Mikkel Keldmann: From one intern to automated analysis at Brondby
Mikkel Keldmann is Head of Quantitative Analysis at Brondby
Written by Mikkel Keldmann — January 18, 2022
THIS is the story of how Brondby moved from just one guy in the corner with a computer (me) to having a fully integrated data and analytics department.
1. One intern and a focus on training load (2015/16)
I started as an intern at Brondby in 2015, while I was doing my Masters in economics at the University of Copenhagen. The role developed into part time and then full time.
In 2016, we installed Inmotio on our training pitches and got a German Head Coach (Alexander Zorniger) who was very focused on pressing and gegenpress, so our focus shifted to training load and match performance algorithms to improve conditioning and work towards the style of play he wanted.
This led us to produce reports like this (pictured), which is an example of player training the day before a match. You can see normal metrics that would show up in any training load, like total distance, high speed distance, but also an example of match load on an individual or team level, in terms of volume and intensity over time.
This could help our coaches plan and analyse training with a focus on slowly increasing or decreasing intensity.
2. Raw tracking data and interacting with coaching team (2017/18)
In 2017/18, we generally had three students writing their Masters with us, usually in human physiology, training and sport.
This shifted in late 2018 towards more data subjects - data science, economics - which was also when we shifted from an intern solution in Amisco (which looked like a 2D Football Manager) and Prozone to raw event and tracking data from Opta and Tracab. The league had a deal for every club to get this data.
We did this by implementing a data infrastructure, primarily in a Microsoft database where we could house our events and aggregate training and match events. Handling data manually in Excel would have been a strain on manpower, so we started to step into automations, first with the help of Alteryx and later R and Python, which allowed us to schedule jobs and do a lot of automated workflows.
Then we presented the data to the coaches and scouts (our interactions had gone from being primarily with the physical coach and performance analyst to more of the coaching team and scouting team) in some kind of BI (business intelligence) tool like Tableau and later Power BI.
This move into raw data forced us to sit down and have a chat about how we wanted to approach analytics. We knew we had to be independent and flexible, so when the Head Coach came to us with a question on a Monday morning we had the answer before he came with another question on the Wednesday or Thursday.
If we can’t do that then we are to some extent failing in our job. We also needed to be able to alter our method whenever we saw fit; not only because of a coaching change but also perhaps because of a change in literature or in the understanding of football.
BIG DATA WEBINAR
Seven top presentations:
- Mikkel Keldmann: Brondby's Data Journey
- James Young: Launching a Data Science Department at Sunderland
- Jan Van Haaren: Evaluating & Recruiting Using Open-Source Tools
- David Perdomo Meza: Democratising Data
- Jens Melvang: Supporting Key Recruitment Challenges Using AI
- Luke Bornn: Using Tracking Data to Improve Performance
- Pedro Pereira: Data-Driven Insights in Opposition Analysis
3. Real-time analysis and interaction with whole football department (2019/20)
By now we were interacting with: the whole of the coaching and scouting team, the Director of Football, the Board AND the Academy.
We now had, on average, one full-time and three-part time students, along with seven students writing their Masters. Two of these tended to be human physiology students, with the others being students in data or software development.
The problem we faced with the move to using more raw data and doing it ourselves was that we now had less time to develop new ways of doing things. So we decided that from the summer of 2020 we would hire three part-time students, on 15 hours a week, who would get paid a little and each be specialised in some area.
We had an analyst, who does the front end; a data scientist, who does the modelling; and a data engineer, who does a lot of the back-end work. We did this to prioritise automation and to free up time for development in order for me to start developing instead of just running day-to-day work.
This, along with our shift into real-time analysis with a new tracking provider, allowed us to start experimenting with live physical and event data using external software like Match Tracker and developing our own apps.
We are also trying to predict more. This (pictured above) is an alpha test of a training load predictor, where we use a streamlet to build a web app for our physical coach or the technical coaches, where they can start planning a training session and it gives a prediction of what the physical load on each player group will be, as well as an average.
We hope this will enable our coaches to start predicting the training before they go onto the pitch, rather than afterwards.
How you can start doing analytics from scratch
That is our journey into analytics at Brondby. Now, what can you do if you have nothing but want to start doing some analytics work with your team?
