need to. The thing is, sometimes you shouldn’t overhire and build a huge team before you have the so-called ‘product market fit’ – you don't even know if the thing that you're building is needed. I think it's best if it's only the founders or a very, very small team working with them, because one day you can just up and say “Hey, we're gonna throw away everything we did. Now let's redo it. Let's change the course completely.” So it's better done in a very small team and you don't want to spend a lot of money on it. But once you understand, “Okay, this is the thing that we're building. There's a demand for it. Now we are moving more to the scaling phase, or to building up more features that the users want, or to being able to serve more users.” That’s when you should hire. But you should be guided by the actual business needs, not because, “Hey, I have a feature list of 100 ideas. I need to hire as many engineers as I can to build all this”. You should be more focused on “Is there anyone who really needs what I’m building?” (57:23)
Alexey: You mentioned at the beginning that you need to “follow startups”. You need to know what others are doing, see what kind of investments they're getting, are they pivoting, what kind of clients do they have? Is there a specific way for people to follow other startups? (58:14)
Elena: I think you can join so many communities to accomplish this. Once you become part of it, this information just starts pouring. You follow some people on Twitter, you’re in some Slacks, you read TechCrunch. I wouldn't suggest just going crazy and trying to keep up with everything that's happening – you just won’t be able to. But it's just helpful, especially if you were not exposed to this before, to look at some investment news, and maybe try to decode, “Hey, what are they doing? Why were they attractive to investors?” Just learn and kind of reverse engineer their success a little bit. (58:36)
Elena: Also keep in mind that what's written on TechCrunch might not necessarily be 100% reality – it's a bit of a vision of what the startup is doing. It's really helpful to get this general understanding just by looking at others. Maybe you can keep up with the news related to your domain so that you know who the big competitors are. You'll definitely be asked about it.
Alexey: Okay, thank you. I think we should be wrapping up. Do you have any last words? (59:28)
Elena: Well, I think that if you want to build a startup, it's a very good thing to try to do. Ideally, in a nicer way – so you don't have to risk all your money or something like that. These days, there’s a good opportunity to do that. There is a lot of capital available, there are programs that support you. But I also suggest being honest about your intentions. Don't just follow the hype and think “Hey, I'm going to be giving talks everywhere because I'm a startup founder!” and enjoy this social part of it. Because it's actually a very hard thing to do. You shouldn’t go for it just for fun or just for money. You should be genuinely interested in what’s you’re doing. (59:32)
Alexey: Where can people find you? (1:00:09)
Elena: LinkedIn, Twitter. (1:00:11)
Alexey: Thank you so much for joining us today and for sharing all your experience with us. Thanks everyone for joining us as well and asking questions. That’s all. Have a great weekend. (1:00:18)
Elena: Thanks everyone. Bye-bye. (1:00:31)
Subscribe to our weekly newsletter and join our Slack.
We'll keep you informed about our events, articles, courses, and everything else happening in the Club.