Mladen Sormaz: Building a data analytics department

MLADEN SORMAZ is Director of Football Analytics at 777 Partners, the multi-club ownership group with a portfolio including Genoa, Standard Liege, Red Star Paris, Vasco de Gama and Melbourne Victory.

Prior to joining 777 in June 2022, Sormaz was Leicester City's first Head of Football Analytics for three years. Speaking at TGG's Big Data Webinar last December, as part of a presentation titled 'Data Analytics Within The Multi-Club Model', he explained how he went about building Leicester City's data department:

3. Action: At what point are you just an extra source of information, and at what point are you actually part of the tipping point that triggers an action? Analytics, to have any purpose, needs to build into strategy and into clear action plans and processes.

4. Synergy: How do we combine expertise in data analytics with expert knowledge from the game so that both become more than just the sum of their parts? In reality, there is information either can teach the other to provide a better view of the questions we are trying to answer.

At Leicester City, we had a variety of data sources, all being used for different reasons. The process (which you can see below) would always be getting the data in a raw format, importing it into some kind of database and then for different practitioners having it either in a cloud bucket, a SQL relational database, or simply in a Dropbox folder.

The final step would always be, 'How do we build in more interactive ways of building insights and doing data analytics?'

Each individual component isn’t as important as it was 20 years ago, because there are so many different solutions, but the general flow of it stays the same.

This conference gave us a lot of different approaches and an idea of how to build around a lot of different people’s processes. We thought about the diverse challenge of how to do data analytics and how to communicate within a club, so that the data guy is not just someone who provides you with answers with numbers.

Your small one or two-person analytics department is likely to have to do a multitude of things.

One of them might be a back-end engineer - the person who creates the data architecture or infrastructure. That’s part of your job when you’re starting a department on your own. You are also the data scientist and analyst, so you have to be able to do some level of data modelling and enrichment.

And then you are also the person doing the front end of that, doing basic UI design and translating those advanced stats into either use-cases that help people or speed up workflows. And, finally, a role you tend to share with technical scouts or performance analysts is that you sometimes do the delivery.

So you have to figure out how all that data that was raw at the start of the process relates to other mediums that coaches and practitioners might deliver in - like video reports in meetings. In companies that have bigger data science staffs, you would have these as individual roles. But when you are setting up for the first time at a club you are expected to wear all of these hats at different times. You feel like a one-man band.It’s important to communicate that, so that people understand.

Expect to do everything, which means you won’t be able to do it to your full potential. Still, you have to find useful ways to add value with your process.

The most important step of all is having some advanced stats and being able to deliver them to either improve performance or to find value in players through recruitment.

This is where all the translation and contact needs to happen between the data scientist and the coach/ scout/ performance analyst. Even if you are at a club where you’re lacking resources, if you focus on this, you can still build a good case for how analytics is making an impact.

If you’re a club that doesn’t have access to a data scientists or the funds to do data science, then a better way to think of the process is project management. This is how you build something useful and the conceptual steps to get there:

1. Step one is have an initial meeting, set aside some quality time, decide on new projects and what the priority order of them is.
   
2. Two is build up a rough spec, often from a performance analyst. Decide how long will it take, what the key components are and how will they all work.
   
3. Create the project - so do the technical work and get feedback for tweaks and changes.
   
4. Allow analysts direct access through different types of software, such as R Shiny. Making it self service is what makes it scalable and automatable and means analysts don’t always have to come through you to get insight.
   
5. The final step is to always ask for ways to improve and iterate consistently.

Something we did for the first team at Leicester was to create a post-game process to turn around analytics quite quickly. Pretty shortly after the final whistle of a match, we would have F7 and F24 Stats Perform files ready for us with all the game stats. We would then run some pre-processing, some visualisation code and manually annotate.

Within 30 minutes to an hour after the final whistle, we would do a WhatsApp delivery of a PDF to the staff that was a basic summary of the match stats.

Two to three hours after, we would have API pulls of the tracking code we got from the Premier League and we would update the whole match round at once, with both the Stats Perform and Second Spectrum data.

Usually the department that needs the data the fastest is sport science, because they have their recovery day the next day to think about.

Twelve hours after the final whistle we would pull the Statsbomb tagged game, quality check it and upload it to the database.

Fourteen to 24 hours after, there would be a longer post-match report delivered to the coaching staff based on the Statsbomb data. We would also run the API and data wrangling code on the whole week of fixtures for all the leagues the recruitment department was using and simultaneously update the tracking database.

It’s a lot of small individual steps, but from the outside all you see is a one-page or a 12-page report!

Another timescale we worked on was the the ideal weekly process supporting the first team, which we developed before I left.

Five days away from a match we would deliver an updated pre-game report for the opposition for one of the performance analysts. We would then simultaneously send a report for the opponents after this match day to the second performance analyst, so they could start early on their prep. The rest of their week is going to be dominated by looking for video and clips.

As you can see, this day, which we call match day -5 in football, is quite frantic! You're making sure everyone has what they need to the correct level of detail.

Matchdays -4, 3, 2, 1 are the ones where you can get in all the data quality assurance work, meetings, process reviews, ad hoc and long-term data work.

Then on the matchday itself, there would be the immediate post-match stats one page report and the speedy physical data turnaround needed for the +1 and +2 recoveries. Then on matchday plus one, there would be a detailed post match summary reports and reflection. This needed to be as soon as it could be delivered, to feed into the pre-match work for the next opponent.

There were also other processes going on, for the recruitment department, ad hoc reports, meetings with different suppliers and upskilling staff, which is vital, because everybody wants to go towards becoming self sufficient, especially with data and analytics. There would also be quarterly long-term trends reports. Juggling and balancing all of that is a massive skill!

And a small shout-out to a couple of people. When I started the department, it was just me on my own. As time went on, there was help in the form of First Team Data Analyst Andrew Peters and First Team Performance Data Analyst Michael Davies.

The key thing was having a process you can bring other people into. It’s no good building a process that only you can execute.

There were other key lessons which I brought forward into my current roles with 777:

• Automation is key. Anywhere there’s spare time to automate a process you know you will need again has to be ruthlessly used. I had an 80/20 rule of thumb - 80% of people’s requests to you on a weekly basis are going to be the same and 20% are probably going to be new. Anything in the 80% needs to be automated to make best use of everybody’s time.
• Ask questions before you build. Time is quite precious, so avoid assumptions and speak to a practitioner and see what the ultimate goal of what you are trying to build is.
• Take extra time to annotate and explain, especially new work. Even if you have built a really good analytics culture, you can sometimes get complacent when your’e presenting new work and sometimes it doesn’t become as effective as it could be, because you haven’t taken the time to explain it and have assumed knowledge.
• Consider the different modes of information delivery. Data visualisation is not the only way. It dominates on Twitter and in blogs, but once you’re in a club some people prefer a brief meeting, some are happy with a list of bullet points, some want it pointed out on video.
• And always play in the staff games! This might not seem relevant to analytics, but a football club is a social thing and there are many staff you might not see on a day-to-day basis. Any opportunity to be involved in wider things helps to build relationships and makes you more approachable.

You will see that the key lessons are inter-personal, not technical, and that being people-facing is the most important part, especially when you’re building a department from scratch.