S02 E08 Data-Driven Fitness: An Inside Look into Urban Sports Club's Innovative Data Platform with Artur Yatsenko, Head of Data Platform at Urban Sports Club

Welcome to the Modern Data Show- a podcast where we explore the latest trends and technologies in data and analytics with some of the brightest minds in the industry. For today's episode, we have with us Artur Yatsenko Head of Data Platform at Urban Sports Club which offers corporate fitness programs including over 50 sports to the network of more than 10 000 partners across Europe. Artur has more than seven years of experience in data and analytics in e-commerce and digital industries working for companies like Zalando and Delivery Hero where he not just helped companies in scaling their data teams but also help in setting up a robust product analytics and AB testing strategy across the whole organization. Welcome to the show, Artur.

Thank you very much, Aayush. Glad to be here!

So Artur let's start with the questions let's start with the very basic question. Tell us a little bit more about Urban Sports Club and your role over there

Yeah absolutely. So Urban Sports Club could be considered the scale-up at this point. The company itself had been on the market for already 10 years. And our main goal is essentially we are a sports platform at this point and what we do is connect people as well with the partners that we offer on our platform that allow them to do sports in various locations and a lot of different diverse sports. An example is that we allow a lot of flexibility for the customers to essentially exercise and our goal is to bring that sport and that mentality as well to our consumers first & foremost.

Yeah. And how would you describe the business as it's a mix of a digital business as well as the actual user experiences is into a physical activity or is it in a physical space? So how would you label the business? Is it a digital-first business or is it a business that is digitally enabled?

It's a good question. I think it's a tech first. I mean focusing is mostly that as a tech company right? And the functionality that we built essentially allows our users to lead a more healthy lifestyle. The digital and sort of more mixed model is something that we also looking forward to moving into as well. I think in the beginning, well not the beginning but in the middle of 2021 that was the point when we started to come up with digital content as well during the pandemic when we realized that there should also be a mix, right. And a lot of gyms were closed so people couldn't do any sports and workouts. And that was the idea that we tried to come up with digital content and I think the strategy that we are looking forward to is to have a diverse offer. The digital content, as well as the physical, and then say the concept of community as well as the users, can collaborate and explore that. So we're trying to look into the mix of both digital and the physical because you cannot really in my opinion completely substitute the physical self-engagement when you actually go to the gym or do any self sports activity with your friends or colleagues as well.

Got it and tell us a little bit more about your role Artur you're head of data platform at Urban Sports Club. Tell us what is this role about and tell us a little bit about your team.

Yeah absolutely. I'm heading the data platform team and that's also something that we focus on in terms of building a new distributed data platform. We have been as I said already quite some time on the market as well, and we are also in the phase of transforming our data stack and not only the data stack as well as the technology stack as well. And the data platform team is specifically focusing on building the new data warehouse that would allow us to use a lot of the different modern data technologies to help different teams. So this is a pure data engineering team that deals with how we can bring the data from a lot of different entities because the Urban's Sports Club is not the only business entity as such as well. So it's about cross-collaboration with other departments and also the backbone of the data that could be used by the analytical teams as well. So we have robust reporting as well. But apart from that it's moving into like more of the data products and how we can build things that could benefit the business users as well. So the data platform vision we have in this team right now also encompasses a lot of different components to it. Not only is the data warehouse such from a technology standpoint of view but also what could it bring in terms of the new data products that could be useful to the business. And we have quite an interesting ecosystem within Urban Sports club.

Yeah and I would love to dive deeper into your stack and how it has evolved. But before I do that I have a basic question. What was the point where like every other company data as a function always starts with something very basic, you would build a bunch of pipelines on your own and you would display our stuff on the basic dashboard or maybe an excel sheet? Where was the time or what was that stage where you guys decided that data is one of the core departments we're gonna invest in, in terms of building a data platform or in terms of building a data team? What was that moment for you?

Yeah I think it's an excellent question. We realized that data was mostly treated as a by-product but not as a product. And I like the statement and what we're trying also to do right now is to say that data is a product on its own. And when you think about even a dashboard or anything that is being naturalized from that bunch of chaos, it has to be treated as a product and it has to go through iterations. It has to go through a proper understanding whether the use cases for the people who wanna use that as well. So I think for us it took some time to transform that understanding that it's not just the numbers to look into but it's showing the additional value that this data could be used for unlocking near real streaming and insights as well. Like working on some machine learning models as well and that was also the point when we got more buy-in from the management and we all as an organization decided that we need to put more effort into building the system or a platform as such which will support us in a multifaceted way.

