AI on the Coaching Bench: When Football Hands Its Heart to Algorithms

On 15 July 2026, I sat in row eleven of the Luzhniki press box in Moscow and watched something that people are still arguing about years later. When nineteen-year-old Kylian Mbappé received Lucas Hernández's long ball in the 65th minute, at least two systems were watching him at once. The first was the human eye: mine, the eyes of two thousand journalists, the eyes of eighty thousand spectators. The second was fourteen cameras mounted around the pitch, plus a sensor chip inside the ball, sampling at five hundred times per second.

Four years later, in Qatar, it was that second system - Semi-Automated Offside Technology, or SAOT - that confirmed Mbappé's goal in the World Cup final against Argentina in roughly twenty-five seconds. The assistant referee no longer had to chase the play. An algorithm did it instead. And almost nobody objected.

I tell this story not to praise machines. I tell it to pose the question I believe every serious football analyst must answer within five years: when a system can decide the fate of a goal faster than the human eye, where does the heart of football - the part I learned to hear during the empty stadiums of 2026 - actually sit?

To answer, I have to start somewhere unexpected: a social media poll by US President Donald Trump, published on Truth Social. The poll proposed renaming Artificial Intelligence, offering three options: Superior Intelligence, Extreme Intelligence, Supreme Intelligence. Alongside it came policy moves: a task force called AI Force, an executive-branch AI coordinating official, and a claim comparing AI's impact to a new industrial revolution.

I read that item one morning in São Paulo, and what caught my attention was not the politics. What caught my attention was how it had been labelled. That item - sixteen information points in total, not one club, player, league or federation - had been sorted by an automated classifier into the football category. No football entity appeared anywhere in it. Yet it entered the sports data pipeline.

And I thought: if an algorithm cannot tell political news from football news, how will it ever tell the difference between a striker with a psychological injury and a striker simply out of form?

That is the starting point for this analysis. Not Trump, not the poll. But how artificial intelligence has entered football, how fast, and what price is being quietly hidden.

Based on my experience covering matches - thirty-five years, from small stands in Newark to press boxes in Russia, Brazil and Qatar - I can tell you one certain thing: football has become one of the fastest-adopting industries for artificial intelligence on earth. And almost nobody is calculating the price correctly.

Look at SAOT. The system uses twelve dedicated cameras mounted under the stadium roof, tracking twenty-nine data points on each player's body - not just the ball, but knees, shoulders, elbows, ankles. The chip in the ball measures oscillation at five hundred times per second. The whole dataset is processed within seconds and sent to the video referee. At the 2026 World Cup, the average processing time for an offside call was about twenty-five seconds, against an average of more than seventy seconds at the 2026 World Cup when VAR was still manual.

But here is the number few mention: every 2026 World Cup stadium had to install camera and sensor systems costing an estimated millions of dollars, and FIFA had to contract technology suppliers to run it. Only a handful of domestic leagues worldwide can afford SAOT. The English Premier League announced a similar rollout from the 2026-2026 season. The Brazilian league, the Argentine league, the Vietnamese league cannot.

This is what I want you to remember. Artificial intelligence in football is not a universal advance; it is a form of capital, and capital always distributes unequally. When a smaller league lacks semi-automated offside, every contentious goal there becomes a lasting scar of fairness. When a mid-tier club cannot afford injury-tracking software, it loses players to problems bigger clubs prevent with an algorithm.

Watching matches in those leagues, I realised: the AI gap in football is not only a tactical gap. It is a health gap for players.

And this is where the story turns serious.

Look at the modern calendar. Two matches a week has become the standard in Europe's top leagues. A player representing both club and national team can play fifty to sixty matches a season. Meanwhile leagues in South America, including the Brazilian league where I live, carry denser calendars still, with state championships, Copa Libertadores and World Cup qualifiers overlapping.

Big clubs now use injury-prediction systems to manage this. They feed GPS data, load data, sleep data and blood-test data into machine-learning models to compute each player's injury probability for each match. These systems can detect early overload signals the human eye misses - micro-changes in stride, in reaction time, in ground-force distribution.

But here is the truth data analysts rarely say out loud: no algorithm compensates for playing two matches a week for ten straight months. The algorithm only helps you choose who gets injured first.

I have written this many times, and I will write it again: the biggest culprit for injuries in modern football is not bad pitches, not boots, not technique. It is the calendar. AI cannot fix the calendar. AI only makes the problem more invisible by giving us a beautiful dashboard to stare at.

I say this because I know club medical departments are judged on days players miss. When a coach decides to push a player who has already played two matches into a third, he is not the one ultimately held responsible. The club doctor is the one asked why the player got injured. And the club doctor does not control the calendar.

