Luís Campos stood in front of a room full of scouts in Lausanne last December and said the thing nobody in Silicon Valley wants to hear. The PSG football advisor, a man who has evaluated more talent than most algorithms have data points, told the UEFA Elite Scout Programme that the hardest part of scouting is not measuring what a player can do today. It is projecting whether his qualities align with the philosophy and needs of a specific club. Then he said something even more important: that skill cannot easily be replaced by technology.
I sat with that quote for a while. Because it captures something essential about this entire AI-in-sports conversation that keeps getting lost in the noise of billion-dollar market projections and injury-reduction percentages. The question everyone keeps asking is whether AI coaching analytics can replace human intuition. The answer is no. It is not close. And the people closest to the game already know it.
The Machine Sees the Data. The Coach Sees the Room.
You know the numbers already. Three out of four professional teams use real-time AI analytics. The injury prevention results are legitimate. Nobody with any sense is arguing that teams should ignore data when it comes to keeping players healthy. I will concede that ground all day long.
But there is a difference between using AI to monitor hamstring load and using AI to decide who gets the ball with ninety seconds left in a playoff game. One is medicine. The other is leadership. And no wearable sensor on earth can measure whether a player wants to be in that moment or is praying someone else takes the shot.
The 2025 coaching margin study laid this bare in ways that should embarrass anyone who thinks the spreadsheet tells the whole story. Researchers built the best model they could using 24 seasons of NBA data. They deliberately left out coach identity. And when the model finished its work, there was a gap. A residual. A cluster of wins that the algorithm simply could not account for. That gap tracked with coaching quality. Championship-caliber coaches produced it year after year. The model, to its credit, admitted what it could not see.
I have seen this movie before. Every few years, someone invents a new way to quantify the game and declares the eye test obsolete. And every few years, the game reminds them that humans are not data points. They are complicated, stubborn, brilliant, fragile creatures who respond to things no sensor can detect: a halftime speech that lands, a veteran's glare across the huddle, the decision to run a play for the young guy because the coach saw something in his eyes during warmups.
You can not measure that.
When the Algorithm Becomes the Coach, the Game Loses Its Soul
Adam Silver admitted it in Paris in January 2025. He acknowledged that analytics have pushed the NBA's offenses in a direction where teams are starting to look too similar. The league is now averaging more than 34 three-point attempts per game, up from fewer than 19 in 2010-11. The math says shoot threes. The math is correct. The game is becoming unwatchable because of it.
This is the part of the story the AI evangelists skip past. When you let the algorithm optimize everything, you optimize out the very thing that makes sports worth watching: the unpredictable, irrational, gloriously human stuff. The mid-range jumper that analytics says is inefficient but that Michael Jordan built a dynasty on. The fourth-and-one call that the numbers say to go for but that requires a coach to look his players in the eye and know, not calculate, whether they have it in them right now.
A PMC study on AI in sports science put it plainly: AI systems often struggle with the unpredictable nature of sports, including unexpected injuries, changes in team dynamics, and psychological factors affecting performance. Those are not footnotes. Those are the entire fourth quarter. Those are the playoffs. That is where games are decided, and it is precisely where AI is most blind.
The Laura Harvey story is instructive. She asked ChatGPT for a formation suggestion. It gave her a back-five. Her staff validated it, adjusted it, and made it work. Good for her. But notice what actually happened: a human coach had the courage to try something new, the judgment to know which suggestions had merit, and the ability to teach it to actual players who had to execute it under pressure. The AI was a brainstorming partner. Harvey was the coach. There is a canyon between those two things.
Tools, Not Gods
I respect AI as a tool the way I respect a good set of binoculars. They let you see further. They do not tell you what you are looking at. A Harvard Science Review analysis suggested a framework of roughly 70% data-driven decisions and 30% intuition. They were careful to note that the 30% is not guessing. It is hard-won expertise about when to override the algorithm.
I would flip those numbers for any decision that matters in the final five minutes of a close game. But the broader point stands: the best coaches in history have always been the ones who synthesized information from every available source and then trusted their gut when the information ran out. Phil Jackson did not need a laptop to read the emotional temperature of a locker room. Bill Belichick studied more film than any analytics department, and his genius was not in accumulating data but in knowing which data mattered and which was noise.
Daryl Morey, the godfather of NBA analytics, has publicly stated that AI and large language models are integral to the 76ers' decision-making. And yet the 76ers have not won a championship. The Houston Rockets ran the most analytically optimized offense in basketball history and came up short when it mattered most. Because in the end, the model could not account for Chris Paul's hamstring, or James Harden's body language in a close-out game, or the particular way the Warriors' chemistry turned them into something greater than the sum of their parts.
That is what winners do. They become more than the model predicts. The coaching margin study proved it with math. I have been proving it with my eyes for twenty-five years. The residual is not noise. It is the whole ballgame.
Use the tools. Monitor the workloads. Predict the injuries. Let the algorithm crunch the opponent's tendencies and surface patterns that save your staff hours of film work. All of that is smart. All of that is progress. But when the game is on the line and the arena is shaking and your point guard is looking at the bench waiting to be told what to do, the answer better come from a human being who has been in that fire before. Not from a dashboard. Trust what your eyes tell you.