Yeah And as you've said, often in a lot of cases data and data teams are often considered to be a cost centre rather than a core product itself. So any example or anything that you can share from your experience at Urban Sports Club or at your previous roles with the companies that you have been involved in where the value that was delivered from data was something that fundamentally changed the business? Is there any experience that you would like to share?

Yeah. I think I can try with more general ones and then I can come up with some more specific good examples. I used to work in mostly product analytic teams and with product analytics as the function as well you can see more you can trace back the results of your work to actual implementations how it's gonna impact the customer. But that's why I like that's results-oriented and also here with an urban sports club for instance we have a product analytics team we have also other teams that give a lot of these recommendations that are later being implemented. But I'm a really big advocate of experimentation as such so the whole AB testing concept as well. And we try to also empower the teams that would have the possibility to think outside of the box, outside of the delivery process as well. We recently have introduced getting stuff done days. So today in a way hackathon is when people can just stop operational work and start thinking about some other project that they want to do and with that there's a point when the data analysts come together with the software engineers can think about the customer problem and can start implementing that as well and can see the effects of that solution essentially on the end business users. So that's mostly how I think it is best to showcase the value of the data by showing how the stats impact the behaviour of the customer down the road. It could be anything it could be a simple functionality say we try to also act on a lot of the user feedback as well try to prototype faster. But that's the way when the data could play a vital role when you solve a customer problem as the first step.

Yeah. And let's dive deeper into your data platform. What does your data platform look like right now like broadly in terms of the architecture the moving components and how it has evolved from one of the earliest versions that you had for the data platform?

Yeah it's a really good question. So I think we find ourselves right now in the process when we are in migration as I said we had Urban's Sports Club has been on the market for some time as well. And so also led to some decisions in the architecture that had been there for some time and have to be upgraded to the modern data stack. So we migrated from the PostgreSQL data warehouse which is not turned out to be flexible for an analytical data warehouse to the Google cloud platform at this point. And I think that's also what we try to apply most of that like modern components like Modern Data Stack essentially what we are trying to do is to make it more distributed as the platform so that we don't have of course like one team that's managing the whole pipelines and the engineering core function. What I also tried to do is like for those people who are the experts in their field let's say, talk about the data analysts who are familiar with analytics engineering as an example. So they could also contribute to the data platform and that the data platform is distributed. And within that migration, we are looking into a couple of different components as well. So we try to break down to let's say what is the ingestion part. like how we stream the data first. And then we are looking into a mix of the also third-party vendors as well as building our own. But to prioritize the speed of the delivery, we're looking into ingestion tools as such from different platforms. And the transformation that we are doing right now within our data warehouse. So not the extract transform load but already the transformation part inside we're doing with dbt which has proven to be like quite useful as well for the analytical use case. Because it doesn't require a lot of data engineering knowledge as such more analytical engineering exposure to it. So that's how we try to democratize parts of that as well. So it's not only let's say you have a data engineer that always takes care of the request of adding something as well but we want to create the main teams. They're able to be responsible for the data that they produce, data that they transform and getting the value out of that themselves as well.

Yeah and as a Head of the Data Platform I'm sure you are also responsible for procuring these technologies that assist as a part of your whole data platform in building the data platform. What are the top things that go into your consideration while procuring these tools? You talked about ingestion technologies you have tons of commercial ETL tools available. Tons of open sources available. Walk us through how that decision starts in terms of either first trying to build this thing on your own or procuring a third-party vendor if you're procuring a third-party vendor what goes in your mind?

Yeah I think it's really good to think about that. I recently saw also a study that I think is from Boston consulting group that the majority of the data professionals feel overwhelmed about the market and the offer that is given by the data tools and that's challenging to decide, right? Even thinking about the data injection part right? I mean our goal is really to start maybe prototyping something fast and test start if it's gonna work instead of building our own because that takes a lot of time to get that done and often we don't have that many resources. The way that we look into this is to try to pinpoint the use case and do a bit of the prototyping testing let's say with a couple of different tools. Let's say take an example for ingestion components. So we normally test, I think, well with two, or three providers that we know for a limited period then see if there are any issues with loading the data or not. Can we have a resilient infrastructure for that? Can we have a good response time as well? And if that all fails we would try to build it on our own. But generally it's really complicated and I think I've been struggling as well coming up with the framework that is allowing you to understand what the requirements for a tool and how do you measure that. There are also a lot of things that come into play, the support as well is quite important, the price of the tooling as well but ultimately I think if that solves the use case and if it does allow you to have less energy and effort of your engineers spending on maintaining and implementing that, I think this is the one that you should always go with because ultimately you would spend a lot more time into trying to fix, try to find the workarounds and everything else and it would be that hidden cost that you don't put into in the beginning when you do this evaluation of the tooling and the vendor stack.

