Most football analysis starts from concepts that already have names, such as a counter-attack or a high press. Pattern recognition applied to tracking data starts from the movement itself and finds groupings that nobody defined.
Unsupervised methods need no labels
Supervised approaches require examples of what is being looked for, which limits them to concepts already articulated by coaches. Unsupervised methods instead group passages by similarity.
Applied to tracking data, they cluster sequences in which players moved in comparable ways relative to each other and to the ball.
The resulting groups are defined by the data rather than by football vocabulary, so some of them correspond to nothing anyone had previously described.
Representing a passage of play is the hard part
Comparing two sequences requires reducing each to a form that can be measured against another, despite differing in duration, pitch location and personnel.
Choices about how to normalise position, how to handle time and whether to track roles or individuals determine what similarity means in practice.
Different representations produce entirely different clusters from identical footage, which makes the representation the analyst's main lever.
Interpretation is where the value is created
An algorithm returns groups without explaining them, so someone has to watch examples from each cluster and work out what they share.
Some clusters turn out to be artefacts of the representation, such as sequences grouped by pitch location rather than by anything tactical.
Others correspond to genuine recurring patterns that a coaching staff recognises immediately once shown, even without having a name for it.
Opposition preparation is the strongest application
Clustering an opponent's build-up sequences shows which patterns they return to under pressure and how often each appears.
This is more informative than a list of average statistics, because it describes tendencies as behaviour rather than as summary numbers.
Preparing for a specific pattern is also easier to communicate to players than preparing for a distribution.
Naming things changes how they are coached
Once a pattern is identified and given a label, it can be trained, defended against and looked for in recruitment.
Football's tactical vocabulary has always grown this way, with practitioners noticing something repeatedly before agreeing on a term for it.
What has changed is the speed, since a season of tracking data can surface in days what previously required years of watching to notice.