AI for Soccer: How Bundesliga and AWS Are Revolutionizing the Game with Real-Time Data

Matthias Patzak:
Welcome to the Executive Insights podcast, brought to you by AWS. My name is Matthias Patzak. I'm an enterprise strategist with AWS. Today with me is Hendrik Weber, Senior Vice President, Sports-Technology Innovation at DFL, Germany's professional football league. Hendrik, thanks for joining us today.

Hendrik Weber:
Thanks for having me.

Matthias Patzak:
Hendrik, could you give us an introduction in what is the DFL and what is your role at DFL?

Hendrik Weber:
Sure. So DFL stands for Deutsche Futball Liga, which is like you said in your intro, the governing body of German elite soccer or let's call it football. So, we organize the competition itself and we commercialize the rights of our clubs. And it's two divisions, like I said, 18 clubs per league. So it's like an NBA, NHL, but just for German soccer. And just to give you some numbers, we make 1.6 billion euro per year on revenue, which we then distribute to the clubs, which is a significant portion of their revenue. So that's the DFL. And in terms of sports-technology innovation, I'm personally responsible for everything related to technology, data and analytics. More on the sports side, but I'm sure we'll get to that a little bit in more details later.

Matthias Patzak:
Could you explain more about what is “sports-technology innovation”? So, what is the data technology innovation part at DFL?

Hendrik Weber:
In terms of sports data strategy, I mean we realize that it's a core ingredient for our media product as well as an important source for our clubs. So, I'm sure you know there's a lot of data being collected during a match. And this has been organized from a league perspective, because it does not make sense that every club on its own collects its data. So, we'll take care of that. And so we distribute this to the club so they can use it for their purposes on the sports side or on the media side. The innovation part is basically that we want to be ahead of the curve and be really innovative, because we as a league, I think we are and we try to stay a pretty innovative league, and this is sometimes it always has to do with technology and data.

Matthias Patzak:
I have to confess, I'm not a soccer fan, so I can't really imagine what type of data do you really collect during a game. So, for sure the number of goals and maybe the number of fouls, but what type of data do you really collect during such a match?

Hendrik Weber:
So you are one of the few Germans who are not interested in football.

Matthias Patzak:
It looks like.

Hendrik Weber:
But fair. But as I said, so if you now focus on the sports data side, so thousands of events are being collected mostly still manually. It's still people who actually observe the game and kind of annotate the action on the pitch. And then where it's getting really into more, the big data game is what we call tracking or positional data. So, we track the ball and the players a lot of moment per second. So, it's like 3.6 million data points we collect from each match. And we have 617 matches a year.

So, it's being actually a fixed camera installation in every stadium. So, it's computer vision technology where the position is being detected. And actually, from next season onwards, we will go into the next generation of that kind of tracking, which we call limb tracking or skeleton tracking. So then factor 40. So, it's really a lot of data where every limb and every body part is being collected and there are a lot of use cases for that. But this is the data collection side on what we call official match data, and we call it that's the single source of truth, which is then part of, as I said, of our media product in which we distribute to lots of multiple different consumers.

Matthias Patzak:
And so, you collect data and is it real time or is it you collect it and then you process it and after a week you sell it or?

Hendrik Weber:
Yeah, you rightly said. So, I think what makes it really special on the sports side that the live content is the valuable one. If we call sell our rights, obviously the live content is really king and key.

Matthias Patzak:
So real time content.

Hendrik Weber:
Yeah, so it's absolutely real time. So, you are absolutely correct. The real value of rights on the sports right is really in the live content, because we compete against obviously other sports but also other entertainment products. So, you either maybe watch Netflix series or you watch Bundesliga or NFL, you hardly can do all of them at the same time. So, the live content where you don't know what actually happens in the next moment, and this is maybe the game changer and something the whole world will talk about, that makes it really the valuable piece. And so, it applies to data as well. So, since we want to use the data and add it to and enrich maybe the media, the experience of every fan, the better we connect real low latency live data with the footage, the better it is.

Matthias Patzak:
And the live data is then used by media outlets and networks or even also by the trainer on the field?

Hendrik Weber:
Both.

Matthias Patzak:
Oh really?

Hendrik Weber:
Yeah, absolutely. I mean, there are analysts sitting on the bench sometimes even on the Press Tribune. So, they're actually interconnected with a headset. They talk to each other, and they have on the iPad the live screen of the footage, they have the live data under the second screen. They annotate it, maybe even on their own live, and then they send information down to the coaches or it's being used in a halftime break to intervene maybe about certain tactical patterns. So, on the sports side it's being used.

