Experiments and Robust Products: Data Analytics at BizMachine

They’ve been on BizMachine’s data team almost since the beginning. Petr Kalát and Matěj Maivald tried working with data while still in school. Matěj as a researcher at McKinsey, Petr on O2’s Big Data team. But both mostly ended up processing data in Excel or PowerPoint. They dreamed of real data work. According to them, they only learned that at BizMachine, where they both arrived in 2017. Since then, BizMachine has moved from data punk to greater specialization, including within the data team itself. From solving ad hoc client requests, the company started building its own robust solutions — its own product. But that doesn’t mean work at BizMachine has become boring. Both Petr and Matěj still see plenty of room for experiments and learning. Currently, the data team has its sights set on more dockerization and automation.

What does their work look like? Take a peek behind the scenes of BizMachine’s data team.

You’ve been working on BizMachine’s data team for five years. Where did your passion for data come from?

Matěj: I studied economics and was always drawn to subjects and assignments that involved some light programming. My first job was as a researcher at McKinsey — lots of Excel and PowerPoint. But I wanted to learn more advanced tools. I got that opportunity when a former colleague invited me to interview at BizMachine. Back then, everything here was done in pure SQL, Python, and R scripts.

Petr: I also studied economics, but at my school I didn’t get much exposure to data subjects — statistical modeling at most. My first experience was at O2, where I ended up on the Big Data team by chance. That’s when I started dreaming about learning to work with data properly. But somehow I couldn’t push myself to do it on my own. I joined BizMachine during a period of wild learning. And then there was no way to not be motivated. I’d come home from work and keep studying. And the very next day, I’d manage to use what I’d learned the evening before.

What do you enjoy about working with data?

Petr: I enjoy cracking hard problems. It’s not revolutionary — it’s gradual. First you know nothing about the problem, then you have to explore it, map it from the inside, and finally conquer it. What fascinates me about data is the efficiency and power hidden within it. If you wash the kitchen one day, it’ll be dirty again by evening or the next day if you use it. Building tools on data is like devising a solution you use over and over, but the kitchen stays clean forever. That’s the power of automation and clean design. When you write a program well, it can run continuously and keep creating added value. And that’s fascinating.

Matěj: My first job taught me that prioritizing form over substance just isn’t my cup of tea. I enjoy seeing the results of my efforts and work. I’m not creating a one-time report that gets thrown away after a single use. I can continually work on improving the breadth of company data. You can beautifully see it in our product. You look at a company profile in Prospector and see all the work behind it. Your work simply doesn’t get lost easily. You can keep it in sight whenever you want. And along with you, at least a thousand people daily.

Has it always been like this at BizMachine?

Matěj: Looking back at the early days of BizMachine, our work was very chaotic. Our task was to deliver results as quickly as possible. Data wasn’t really viewed as an investment in the future. It was simply a one-time or time-limited project. But we gradually started to transform. Now we’re building B2B solutions. Everything we do for clients, we actually build so it’s not a one-time script.

Petr: We build our internal development as a generalized client use case. Because we believe that if one company needs to crack a certain question, the solution will find use elsewhere too. We build internal competence, a tool, or a dataset that can then benefit multiple clients. And we can keep developing it and gradually increasing the added value it brings to us and our clients.

‍

‍

“I’ve been on BizMachine’s data team since 2017. I often fix my own legacy solutions and scripts, but at the same time I frequently create something new. I enjoy cracking complex problems. For example, on one of my recent ones — writing a financial statements parser — I had to figure out how to define columns and rows in unstructured data so we could assemble a tabular structure. I verified my ideas in my free time using Lego bricks. Then I studied up on data science, read about
clustering algorithms, and used the one that fit my problem best. When I find the right solution that works
and works perfectly, I feel good. I used the right method in the right place and created a solution I don’t have to be ashamed of.”

Petr Kalát

I see. How does this affect your work? What does your workday look like?

Matěj: Our data team is now essentially split into two parts. One focuses more on fulfilling client requests. Based on our dataset, they find answers to a specific client’s questions. They create insights tailored to a given client (based on their parameters and attributes). The other part of our data team focuses on building solutions that benefit us long-term. They build solid foundations to enable all kinds of analyses on the data. Petr and I belong more to the second group. At the same time, we’ve both been here a long time, and our work has elements of both branches. Our newer colleagues can already choose whether they want to focus more on client requests or data engineering and building more robust solutions.

Petr: Previously, positions on our data team were more universal. Now there’s room for specialization. Matěj and I have it a bit different still. Over the past year, while working on a large internal project, I actually had to jump to a client task several times, solve it, and then get back to our internal tool. I enjoy that variety, but we now also have dedicated data engineers on the team who can focus on a narrower spectrum of things, go deeper, and improve our datasets and Prospector faster. That’s the advantage of being completely shielded from client requests. What’s great is that you can choose what suits you more. Personally, I enjoy the combination and variety.

How would you describe yourselves? What’s your job title and responsibilities?

Petr: I still have some client work, so about 5% of my time I’m a consultant. Sometimes it’s about faster communication with a client, cutting out the middleman; other times it’s about attending a workshop. Then there’s a bit of Data Scientist in me, or perhaps more of an analyst. But you can’t do analysis without prepared data. And that’s Data Engineer work. That’s my whole work mix. I work very efficiently — like much of my colleagues. I can prepare data myself, without waiting for a ticket from the engineering team. I can simply prepare the data, analyze it, find the answer for a client, and then generalize it all so we can use it in developing our own internal product.

Matěj: I have it very similarly. I see myself as a data engineer, but I also have one long-standing client historically. I know in detail how their system works. So I don’t just create robust solutions — I sometimes also experiment and look for quick answers to current questions. Our data team’s work — currently there are eleven of us — also involves using a DevOps ticketing system. We run weekly sprints here. Every Monday we share what each of us is working on and where we’ve progressed. We need to know about each other. And on Thursdays we share what we’ve learned. We share new knowledge, show how and where we’ve used new technologies, and so on.

