Football measures more than any sport in history and forgets faster than most. Every touch is logged, every metre is tracked, every match is filmed from eight angles, and then a manager leaves in November and the reasoning behind a season of decisions leaves with him. The spreadsheets stay. The judgement does not.
Tracking, event feeds, video, medical systems and scouting platforms have all improved. The decision they exist to inform, which is what this player should do this week and why, is still made in a meeting, from memory, and is still very rarely recorded in a form that survives a change of manager.
Key findings at a glance
- Availability is a performance variable, and the evidence is unusually good. Across 11 seasons and 7,792 injuries in the UEFA Champions League injury study, lower injury burden was associated with better league ranking, more points per match and a better European ranking.
- The injury picture is improving overall and worsening in one place. Across 18 seasons, overall injury incidence fell by around three per cent a year. Over 21 seasons, hamstring injuries roughly doubled as a share of the total, from about 12 to 24 per cent, with 18 per cent of them recurring.
- Congestion is dangerous in a way the running data will not show you. A meta-analysis of 16 studies found fixture congestion of 96 hours or less had no significant effect on total distance covered. Muscle injury rates rise with congestion regardless. A department watching only distance will conclude the squad is fine.
- Tracking systems agree about the easy variables and disagree about the useful ones. Total distance is robust. Accelerations, decelerations and derived metabolic measures are considerably less so, and they are the variables load models lean on hardest. A club that changes provider has changed its own history.
- The strongest AI result in football is a decision-support result. In the TacticAI study, expert raters at a professional club favoured the system’s corner-kick suggestions over existing tactics in about 90 per cent of cases. The system proposed. People chose.
- The barriers clubs report are organisational, not algorithmic. In a survey of 29 national federations and 32 professional clubs, the leading obstacles were resourcing, in-house expertise and the absence of a shared metric vocabulary, with only around 30 per cent finding provider positional data sufficiently transparent.
The data-rich, insight-poor club
The thing that strikes an outsider about elite football is not the sophistication of the analysis. It is how little of that analysis reaches the decision. An analyst produces a report. A meeting happens. A decision is taken. The report is archived and the reasoning is not written down anywhere, so when the same situation arises in March nobody can reconstruct why the club chose what it chose in September.
That is a memory problem rather than an analytical one, and it is the problem a change of manager makes catastrophic.
Congestion, and the metric that hides it
Fixture congestion is the clearest example of a club watching the wrong number. Congestion of 96 hours or less has no significant effect on total distance covered, so a department monitoring distance sees a squad coping. Muscle injury rates rise with congestion anyway. The metric that is easiest to collect and most consistent between providers is precisely the one that will not show the risk.
The variables that would show it, meaning accelerations, decelerations and derived metabolic measures, are the ones providers disagree about most. That is an uncomfortable combination and it is not solved by buying more data.
What the market sells that clubs should refuse
Three things. Single-score readiness indices, which hide the disagreement between their inputs. Individual injury prediction, whose published performance does not support the decisions it is sold to make. And models trained on a club’s own past selections, which recommend the same selections back with a decimal point attached and call it insight.
What follows for a football club
Join the records under one player identity before buying any intelligence, because a system reasoning across five incompatible platforms will reason confidently about the wrong player. Insist that every recommendation names its sources, states its confidence and offers alternatives. Place a named person between any recommendation and any player.
Above all, record decisions. A decision record costs nothing at the moment of the decision and cannot be reconstructed afterwards at any price. It is the only asset in this field that appreciates, and it is the one thing a change of manager currently destroys.
Free download
Get the full paper
Read The Augmented Touchline in full, with every claim carrying an evidence grade and the full reference list. Give your name and email and the PDF opens straight away.
Frequently asked questions
Does player availability actually affect football results?
Yes, and the evidence is unusually strong for this field. Across 11 seasons and 7,792 injuries in the UEFA Champions League injury study, lower injury burden was associated with better league ranking, more points per match and a better European ranking. Availability is a performance variable rather than a medical footnote.
Is fixture congestion dangerous for players?
Yes, and the most commonly monitored metric will not reveal it. A meta-analysis of 16 studies found congestion of 96 hours or less had no significant effect on total distance covered, while muscle injury rates rise with congestion regardless. A performance department watching distance alone will conclude the squad is coping when it is not.
How comparable are football tracking systems between providers?
They agree on the easy variables and diverge on the useful ones. Total distance is robust across systems. Accelerations, decelerations and derived metabolic measures are considerably less so, and those are exactly the variables load models depend on. A club that changes provider has effectively changed its own history, which makes multi-season comparison unreliable.
Can AI pick a football team or predict injuries?
It should not be asked to. A model trained on a club’s own past selections learns the club’s history rather than the game, and returns the same biases with a decimal point attached. Individual injury prediction is not supported by its published performance. The strongest published result in football AI is decision support rather than decision making: in the TacticAI study, expert raters favoured the system’s corner-kick suggestions in about 90 per cent of cases, but the system proposed and people chose.
What stops clubs getting value from their data?
Organisation rather than algorithms. In a survey of 29 national federations and 32 professional clubs, the leading obstacles were resourcing, in-house expertise and the absence of a shared metric vocabulary. Only around 30 per cent found provider positional data sufficiently transparent.
About this series
This is the sixth of seven papers on the responsible use of artificial intelligence in performance sport. The governance framework it draws on is The Responsible Performance System, and the youth football counterpart is Casting the Net Wider. The series also covers elite athletics, grassroots athletics, triathlon and golf.
Legal and regulatory notice. This article and the paper it summarises describe legislation, regulatory guidance and governing-body requirements. Nothing here constitutes legal advice, and the provision of legal advice sits outside the terms of any engagement with the author or with Athleet.AI. The material is presented to support discussion and further review by qualified advisers. Every club, player and scenario used as an illustration is synthetic.