By the way, maybe another example is that the data is being imported and via the data you can clip the video quicker. So, you get a little snippet ready to go for the halftime intervention by the coach. But since really, it's about minutes, you need to be quick and you need the data to automate the process. On the media side, there is the TV product program where you use live TV graphics or the commentator. I mean we love to talk about more about that because there are actually a couple of interesting projects we do together with AWS in terms of how to support the commentator, who actually comments live the game, what kind of data he can use for his commentary to make it more intriguing and interesting.

Matthias Patzak:
As a technologist, I find this very interesting and as this podcast is about data and data strategy, I would be interested in what is data strategy for you?

Hendrik Weber:
Let me start maybe what is data strategy not for me because I think it's not really starting, I would call it “bottom up” in terms of what is the data model, what is the infrastructure, how do we host the data, how do we process? It's more a top-down approach where I think it must be heavily linked with a business strategy. There's also another aspect which is more linked to the fan data which is a totally different collection process.

But it's basically the idea as a league, obviously you have more classical B2B business model where you basically sell your rights to a broadcaster or an intermediary. And actually, they have the direct contact and not us. But that's what we really now got to get into it that we think, “Who is actually the Bundesliga fan? What are his/her preferences, what does it like and not like?” To use this information to make or tailor our product better, to make our clubs better on the sports side, so we have a more attractive league, and a more attractive competition. And the more competitive and attractive it is, the more it's easy to commercialize.

Matthias Patzak:
For many organizations, I observe for them a data strategy is a plan to set up technology and architecture and software. When do you know that your data strategy is successful?

Hendrik Weber:
The challenging piece in the sports industry I think, and especially in our case is there are right cycles and they have 2, 3, 4 years. So, you sell the rights and then the period is for four years, and then you do things and only after four years if you sell them again, the value increased or decreased. So that's not like if you sell, I don't know, cars or books and you have a weekly report and it does go up and down. So, the telling piece is really how to detect what works and what does not work. And obviously we should like AWS I guess be customer-centric and really have fan surveys and really ask the Bundesliga fan, “Do you actually enjoy the program?” Do you enjoy if we integrate more data content pieces instead of relevance, for which age group, for which generation and which channel and which distribution way? And then react whatever the feedback is.

Matthias Patzak:
Interesting. Can you walk us through DFL's journey becoming a data-driven organization and leveraging data and data strategy?

Hendrik Weber:
It was 2010, 11, 12 when we realized that sports data is really a strategic asset. Because at that time a lot of providers collect data. They were not consistent, not a lot of high quality. We did not have enough processes in place to deal with that. So, we really started from scratch in terms of a data model, a definition catalog, how do we define things, how do we operationalize, then set up a data management hub, how to distribute the data. And then we actually created a company called Sportec Solutions, which is a joint venture but it's a subsidiary of the DFL group where all the sports data activities are being bundled in one corporation.

And it was linked with the overall idea, we call it “glass to glass” strategy. So, from the lens, the glass of the camera lenses where from the camera where you actually capture the action on the pitch, until the glass of an iPad or TV screen, how you actually consume the Bundesliga. So, everything in between is the value chain, which we believe we should at least think, “Where are the crucial and strategic points and where do we need to own them or at least to control them?” And for us then data became more and more a strategic ingredient and it's within this process. We had internalized resources, built our own software systems and really became more and more mature in that field.

Matthias Patzak:
It really sounds like a journey with a lot of learnings. Many organizations I observe, for them, a data strategy they follow this plan, build, run approach. So, they make a plan for a data infrastructure, then they build it by themselves or they outsource it, and then they hope that the data strategy deliver value. But it sounds like it was a lot of experimentation involved in your journey. So how did you optimize the probability of success of your data strategy?

Hendrik Weber:
I think one of the key things is to manage those experiments, or let's call them POCs, are big enough but not too big so they have the right scoping so you really can learn from them. But it does not jeopardize too much everything, because as I mentioned earlier, we are in an industry which is rather, I would call risk averse, you can experiment, but really big errors are really a problem.

And in football and sports where especially in football where it's so much cultural inherited and so the fans, they don't want to have too many experiments. If you try something it should rather work. So, it's to find the right steps, step by step, and then as you said, learn from them and then adjust and then redo it. So, it has been lots of small steps sometimes one step back and then hopefully one and a half steps in advance.

Matthias Patzak:
In the German community, you were one of the very early adapters of generative AI. Can you share a use case, how your data strategy helped you to adopt generative AI in your real time or in your fan-based data business?

