Craig Hepburn worked at UEFA before becoming an independent AI strategist and Perplexity Fellow, and writes Ground Truth, a Substack on the economics of artificial intelligence. Speaking ahead of his participation in WFS Madrid 2026, he argues that AI will do more to football’s business over the next decade than the last one did, with one exception: the ninety minutes, the one thing no algorithm can fake, only grow in value as everything around them turns synthetic.
In this edition of The Pundit’s Seat, he talks about why the resource gap between the biggest and smallest clubs may be closing for the first time, why football can finally be run like the data business it always was, what he refuses to delegate to AI in his own work, and why he chose to answer WFS’s written questionnaire by asking Claude to interview him aloud rather than type a single word.
You have argued that agentic AI is not an evolution of digital tools but a shift in the economics of production. What does that mean for a football organisation?
The shift is in the economics of production itself. For twenty years, the question behind any digital project was whether you could afford to build it and who you’d get to do it. That question is going away. The cost of producing software, content and operational capability is falling towards the cost of compute, and that changes what a club or a league can sensibly take on.
Take it in three parts. First, building. The coding models, Codex, Claude Code and the others, take you from an idea to a working prototype in an afternoon. A fan app, a second screen product, an internal tool that used to need an agency, a design team, an IT department and a six figure discovery phase, you can now stand up for the price of a subscription. A club can build its own supporter app over a weekend and see it working before it commits any capital.
This doesn’t remove real production. Design, experience, brand, security, compliance and data still take skill and still matter. But the part that used to need a team assembled just to reach a first version has come down to very little, and builds that ran into the millions are going into production for a fraction of that.
Second, data. I wouldn’t tear out your CRM, but most organisations sit on data they never use, because they never had the analyst hours to go through it. A model with the right context will work through large, messy, inconsistent information and surface the insight or the commercial opportunity that was always there and always out of reach. The constraint was never the data. It was the cost of looking at it.
Third, operations. Any process with several parties, competing systems and tangled workflows, broadcast, ticketing, sponsorship servicing, compliance reporting, you can map it, connect the right agent frameworks and tools, and rebuild it as one platform instead of a chain of handoffs and spreadsheets.
This is what agentic means in practice. It’s more than building faster. What you build can then run itself. The unit of production stops being a person doing a task and becomes a system doing it continuously, across every market and language, on matchday, without someone watching over it.
So the useful way for an executive to hear this is in terms of what you stop doing. You stop commissioning long discovery phases to test whether an idea has legs, because you can build it and look at it. You stop treating “can we even make this” as the ceiling on ambition. And you stop letting data sit idle for want of someone to read it.
But the cost doesn’t disappear. It moves. Off the balance sheet as labour, onto it as compute, and most of all from building to judgement. When anyone can produce almost anything cheaply, the scarce thing is no longer production. It’s knowing what’s worth building, and holding the data, context and relationships that let you build it better than the club down the road working from the same models. The advantage was never in the model. Everyone can buy the model. It’s in what you build around it.
Football has an enormous resource asymmetry. Does AI amplify that gap, or does it open space for organisations that until now could not compete with the biggest players?
My honest view, and it isn’t the diplomatic one, is that this is probably the biggest opportunity smaller organisations have ever had in technology. For the first time I can point to, everyone holds the same capability. The same frontier models, the same frameworks, the same intelligence, available to a grassroots club and a Champions League side on much the same terms. That’s never been true before. Capability used to be something you bought, and only the big clubs could afford it.
That doesn’t mean the gap closes on its own. If you have money and you hire the right people to use it well, you still build an edge, and the biggest clubs will. But the floor has moved far enough that resource is no longer the thing that decides the outcome. Direction is. A small, smart team that understands these tools can now produce work that used to need five or ten times the headcount. We’re watching it across technology, where AI first companies of eight or twenty people reach a scale that used to demand a much larger business. The same thing now applies to a federation, a smaller club, a national body that was always told it couldn’t compete.
And the constraint, everywhere I look, isn’t budget. It’s imagination. Most people don’t know where to point this. The ones who use it best are often the ones who think differently, who push the model, feed it context, and don’t accept the first answer. So the job for a smaller organisation is concrete. Get the leadership properly fluent, in the room and building, not sitting through courses. Turn your own knowledge into models and agents. And use the capability constantly.
If you were advising a football organisation on AI today, where would you tell them to focus?
Start with a diagnosis, not a deployment. Before anyone talks about a model, I want to understand where the organisation actually hurts. Where the work is slow, complex, frustrating, repetitive. Every business carries it, in the commercial team, in legal, in finance, in marketing, in the daily friction of admin, contracts, reporting, sponsorship servicing, reconciliation. That map of the pain is the brief. You point the capability at the places already costing you time, money and patience, and the returns come quickly, because the problem was real to start with.