Now look at scouting. This is the field where AI has changed football most deeply, and few notice.

Brentford in the Premier League is built on a data model developed by owner Matthew Benham through his company Smartodds. Liverpool under Jürgen Klopp used data science as a core part of recruitment. Brighton & Hove Albion under Tony Bloom turned probabilistic analysis into competitive advantage by buying players in under-scouted leagues. These clubs do not buy stars. They buy hidden metrics.

What these models achieve is genuinely remarkable. They can spot a twenty-year-old full-back in the Brazilian second division with the potential to convert into a Premier League wide midfielder, based on movement and decision patterns the naked eye will not catch in a single match. They can value a player on progressive passes into the final third per ninety minutes, rather than on goals scored.

But here is the downside I have seen directly: these models are mostly trained on data from big leagues. Applied to South American, Asian or African football, they are often wrong. A player performing well in Brazil can score badly in a European model because the model does not understand tactical context, teammate quality or pitch conditions.

So AI in scouting is creating a paradox: it opens global opportunity for players while simultaneously imposing a European standard on everything it sees. Whoever does not match that standard is overlooked - not because they are worse, but because the model was never taught to understand them.

I remember a conversation with a scout in São Paulo. He told me something I have carried ever since: an algorithm does not lie, but it only answers the question you ask it. If you only ask about speed, it will only answer about speed. And so a slow player who reads the game superbly can be excluded from every list.

This leads me to what I consider the most important part of this analysis.

There is an implicit assumption almost the entire football-AI industry operates on: that everything important in football is measurable, and everything measurable is optimisable.

I do not believe that. And I have a specific reason.

In 2026, when the pandemic froze every league and I lost my television commentary contract, I launched a livestream series called Locker Room in the Dark. I invited assistant coaches, stadium cleaners, unemployed former players to tell their stories. The episode with João - a man who had cleaned the Maracanã for twenty years and lost his job overnight to social distancing - reached 2.1 million views. More than 340,000 reais was donated to an emergency fund. In 2026, I learned to hear football with my heart when the stadium no longer had a human voice.

And across those months, I realised something no algorithm can compute: football runs on tens of thousands of people whose names never appear in any dataset. The grass cutter, the kit washer, the team bus driver, the floodlight repairer, the ticket seller. When AI talks about performance optimisation, it talks about eleven players on the pitch. It does not talk about the two hundred people behind them who make those eleven able to play.

An assistant coach I once interviewed told me: a player is not a collection of metrics. He told me that when nineteen-year-old Kylian Mbappé walked into a press conference in Russia, the most important thing was not his running speed. The most important thing was whether he could sleep the night before the match.

People laughed at me for asking Mbappé about emotions; then the whole of France wept with happiness. And that reminds me that in every AI model I have ever read about football, there is no input field labelled fear.

Let me be clearer about that missing input field.

When a twenty-two-year-old moves from the Brazilian league to Europe, analytical models talk about tactical adaptability, integration speed, resale potential. No model talks about him being alone in a Manchester flat, eight thousand kilometres from family, in the first winter of his life. No model talks about him having nobody to speak his mother tongue to for the first three months.

I am not saying AI is useless. I am saying AI is being used to answer the easy questions, while the hard ones - the ones that decide a person's career and mental health - are left behind.

And here I want to address the economics of AI in football.

Data analytics systems, injury prediction systems, match-tracking systems, betting-integrity systems - all are commercial products. Companies like Sportradar and Genius Sports monitor hundreds of thousands of matches a year to detect match-fixing signals, and sell that data to federations. Other companies sell motion-analysis software to clubs for hundreds of thousands of dollars per season.

A football AI market is forming, and it runs on market logic: whoever pays most receives most. This is not wrong in itself. But it raises a question I have never heard anyone in the industry answer satisfactorily: if AI helps rich clubs prevent injuries better, buy players better, read matches better, is it making football more unequal?

I believe the answer is yes. And I believe it not because I distrust technology, but because I have watched the opposite happen in another field.

Look at the Mbappé goal I opened with. When SAOT confirmed it, the whole world saw fairness. But behind that fairness sits a system worth millions of dollars that only some stadiums in the world possess. Football has achieved fairness at the summit, and unfairness at the base.

Now let me address what I consider AI's biggest risk in football - bigger than the financial risk, bigger than the technical one.

That risk is: AI makes decisions invisible. When a player is left out of the Sunday squad, nobody tells him an algorithm calculated his injury probability as higher than others. Nobody tells him a model ranked him below others. He only knows he is not playing. And the question of why has no clear answer.