So you both have a broader scope. You describe your work quite generally — could you give me a recent example?

Petr: The answer is a bit complicated given the scope that both Matěj and I have. We both create something new and aspire to do it really well, while also fixing existing solutions and various one-off scripts. But if you want an example, I recently worked on a financial statements parser. It was a combination of major software engineering and solving smaller sub-problems. For example, I had to figure out how to define columns and rows in unstructured data so we could merge data from different sources. First, you have to explore the problem. You often verify your ideas in your free time, like while playing with your kids and their Lego. You build strategies for how to approach it, and eventually you have to find an ideally at least partially ready solution. We can’t keep reinventing the wheel. For this specific task, I studied up in Data Science — read about clustering algorithms and used the one that fit my problem best. That’s the necessary experimentation phase in client assignments. You want to quickly test that your chosen strategy will work. When you then find the right solution that starts working, it usually works absolutely perfectly. And then you finally feel you’ve written a good solution. That it’s not cobbled together, not nonsense that no one will ever understand or decipher if they had to fix it after you. You simply feel good that you used the right method in the right place.

Those sometimes chaotic-looking one-off assignments from clients actually expand our horizons and define what we could start doing. Through them, we learn that a problem exists and that a solution can be found. And so an entirely new solution is born — not just another feature for an already established platform.

So you essentially learn something new with every project. What specifically have you two learned at BizMachine?

Matěj: I’ve learned almost everything here. Like using countless technologies. At the beginning we had almost none, and now we just keep adding them. I learned here to create quick solutions, to experiment, to verify that the approach I want to use makes sense. I learned to balance between how quickly you can get a given answer and how stable the designed solution can be. It’s an important basis for deciding how much time and energy to devote to developing a solution. Our engineering team then tries to build every solution that deserves it from a business perspective robustly from the start. But I still have the option to do something quickly as a prototype and see results, and only then take the robust path. I really value being able to choose which path to take.

Petr: I also learned everything relevant that I know here. We have the advantage of quick client feedback. Even when we do something internal and robust, at the end it’s a generalized client case. And that’s why I really like the ability to jump between doing something quickly and maybe a bit worse — experimenting — versus doing it slowly and properly. It’s a skill for us, a matter of choice. We choose based on the client brief, the situation. It’s not that we only know robust solutions or only quick experiments. We often do completely new things that no one has tried before us. That’s where you need an experiment. It can’t be automated from the very start. That would take twice as long. We’ve all gradually learned here to change our approach, our way of working, and even our thinking. To move from pure experiments to systematic solutions. But while seeing the advantages of both. Because those sometimes chaotic-looking one-offs from clients expand our horizons and define what we could start doing. Through them, we learn that a problem exists and that a solution can be found. And so an entirely new solution is born, not just another feature for an established platform.

We’re talking about experiments and learning. Is there still something you want to learn? Where do you want to grow?

Petr: Of course there’s always more to learn. Data and our entire field are changing and evolving rapidly. And we have the experience that with every senior person who joins us, we learn new technologies faster — ones we’ve been eyeing for a long time.

Matěj: Thanks to one data engineer, we moved from very rigid SQL to using MongoDB and NoSQL databases. We knew their advantages, but with someone who had prior experience with them, we simply started using them faster.

Petr: We know the technologies we’d like to start using, but when you have an evangelist on the team who knows them and introduces you to their secrets, it brings much faster adoption. And that’s fortunately happening here.

And what’s the technology you want to use more than before?

Matěj: Dockerization. And then automation end-to-end in general. Both are very broad topics that don’t involve just one specific technology, but rather a way of thinking.

Petr: Definitely Docker. But the scope of what we want to learn is incredibly broad. And although we’re gradually working through it, new things and technologies keep arriving quickly. It’s also driven by our field. You can’t know everything.

One last question. You’ve been at BizMachine for a long time. Why? What do you enjoy here?

Petr: We have a great team here. Not just in our data team — we go for beers after work and often chat and spend time together outside of work. The advantage of BizMachine is also our startup nature. In the beginning, it was pretty wild here — rapid learning, speed, constant forward motion. That’s stayed. And what’s also persisted is our flat structure. There aren’t five layers of managers between me and the company founders. One of the founders sits a meter away from me, and when I need to discuss something with him, I turn around, walk over, and we just talk normally about client assignments and their solutions. But we’re also no longer the type of startup where all solutions are built on scripts written in the evenings. Far from it. But we’re not a corporation either. For example, we don’t have predefined, fixed, and unchangeable technologies or a single service provider. If someone comes with a good proposal for how to solve something, and our current solution is worse, nobody here resists gradually switching to something new. The freedom to choose, learn, and experiment is what I enjoy here.

Matěj: I’d add that working with the Martins is great. Each of them has a different opinion and a different perspective. They’re not authoritarian, and debating with them is constructive and demanding at the same time. This also means that the tasks we solve here aren’t easy. But that attracts people who don’t rest on their laurels and are eager to constantly push themselves and our product forward. That’s what I enjoy.

‍

“I enjoy working at BizMachine. Collaborating with our founders is interesting — I can discuss even detailed parts of proposed solutions with them,
on a daily basis. There aren’t 5 layers of managers between us. Plus, our three Martins are inspiring. Each has a different opinion, a different perspective. But they’re not authoritarian, and debating with them is both constructive and demanding. This also means the tasks we solve here aren’t easy. But that attracts people who don’t rest on their laurels and are eager to constantly push themselves and our product forward. That’s exactly why I’m here.”

Matěj Maivald

‍