Football scouting with data and AI: metrics, models and bias

What data scouting in football actually involves

At professional level, data scouting usually combines three layers. Event data logs every on-ball action with a location. Tracking data adds the positions of all players over time, which allows measures of off-ball movement, pressing and space. Video links both back to what the eye can judge. On top of these, clubs build models: expected goals and assists, possession-value models that credit passes and carries for moving the ball into dangerous areas, similarity searches that find players with a statistical profile close to a target, and projection models that estimate how a player's numbers might translate to a different league.

AI enters in two places. Computer vision produces the event and tracking data from video at scale. Machine-learning models then rank, cluster and project players. Both depend on the quality and coverage of the data: leagues with full tracking and many seasons of events support richer models than youth or regional football, where data may not exist at all.

Metrics that help and the traps behind them

Three questions apply to every number. Is the sample big enough? Set a minutes threshold before looking. Is it adjusted for context, such as team possession and league level? And does it describe the role you are recruiting for? A full-back who will be asked to invert into midfield needs a different profile from one who overlaps to cross.

Youth recruitment: relative age and maturation

In youth football, performance data mixes ability with age and growth. Two effects matter most.

Relative age effect. Players born early in the selection year are over-represented. In England the cut-off is 1 September, and the CIES Football Observatory found that among English players in top European leagues, those born September to December made up 39.9% against 24.4% for those born May to August. The effect was strongest among players aged 21 or under and at the big training clubs.

Biological maturation. Two 13-year-olds can be several years apart in physical development. Early maturers run faster and win more duels, so their physical and duel data look better, and they are more likely to be selected. Bio-banding, grouping players by maturity rather than age, has been used in Premier League academy festivals since at least 2016, with maturity estimated by methods such as Khamis-Roche, which uses height, weight and parents' heights. Growth also affects injury: a study at an elite Spanish academy found that players growing faster before peak height velocity had higher injury incidence and burden.

The risk with AI is that it learns these biases efficiently. A model trained on which players a club selected in the past will treat being born in September and being tall at 12 as signs of talent, because that is what past decisions rewarded. More data does not fix this; it can make the bias look objective.

• Record birth quarter for every player assessed, and look at the distribution of your shortlist.
• Compare players within the same birth quarter or maturity band, not across the whole age group.
• Weigh technical and decision-making evidence more heavily than physical outputs before and during the growth spurt.
• Revisit released players: a late developer at 12 may be a different player at 15.

Designing a fairer trial or scouting day

A club that cannot afford data platforms can still make observation more consistent. The aim is to see each player in comparable situations, judged against the same criteria by more than one person.

Rotating small-sided games with banded groups

• Format: 4v4 plus goalkeepers on 30 x 20 m pitches, or 7v7 on 50 x 35 m for older groups.
• Grouping: first round by birth quarter, second round mixed, third round by approximate maturity if a qualified practitioner has estimated it.
• Duration: 6 x 6 minutes, 2 minutes' rest; rotate teams so each player meets different opponents.
• Observation: three observers, each with the same four criteria (receiving under pressure, decision with the ball, defending 1v1, reaction after losing the ball), scored 1–4 with one clip or note as evidence.
• Data: record minutes, position played and birth quarter alongside scores.

Running tournaments is one of the most time-consuming jobs at a grassroots club: registrations, the draw, the schedule and results. DorsalCUP does it for free, and the tournament fixture generator builds a one-day schedule in minutes, no sign-up.

Ethics, transparency and the law

Recruitment decisions affect careers, and in youth football they affect children. A few principles keep data on the right side of that responsibility.

• Human decision. UK data protection law restricts significant decisions based solely on automated processing and requires safeguards such as human involvement. A ranking can inform a shortlist; it should not release or sign a player on its own.
• Transparency. Tell players and parents what is assessed, how, and who sees it. Keep notes factual and evidence-based.
• Data minimisation. Collect what the decision needs. Maturity and health data are sensitive and should be handled by qualified staff.
• Scraped and third-party data. Check the licence and the provider's lawful basis before building a database from external sources.
• EU rules. For clubs operating in the EU, the AI Act lists AI systems used for recruitment or selection of people in employment, including evaluating candidates, as high-risk (Annex III). Professional players are employees, so systems that evaluate them for signing may fall into this category. After the 2026 amendments, the Annex III obligations apply from 2 December 2027.

The safeguards in filming youth football and data protection apply to scouting video too: lawful basis, secure storage and limited retention.

What data cannot see

Data describes what happened in the matches that were recorded. It sees little of how a player learns, responds to a mistake, communicates, or adapts to a new role, and nothing of their life outside football. It also reflects the team: a midfielder in a side that rarely has the ball will look different from the same player in a dominant team.

The most useful role for scouting data is upstream and downstream of the scout. Upstream, it widens the search and flags players who would otherwise be missed. Downstream, it tests a scout's impression against a larger sample. The judgement in the middle, especially with young players, stays human. For the tactical side of reading players in context, see opposition analysis and the match plan.

Frequently asked questions