Hendrik Weber:
I think a couple of nice use cases are already being POC'd and implemented. So, for example is, like I mentioned earlier, we want to provide relevant data insights, for example, to the commentators. So, we call it's like an AI life ticker or AI data finder tool, which we, together with AWS, built. So basically, we have all this live data coming in, but obviously it's lots of data and if in a live scenario you do want to have time to scroll down the numbers and find the right pieces. So really to have automation there that what are the relevant pieces, and then to prepare them for the commentators so they just can use it in a live commentary. So, this is something which we already use now for a year or so.

And the next thing I think really interesting we're working on is that this is really only data itself, but if you then add video or audio or more content with it, then it gets really intriguing. So, then it's about the guys who take the photos basically and then they have the agency, so they take the pictures and then they're being somehow uploaded somewhere and then later post-match you can pick the photos. But that's far too long.

So, as I said earlier, it's about zero latency. The idea is that the AI finds the most fitting picture, which is related to the stats he already collected. Then maybe crops the picture, then because it's cropped, it's maybe not as of quality. So, you upscale it, so that it's a really nice picture and then even connect it with a graphic. So, it's like an infographic, which then instantly you can send out, distribute to your own channels, social media or even using in the TV graphic. So, there's an idea where those processes are being automated and smartly merged and then at the end really is unique content, which especially for I think younger audiences is of real value.

Matthias Patzak:
I really wonder, so you talk about innovation, it looks like you have a fast pace of innovation at the same time you say the fans expect very high quality. So how do you balance this high quality and data integrity and the fast pace of experimentation?

Hendrik Weber:
Yeah, that's a balance, classical I guess in any data projects where you have the human in the loop process there, which we use as well. So, for example, a project we collected we jointly did with you guys is what we call Bundesliga Match Facts. So really advanced stats based on even tracking data on a live scenario, but there are always edge cases, things which sometimes happen, like a player who even maybe missing data, maybe the data can be false sometimes, even though we quality-insured, but it can happen. And those are being covered then with a human-in-the-loop process, which then obviously we look at them and then check how to adjust the model or adjust the algorithm.

Because as I said, the official match data is the single source of truth. And for the public, that's the written bible, basically. If we say it's, I don't know, 10K distance covered, it's 10K distance covered and nobody... So, we really need to make sure that it's of highest quality.

Matthias Patzak:
You are an AWS customer for a long period of time and you're not just using our services, but we really have a partnership on ways of working on how to innovate. Could you share a bit with the listeners to the podcast how this partnership helped the DFL to innovate?

Hendrik Weber:
A couple of things which I think we really learned from you guys was really the customer centricity approach. So really be alert, and to really listen to the fan, and maybe listen to the upcoming fan, or the younger generation, and then to transform your organization, to transform your business model. I think this is really something we learned from you guys and we try to adapt.

And then as I said earlier, this experimenting approach, because I already mentioned it's a rather risk-averse environment. It's sports, it's very emotional, especially football. It has been there for century and it is successful because it never really changed. It's good because it did not change. That's a little bit the notion, but we see this differently because we think we need to be innovative, we need to transform, we need to adjust to the next generation, to the really very competitive landscape. We need to develop our product further, and how to do this and experimenting and fail, but then you stand up again and do something else. This is really an approach which we adapt and is I think crucial to our success.

Matthias Patzak:
As a final question, what would be your advice to your peers and other senior leaders on how to build a strong data foundation and a culture of experimentation based on this data foundation?

Hendrik Weber:
I think it's really important to start with a business strategy. Not to start too technical, even though the tech is important and so on, but really think, “Okay, what is our core goal of our organization and how do we actually generate revenue? And what is our role and how does the data help us?” So really this is the starting point and should lead us.

And then I think the second thing is before tech for me is really people. You need to have a culture that enables that approach in terms of experimenting and be open to that. I think you really need a top-down buy-in from senior management that this is really the approach and people know this is being supported, so you can actually fail. If you understand why and to react in the right way, then it's okay to fail. Sometimes it's easy to say, but an organization, sometimes people, it hinders innovation sometimes. So, if you really come up with this culture and have the right people there, then the third thing for me is then to get the right technology and get it done, and which serves the purpose of you and those three pieces all together. I think you are in a pretty good spot.

Matthias Patzak:
Hendrik, thank you very much for joining me on this podcast. It was really a pleasure. Thank you.

Hendrik Weber:
Thank you.

Listen now

Listen now

Listen now