So does that mean, do you guys try to avoid open source?

No, actually we adopt open source as well. Openly we use dbt as an open source as well. But some things that are available of open source do not often fully maybe the level of support or something or specific, maybe let's say connectors, right? And that sometimes is the challenge. If you're also interested, you can check out our Tech Radar. We have published what we adopted in terms of the technologies and the frameworks as well so that might be interesting and then we experiment with open source as well. I think for us most important is that try to do if there's any Security Threat Analysis in case we want to be more compliant as well. We are working with Urban Sports Club is also a marketplace from different sides as well. We also work on the corporate end. So you wanna make sure if the data could be treated in this case compliantly as well. So we try to analyze if there's anything that could be potentially red flags as well before we decide on the software itself.

Okay. Brilliant. And speak a little bit more about the data platform itself. Is it mostly batch processing or what real-time streaming components that you have currently as a part of your data product right now or what you have planned in the near future?

Yeah, so at this point, mostly analytical as an analytical data warehouse. I mean that mostly is the batch, right? We've been seeing adoption historically as well. And then we try to continue as the basis of our operations as well. But looking into the real-time is something that we started also doing. And we have major plans forward too. One of the reasons is that we have our essentially microservices being rebuilt and with the microservices, we try to build our stack as well that can be more independent way and stream the events-driven infrastructure towards that as well. So while the tech stack is being evolving as well. What we as data platform are doing is trying to read a lot of those events from these microservices as well for a lot of the analytical purposes, all as well, enriching some of this like data. So essentially we get all of this information in one place.

Was there any specific trigger or reason why you started incorporating these real-time processing capabilities? Was there a kind of a product requirement or was it a very specific business function requirement that kind of maybe even think about this?

Yeah so, I think for us it was quite important to give more autonomy to the teams like we've been operating like technically as well a bit of the more monolithic architecture. And then at some point, we've said we need to decouple that as well and give more ownership to the teams that they would have. So the reason for going event-driven is essentially to give them more autonomy to the teams for managing the services as well as expanding that to Microservices that the team could have end-to-end responsibility. And as the data platform we decided mean there is a cloud event that we are sending data through. We are using Pub/Sub up for that reason as well. So why don't we actually, stream the information on our platforms as well? And this is just like a stepping stone because for us is also building more use cases based on that data, based on their services that could be and should be processed in real-time for a lot of other business use cases.

Got it. And just curious what was, what was the biggest challenge that you faced in terms of implementing this entire data as a product strategy? Both from a maybe from technical as well as from a socio-technical perspective, what was the biggest challenge that you faced?

Implementing the data as the strategy generally, you mean?

Data as or maybe data product as a strategy.

Yeah, I think the biggest challenge is also understanding again and well convincing people that the data is not the byproduct but is a product right? First for some of the people it still could be considered as this is just the reporting, this is just the numbers, but it's really hard to put like into perspective, like a lot of things that are happening behind the scenes, right? And even to provide you, let's say the correct number of the count of your daily active users and as an example, there are a lot of things that are happening in the background from the data quality perspective a lot of different things. But I think for us that was one of the reasons that like one of the challenges and still showing the value of that data and how much work is being put into that is, is something we can help the stakeholders, especially business stakeholders who really might be quite detached from the tech context. That sort of helps to win that over right.

And, but it's really hard to put any quantifiable metric to a data team, right? For example, for sales team, you would've revenue KPIs like your revenue, your bookings, and so on. But for the data team, it's very hard to put a quantifiable metric. How do you justify the ROI of investing in data products or data teams per se internally at Urban Sports Club?

Yeah, so we also try to adopt the more engineering metric approach for the data platform. So there are things that we can still measure, right? There are things that we can measure like days without incidents as an example or the

You also talked, one thing is one of the, one of the biggest driver for you in terms of building the data platform is to be able to, democratize the access to data to various teams and departments. And one of the kind of counter argument to that kind of approaches is how do you deal with then data governance, especially considering the, these decentralized approach to your data platform? And how do you mensure that the data quality is maintained across all teams and departments. You just said, one of your KPIs is how many incidents happening per week or per month. But if it's, that's decentralized, how is that affecting you and your team?

