Vanni Di Febo is a seasoned data scientist and Sportlight alumnus who now serves as a football data analyst for the Italian Football Federation and Italy National Football team. Although Di Febo’s focus in his current role with FIGC centers on technique, tactics and game play, his background in sports science and performance gives him a unique perspective on where those elements meet and how they interact. A recent conversation with Di Febo ran the gamut of data, sports science, performance tracking and practical match-day applications.
How does working for one of the most successful football associations in the world compare to working for a sports tech startup?
It has of course a lot of differences, from the size of the company to different work procedures. But in the end, my job hasn’t changed that much. I’m still a PySpark and Tableau guy, crunching numbers and deploying models to obtain insightful information for my employer. If you like football and math, this job is enjoyable at whatever stage you are doing it.
What are some trends you’re seeing in tech, data or similar systems in sports – and particularly where they meet?
I see that AI is getting more and more attention and people are involving it much more than before in sports analysis. At Sportlight, we used it to track the players, but there are also people developing models to understand the level of pressing of a team, the effectiveness of player selection, and so on. A fact worth pointing out: the smartest mind in the field is still going to be the coach. He knows his team and his players. As sensationally important as data is, its value lies in how that information is delivered to a coaching staff.
Coaches will rarely accept information they can’t easily understand, like when it comes from a machine learning model they know nothing about. The information has to be clear and not super-complex. It has to talk about football and not about the quality of your model. And it has to have an immediate, practical use. You can tell them that the team “availability for passing” metric is 66.5, that it comes from a very good model you developed over several weeks and that you’re very proud about. But how can they use that information in training? What’s going to change their approach is showing them the correlation between the passes received by a certain player in a certain zone and the overall offensive production of the team, so that they can train the team to direct the attack to that certain player in that certain zone.
Are there any differences you’re seeing between football associations and clubs, or ways they could be using data and/or performance tracking differently?
Yes, there are a lot of differences. First, on a team, you are playing every three (or, at most, seven) days, so it’s always pre-game time or post-game time. You have to analyze what happened in the last match or give information about your next opponent as soon as possible. In a Federation, you will likely have a very intense period in which seven different national teams are playing 20 games in 10 days, and then maybe an entire month without a single game.
Another difference is that a club analyst is mostly focused on a single team (first team or one of the academy teams). But in Federation, I work together with the coaching staff and management of all teams to bring value to the entire organization. Also: scouting. For a club, scouting means understanding whether or not another team’s player is undervalued by the market and can bring an upgrade to your team. Ultimately, the purpose is to decide whether or not to buy him. In a Federation, you have a certain group of players from which you have to select and slot for the most appropriate duties within that federation.
What are the challenges of applying performance data silos to tactical situations? Has there been much recent progress in that regard?
The main challenge is that football – vastly different from U.S. sports in this way – is a continuous, low-scoring game with no defined possessions and no clear outcome for any choice a single player makes. Furthermore, the team has to be considered on the whole, even if it is made up of 11 separate entities who must be monitored frame by frame. With all that in mind, I think tactical insights from tracking data are the future of this category.
Data is just a different way of describing a football game. The more granularity you are able to capture, the better you can tell the story. How can you describe such a complex game with only game events and sports science? For instance, a certain player may make a positive, productive pass. But what if, in the process, he missed a completely unmarked teammate? And what if that teammate could have easily scored, but the player instead chose to back pass to a teammate who, despite being able to control the ball, was immediately surrounded by three opponents, has no further passing options and is thus likely to lose the ball? To understand why a game unfolds a certain way, or a particular action was effective, you need as much information as possible.
Do you foresee a time when data analysts, trainers and coaches will take a more synergistic approach to game planning, tactics, etc.?
I think we are rapidly headed in that direction. In Italian Serie A, coaches have at their disposal a tablet with live information coming from tracking and events data, and usually an assistant is monitoring those insights to note if they can have any practical impact on real-time coaching decisions. Match and data analysts are often seated at the same table with coaches to develop game-plan strategies. Coaches, as I mentioned, are still the best minds on the job, and they are increasingly able to recognize the relevance and value of data. At a time when data is changing and improving so much of the world around us, leaving it at the door in a well-served sector such as football would be a waste of resources and growth potential. The game has changed so many times in the course of its history, and now data is helping to change the choices that coaches and players make for the better.