TI #025: Tracking Data Reveals the Tactical Behaviours That Win Matches?
Timeless Insights #025
Width. Depth. Mobility. Support. Superiority. Compactness.
They shape how coaches design sessions, how analysts clip games, and how teams talk about performance internally. But there is always a gap. These ideas are clear in theory, yet hard to measure in a way that consistently supports decisions.
That is what makes this paper interesting.
It does not start from data and search for patterns. It starts from football language — the general principles of play — and asks a more practical question:
Can we translate those principles into measurable features, and do those features actually relate to winning?
Using tracking data from 302 professional matches, the authors build a set of features that represent different attacking principles. They then test whether those features can predict match outcomes, both across full matches and within smaller windows of time.
The result is not just a model. It is a structured attempt to connect how coaches think about the game with how analysts measure it.
And that’s why the paper still feels relevant today.
Because even now, with better data and more advanced models, that translation layer remains one of the hardest problems inside clubs.
When tactical ideas become measurable
What stands out is not the final accuracy number, but how the features are constructed.
Each one is tied to one of five specific principles of play.
Possession is not just “having the ball,” but expressed through pass volume, accuracy, length, and direction.
Mobility is captured through how much a team expands after regaining possession — a proxy for how quickly space is created.
Superiority becomes more concrete through metrics like outplayed opponents, outplayed defenders, and numerical advantages in key areas such as the final third.
Disruption focuses on what happens to the opponent, measuring how much a pass shifts defensive structure or forces movement.
Chance creation is represented through the potential of a reception to lead to a scoring opportunity, combining location and defensive pressure.
Individually, none of these are perfect. But together, they form a measurable version of tactical behaviour that can be tested against outcomes.
One of the recurring issues in applied analytics is that models often produce variables that are hard to interpret in football terms. Here, the direction is reversed. The model inherits meaning from the concepts it is built on.
Football is always relative
One of the most important findings is easy to overlook.
The model performs better when features are expressed relative to the opponent, rather than in absolute terms.
That aligns closely with how the game actually works.
Completing 500 passes has no fixed meaning. Expanding quickly after a transition is only useful if the opponent is not doing it better. Creating numerical superiority in the final third matters more when it is sustained against an opponent trying to prevent it.
Everything in football is relational.
A team does not just perform. It performs against something. And that “something” is constantly adapting.
This is where many models still fall short. They measure output without fully capturing the interaction that produces it.
The paper reinforces a simple but important idea: tactical behaviour only becomes informative when it is placed in context.
When does performance become visible?
The second part of the study moves from description to timing.
Instead of only using full-match data, the authors train models on the first 25%, 50%, and 75% of matches. The drop in performance is smaller than expected. Even with just the first quarter of a match, the model retains a similar level of predictive accuracy.
This does not mean matches are decided after 20 minutes.
But it does suggest that tactical patterns emerge earlier than we often assume.
A team’s structure, spacing, and interactions shape the game early on, but the challenge is detecting it reliably enough to act on it.
And that leads to a more interesting question.
Not whether we can describe team shape — we already can — but whether those behaviours actually tell us something about performance while the game is still being played.
The real question
We can map positions. Measure distances. Describe compactness. Visualise shape.
But description is not the same as decision support.
The real value comes when those patterns are linked to outcomes in a way that is actionable.
So the question is…
Which of these behaviours actually matter most, and how stable are they across matches?
