Artificial Intelligence in Football Coaching and Decision Support: What It Can and Cannot Do

Artificial intelligence in football coaching and decision support is most useful when you treat it as a second reader of the same evidence your staff already handles: form, strength, injuries, schedule, the opponent's habits. It does not see anything a good analyst cannot see. Its value is speed, consistency and a stated probability instead of a vague "should be fine". The coach still owns the decision, and the tool earns its place only if it changes a conversation for the better.

Start with how hard the next fixture really is. A good decision-support output gives you probabilities for home win, draw and away win, not a verdict. That distinction matters in a congested week. If your side is given, say, a 40% chance of winning away, the staff can ask what that implies: a match you are expected to contest rather than control, where a rotated side might cost you little or a lot depending on who is missing. The number frames the argument. It does not settle it.

Rotation is where this helps most, because the cost of a wrong guess is spread across several matches. Used well, a forecast lets you compare fixtures against each other: where a weakened lineup still leaves a reasonable chance, and where it does not. It also shows how much the picture shifts once confirmed line-ups are known, which is why timing matters. A call made days out and a call updated when the team sheets land are different products, and you should know which one you are reading.

Scenario preparation is the third use, and the least glamorous. Exact-score estimates are not predictions of the final scoreline. They are a way of seeing which game states are plausible: an early goal, a tight 1-0, an open game where both sides score. A goal simulation built on each side's recent scoring and conceding, calibrated to the market's view of the favourite, gives a staff a menu of likely shapes to rehearse. Treat it as a prompt for set-piece and substitution planning, not a script. Staff who ask what AI prompts for football coaches should look like will get more from a plain question: which two or three scenarios carry most of the probability, and what do we do in each?

Judging whether a forecast is any good is the part clubs most often skip. Three habits help. First, ask for a record over many matches, graded on a fixed definition, because a single result proves nothing: a 70% call is still expected to miss about three times in ten. Second, compare it against a naive baseline, such as always backing the home side, since a forecast that cannot beat a simple rule is not adding information. Third, check calibration and a proper scoring rule such as the Brier score, which rewards honest probabilities rather than confident guesses. Lower is better, and the comparison that counts is against a baseline, not against perfection.

A worked example shows what an honest record looks like. Kickwise publishes a pre-match probability for each match it covers, blending a language model's reading of the match facts with the market's consensus, or a form-based estimate where no market exists. Its public record keeps the last call made before kickoff and grades it on the result after 90 minutes plus any extra time. As of 2026-10-09, calls at 65% confidence or higher over the last 90 days were right 45 times out of 60 (75%), in matches where the home side won 52% of the time. Across all graded calls over the last 30 days, at every confidence level, the record shows 36 right out of 62 (58%), with a Brier score of 0.549 against 0.656 for a baseline using the sample's own home, draw and away rates.

Those same facts carry the limits you should expect from any such tool. The draw is the hardest outcome to call. Friendlies and matches with thin data are close to coin flips. And a short run of results, good or bad, says little about quality, so a record has to be read over many matches. If a data product, bought or built, cannot show you its own misses and its comparison against a simple baseline, you cannot trust its hits either.

Where does it stop being useful? At the things a probability cannot hold: a player's state of mind, a dressing-room dynamic, a tactical idea your coach wants to try precisely because the numbers do not expect it. The sensible use of ai tools for football coaches is as a check on your own priors and a cleaner way to brief a board: here is the expectation, here is the range, here is why a poor result does not by itself mean a poor process. The tool informs match preparation. The coach decides.