Yeah, I think what we're trying to do is that, our team would be like more of central guidance such, to provide this and essentially like everybody who are skilled enough, like to do, the data transformations and having their own individual like data teams in the future. More like data engineering teams that should always be able to do that as well. We don't have that at this point. So it's mostly like data professionals who work with the data that's being prepared by the well data engineering team which is like more of a central team. But at the same time, the governance is a good question because I think there should be some balance between true democratization and actual governance. Cuz in the end you also don't want the stakeholders, creating, let's say different definitions of the same metric, right? And I think that's something that we have seen in the past as well within our organisation. We try to change right now as well, how can we provide more metadata, more context, more robustness into the definitions itself. Like having one stop place for everything else as well. So that it doesn't get misinterpreted. And this is one of the examples that we are trying to do with our approach.

Yeah. And now let's talk a little, something a little bit about what happens once your job is done. As a data platform, as a data team I think one of, one of the kind of outputs of this team is delivering those data-driven insights, right? How do you ensure that these data-driven insights are communicated effectively across the business and are integrated into the decision-making process?

Yeah, I think this should be targeted at some point, right? That it could be like really something generic in a way. I should originate from the like requirement in the end, right? If, let's say I have a business person asking us, let's say we wanna understand something more on the partner's side, and we wanna understand maybe which cities are most relevant to you to work in or increasing to, right? This is something more, we have a clear a business requirement in mind that we can, start profiling the data for it. And the result of that is something that you know, like people who work on that the data professionals can control. Cuz in the end they have the context. They are embedded within that setup essentially. They know the business problem that can deliver insight they will be taken into account, right? So I think the end lies within the individual responsibility of the data professionals. So there, we provide the data, right? We like to provide clean data and everything else, but the context and how this will be used should be also back on the analytical professionals as well, should be back on the business professionals as well who can come together into, bringing those insights into action.

And how much of and, I'm sure you wouldn't have a kind of a concrete number to that, but how would you say. The work that you have as a part of the data team is push versus pull. And what I mean to say by that is pull is you have been requested to, let's say, build a new dashboard or build a track, a new set of KPIs or matrix versus. how much of that part of work is, you built something which you feel can add value to various other business functions and where you go to those teams in terms of presenting those capabilities to them. What's the mix between those things?

Yeah, I think it's still a lot of that, we pushed right, like request-wise, but that's something that we want to change in ways like by, we bring more like data literacy to it. I think that, having something fundamental that should allow the self-serve analytics that knows people should be able to make their hypotheses, and insights and test them, right? If this is something more complicated, that requires, the assistance of the data professional, I mean that they should be able still to do the first hypothesis and then come to us for help. So I say that we're still being requested to do some more things a lot, but we also try to change that like allowing the sort of more governed democratization of the self service data to be used so that's, we don't have to deal perhaps with any, minimalistic changes that have to be added. And with the distributed setup as well, we don't often need a data engineer to do those changes to produce a dashboard. The analysts themselves could also be the ones who will create a data product, right? I think that's the approach that we're trying to be taking.

Got it. And now that we are, towards the close of our episode, let me leave you with one last question. In terms of looking ahead, what do you think would be the most significant like trends and challenges in the world of data and analytics? How do you think, things would go from here, especially from a viewpoint of a practitioner? I know that we have a lot of noise in the industry in the past few years, but as a practitioner, Where do you see this going from here?

Yeah, it's an excellent question. I think yeah, I mean for us, perhaps as a practitioner says like how to keep up with the modern thing, how to keep up with modern technologies. How to make sure that we're not obsolete in terms of the things that we use, how to integrate, AI machine learning as well more effectively. I think everybody is talking about chatGPT 4, and things like that as well. Just like one example, you know how to essentially be more proactive, not reactive. Building the next level of analytics and it's not only about serving analytics, but it's more about predictive analytics as prescriptive analytics as well. And that requires a bit more context also like a bit of a different set of skills instead of maybe just building and modelling the data. But thinking about the data as they're creating products that can help also maximize your revenue stream as well or yeah, so make life easier for some of the company or like new functionalities. And I think we have to be open and flexible and try to pick up the new trends and how we can utilize them to our advantage in the industry we're working in.

Amazing. So thank you. Thank you so much for your time, Arthur. It was lovely having you on the show, and I hope our audience had a great time listening to this.

Thank you very much.