“AI is just one small part of the wider world of football analytics”

Interview with sports informatics expert Professor Daniel Link

"AI is just one small part of the wider world of football analytics"

At the 2026 World Cup, intelligent cameras capture players and the ball - the system even tracks the exact position of each player's nose at any given moment. In this interview, sports informatics expert Professor Daniel Link explains how limb tracking data can be used in match analysis and how to determine which AI algorithms currently make a difference.

You previously outlined the potential applications of AI-powered game analysis for the 2024 European Championship. Have there been any new developments over the past two years?

Yes, significant progress has been made in this field. Personally, I find skeletal data, i.e. data on each individual limb, particularly interesting. Thanks to advances in camera technology and computer vision models, this data can now be captured with good accuracy.

What is this data used for?

The collection of skeletal data was initially introduced mainly for video assistant refereeing, such as the semi-automatically detection of offside positions, and for creating animations from a player's perspective to make match coverage more engaging. However, it is also highly valuable for match analysis, as positions such as those of shoulders, ankles, ears and the nose, as well as knee angles, can be used to answer performance-related questions.

How do nose movements help with match analysis?

By knowing the positions of the ears and the nose, we can estimate where a player is looking. From this, you can infer "situational awareness", such as which teammates and opponents a player actually saw when passing the ball. If the AI identifies that a player is focusing too focused on the ball and not scanning their surroundings enough, this can be addressed in training.

Can other insights be derived from this?

At this year's TUM Football Analytics Hackathon, we asked students to extract insights from skeletal data. This led to some very interesting projects. For instance, changes in running parameters such as stride length and leg extension were used to estimate player fatigue.

Is there such a thing as too much data? A critical mass that we'll eventually reach?

In fact, we already have more than enough data. The challenge lies in extracting the two or three pieces of information a coach needs to make decisions during a game from this enormous volume. However, this is an individual decision, shaped by each coach's training philosophy. In principle, you can recognize a coach's "fingerprint" in how AI is used. For instance, does a coach place more emphasis placed on identifying their team's weaknesses, or on finding flaws in the opponent's pressing behavior? What is actually used in practice, however, is rarely disclosed - that remains a trade secret.

So how do you determine which indicator is truly useful?

This is where science comes into play. We can examine whether a suspected performance indicator is actually associated with performance and then apply this knowledge in practice. For example, we recently showed that an AI model which calculates a player's availability based on positional data captures a performance-relevant aspect of play: higher availability is associated with faster space gain, which in turn is related with greater success. The may sound trivial, but without objective evidence, everything remains a matter of perceived truth.

Will amateur teams also have to use AI in the future?

AI is already being used in amateur soccer today, albeit not with the expensive 24-camera systems deployed at the World Cup. Even with just two cameras, however, a considerable amount of information can be captured. Commercial systems are available for this purpose. The error rate is higher, of course, but the use of such systems will continue to grow in the future. AI is just one small part of the wider world of football analytics.

Has AI already changed the way soccer is played today?

AI can have a significant impact on coaches' tactical decisions and interventions. For instance, it can identify behavioral patterns in opponents and in a team's own play that even professionals might not spot immediately. There will certainly be more and more applications in the future. However, to answer this question rigorously, we would need systematic scientific studies - and we are only at the beginning of this process. Moreover, it will never be possible to prove with certainty whether AI is the decisive factor in a team winning a game.

• Li, Y. & Link, D. (2026). Intention Driven Identification of In-Possession Match Phases in Association Football through Temporal Graph Learning. arXiv preprint. Advance online publication. DOI: 10.48550/arXiv.2606.09289
• Jiang, J, & Link, D (2026). Availabilty is asssociated with success in football. Zenodo Electronic Journal. Advance online publication. DOI: 10.5281/zenodo.20633636
• Dick, U., Link, D., & Brefeld, U. (2022). Who Can Receive The Pass? - A Computational Model for Quantifying Availability in Soccer. Data mining and knowledge discovery, 36 (3), 987-1014. DOI: 10.1007/s10618-022-00827-2

• Prof. Daniel Link is a member of the TUM School of Medicine and Health and a core member of the Munich Data Science Institute.

More issues