I am not saying clubs must publish their entire algorithm. I am saying a responsibility gap is forming, and players - flesh-and-blood people - sit on the losing side of it.

This is exactly where I recall my own story.

That transfer window was not merely a purchase; it was a crack in the whole of football. I am talking about the Neymar transfer of 2026, when Paris Saint-Germain triggered his 222-million-euro release clause. As the football world praised the historic deal, I wrote that PSG was buying a brand, not discipline. A group of supporters and male colleagues mocked me. I did not retract. I attached the table of PSG midfielders' average touches in the previous Champions League season. In March 2026, PSG lost 5-2 on aggregate to Real Madrid. Brazilian football forums scrambled back to my old tweet.

I retell that not to boast. I retell it to make a point: data has value when it answers a specific question. That PSG side did not fail for lack of data. It failed because the data they had did not answer the right question: how does a team of stars play like a team.

And AI today is the same. It can give you thousands of data points on every player. But it will not give you the answer to how eleven human beings learn to trust each other.

Now I want to return to Trump's poll, because it teaches an unexpected lesson about football.

Looking at how the media handled that story, I see exactly the pattern I have seen hundreds of times in football. A big statement is made. Media report it in a tone as if it had already happened. But read closely and you realise it was only a proposal, a question, a non-binding poll. No official document. No change. Nothing at all.

Sound familiar? It is identical to how a transfer rumour spreads. One source says a club is negotiating. Sites repost with headlines as if the deal is done. Fans buy the player's shirt before he has signed. Then the deal collapses.

The truth is: in both politics and football, a proposal is not a decision. And a poll is not a policy. This is a lesson football fans have learned by heart through thousands of failed transfers.

And this is what I want football's AI analysts to understand: your model is only as good as the data you feed it. Feed it a rumour and it will output a conclusion. But that conclusion is not the truth.

I have written for years that I never press publish unless there are at least two concrete numbers in the piece. That is not a habit. It is a safety fence.

Now let me address what I believe is the real prospect of AI in football - the part I think will change the game within a decade.

It is not scouting. It is not tactical analysis. It is players' mental health.

I have said I am known for questions like can he sleep. I ask that because I believe it is the most important question nobody is answering. In the next decade, if AI can help clubs detect early signs of depression, anxiety, burnout in players - not to push them to play on, but to help them rest at the right time - that will be technology's greatest contribution to football.

But to do that, we must change the question we put to AI. We must stop asking only how to make players run faster, and start asking how to make players live better.

I know this sounds dreamy. I am a fifty-one-year-old woman working in a male-dominated industry, and I have earned recognition through competence, not identity. I am not dreamy. I only observe. And I observe that things once dismissed as soft - emotion, mental health, kindness - are increasingly becoming competitive advantages in elite football. Not because the industry became kinder. But because it realised that a mentally healthy player performs better.

That is the kind of optimisation I can support.

Now let me be honest about where I might be wrong.

I am not a data engineer. I do not build machine-learning models. I am a commentator who has spent thirty-five years observing this industry from outside the analytics rooms. Perhaps I am underestimating how quickly AI will solve the problems I have just raised. Perhaps in five years models will learn to measure even what I consider unmeasurable.

Perhaps I am wrong to say the calendar is a problem technology cannot solve. Perhaps I am wrong to doubt scouting models that impose a European standard on the rest of the world.

Last week I got a match prediction wrong after analysing it very carefully. I write that down whenever I can, not because I enjoy self-criticism, but because I believe an analyst who never admits error is not an analyst. He is a salesman.

But there is one thing I do not think I am wrong about. It is this: if AI becomes the only voice in the dressing room, football will lose what makes it different from every other sport. It will become an optimisation problem, and the winners will always be those with the most data, not those who understand football best.

And that is a prospect I, as someone who has spent a lifetime hearing football with her heart, do not want to see.

I will end with a question anyone reading this can verify within three months.

Watch the coming season and note: when a top club changes manager mid-season, what reason is given? If the reason is poor results, AI has not yet taken over. If the reason is declining performance metrics and the board was warned by a data model before the results turned bad, then AI already sits on the coaching bench.

And watch this: among the injured players that season, how many played more than forty matches the season before? If that number is high, what I said about the calendar holds. If it is low, I was wrong. The data will answer for me. And I will accept the answer, whether or not it favours my argument.

That is the only way I know to do this job: make a verifiable prediction, then let the truth judge.

And if you ask me what I believe most after thirty-five years watching football, I will say this: technology will keep changing how we play this game. But no algorithm will ever run faster than the emotions of a nineteen-year-old walking out for a World Cup final.