Firstly, it will help if you think about the philosophy and way of working you would like. Everything you do depends on how you balance flexibility, complexity, and cost; how easy things are to use and implement; and how much time will go into development compared to production.
You can do things on a spectrum. For example, most of what we do at Brondby is in-house, which takes a lot of development time and effort, but means things can be customisable. On the other end of the spectrum, end-to-end software is easier to implement but less customisable.
Alternatively, you could have a mixed set-up, where you mostly rely on an outside company and do your own internal work for specialised bespoke things.
Assuming that you want to do what we did at Brondby, what would you need? The answer is actually pretty simple:
- A computer.
- Some free software, like R and/ or Python.
- Access to some kind of data, which could be from very aggregated to very granular.
- A little manpower to implement these things.
The next step would be to get some storage. You can do this for free, or at least for a low cost, with MySQL, Microsoft, and you could also do Power BI for free.
In total, you could do ALL of this for free. You could do your visualisations straight from R or Python in web apps or in PDF reports, again enabling you to do it for free.
What would be the next step? Again, it depends on your goal. If you are trying to analyse throw-ins or corners or some kind of set piece, you could do this by having some event data, R or Python, and looking at end touches or second touches during these set plays.
Gabriele Gnecco: At Hudl, we've looked into the organisation and structure of data departments and recognised four main figures:
- Data Architect: In charge of building the data warehouse infrastructure and pulling data from different sources.
- Technical Analyst: Has to build the dashboards and make the insights understand the coaches and players.
- Data Scientist: Has a statistical background and is in charge of building predictive mathematical models and expanding the amount of information that can be retrieved by the club from data.
- Tactical Analyst: The football wise guy of the department, in charge of linking it with the football side of the club.
All these roles can be hybrid, because one person can cover more than one of them at the same time.
Gabrielle Gnecco is Solutions Consultant at Hudl and presented at the Big Data Webinar.
If you want to put together a fully-customised report on your playing style, you can still do it using the basic set-up of R or Python and some data. A BI tool that would enable you to quickly share the report with coaches or the Director of Football would probably help - as would having somewhere to store your data, instead of having to go through 10, 15 event data files to build up a season report.
Lastly, if your goal is to find the weaknesses in your opponent using deep learning or AI, then you probably need a more fully-automated data infrastructure. If you manage that then please send me an email to say how you did it!
KEY MESSAGES
The first thing is the closer the better. If you are in the room, people get used to you being there.
Physical proximity and availability are key. Do not put your only data intern in an office far away from the coach, in the basement somewhere, because you won’t gain anything from the analytics department and won’t move closer to implementing it.
Buy-in is also very important - not just from the executives and the Board, but also from the end-user. If the Head Coach does not buy into the project then the Board might spend the money but it still won’t get to the field.
The buy-in you get from being in the room and solving problems allows you to make small steps towards growing the analytics department. Particularly in the beginning, you should tailor the output to the end user.
Some coaches will buy in from the moment you get there, because they are interested in what this can do for them, but some will definitely not. So tailoring your output or the way you do things to the people in the club is paramount. You should think about not only the data literacy of coaches and specialists, but also the tech literacy.
If you start by putting a dashboard in front of someone who is not used to using a computer then you will get nowhere. There might be a better and bigger gain by printing the report that the user is expecting. Or saying, 'I’ll send it on a PDF or an email, tell you how to open it and help you the first 10 times.'
After half a year you might then move on to the dashboard. The most important thing is getting it implemented, not how fancy your dashboards or apps are. Trust is earned, just as in real life, and you have to create a working relationship with people.
You will not be able to revolutionise football operations overnight. Start by solving simple practical questions, or by helping people to do their jobs faster, more efficiently, more easily.
If you do that, you will get buy-in and be allowed to move on to the next project, something bigger, something more exciting, something more valuable for the club maybe. Then it allows for that organic growth. Instead of just saying, 'We want to do analytics,' start by implementing a culture of analytics and an understanding of what it can do and help with.
These TGG Masterclasses might help you on your analytics journey:
From there you can go anywhere. You could potentially scale to a Liverpool-sized analytics department, or you could stay with two guys - whatever your philosophy and way of working with analytics tells you.
Good luck to everybody on that journey.
- This article is an edited version of Mikkel Keldmann's presentation at TGG's Big Data 2021 Webinar.