Then put people in the room and let them build. Open the laptops and have the team stand up their own small agents, apps and solutions to their own problems. The point isn’t that this becomes their job. It’s that they feel, for themselves, how an idea connects to their data and their context and gets solved. And here’s something new. This is the first tool that carries its own instructions inside it. You can sit with it, describe your problem, and it will walk you through the solution, what to connect, what context it needs, what to do next, for as long as you want to go. Nobody has to be trained to start.
What I’d tell them not to chase is the big bang. Don’t sign off a company wide AI rollout before you understand what you’re deploying. The way to think about it is this: the model isn’t a piece of software you install. It’s a resource, a new general purpose input, closer to suddenly having a pool of capable people you can direct. So the question stops being “where do we install AI” and becomes “now that we have this, how would we run the business differently if we were designing it today”.
The last decade has fundamentally changed football: how it is financed, where it is played, how it is consumed. Looking at the next ten years, do you think the impact of AI will make that transformation look modest by comparison?
Yes, but not quite in the way people expect.
Start with what doesn’t change, because it anchors everything.The ninety minutes on the pitch, a result nobody can predict or fake, is the one thing AI can’t manufacture and shouldn’t. In a decade where almost everything else, the films, the music, the advertising, the content we’re surrounded by, becomes generated and abundant, that real, unrepeatable, uncertain live contest doesn’t lose value. It gains it. You can already see the money moving this way, into live rights and live experience, because they can’t be faked. People are pulling back from their screens and paying for things they know are real and shared. A World Cup night, a Champions League comeback, those are real in a way nothing generated will be. So the game itself, the part everyone assumes AI changes most, is where it changes least.
Everything around the game is where the real change happens, and it’s large. The experience layer comes apart and reforms. The single broadcast feed, one picture sold to a set of broadcasters, gives way to millions of individual streams: your camera angles, your commentary, your language, your level of tactical detail, generated live for you. The new world models, trained on physical reality rather than text, finally deliver the immersive experience that VR and augmented reality kept failing to. The thing that was always missing wasn’t the headset, it was the content engine, and now we have it. A fan who can’t be in the stadium will feel close to the match, inside it from any angle. And every fan carries an agent that curates, explains, predicts and transacts, which means clubs hold a direct relationship with each supporter at a scale that wasn’t possible before.
Then the backend, which is where I’d put my conviction, because it’s the part nobody outside the building sees and the part that changes most. Performance becomes systematic: recruitment, scouting, injury prediction, individual training, the tactical edge made repeatable. The operating model turns over. A club runs from a small, sharp core, with agent systems carrying the commercial servicing, the content, the operations, the legal and the finance that used to need whole departments. And the lasting advantage, as I keep saying, isn’t in the model, because every rival can buy the same model. It’s in the context each club owns and directs: its performance data, its fan data, its history, its relationships. The club that turns that into capability builds something rivals can’t copy. The club that doesn’t will watch its rivals do it.
For the league, two things move. The unit of rights itself is in question, because once distribution is direct and personal, the old model of one feed sold on starts to break down, and that reshapes the commercial spine of the sport. And integrity becomes a front line job, not a back office one. In a world of convincing fakes and manipulated clips, the league’s work grows to include proving what happened on the pitch.
So the last decade changed how football is financed, played and watched. The next decade changes what the football business is, while leaving the ninety minutes almost untouched, and in fact making them worth more. Bigger than people expect on the business and the experience. Smaller than they expect on the game.
So the game itself, the part everyone assumes AI changes most, is where it changes least.
Football sits at the intersection of business and entertainment. How do you see AI reshaping both sides of that equation, the way the industry operates and the way people experience the game?
Start with something people don’t like to say. Football is, at its core, a data business. The game is driven by data, the commercial engine runs on it, sponsorship is priced and sold on it. And yet football has rarely been run like a data business. Look at Formula One for the contrast. It’s obsessive about engineering, telemetry, insight and operational precision, a culture of technical experts working off hard data. That rigour was expensive, and for a long time football either couldn’t or wouldn’t make that investment on the business side. A few elite clubs and leagues managed it, because they could afford the people, and even they are dealing with enormous complexity. For most of the sport the capability was out of reach, and the people running these organisations never fully embraced technology where it counted, so the returns never came.
What AI changes is the price of that rigour. The kind of capability Formula One has, taking large, messy, fragmented information and turning it into operational intelligence, is no longer something only the wealthiest can buy. For twenty years the sport built complicated software systems and then didn’t resource them properly, so the data sat there unread. That’s no longer the constraint. Football can finally be run like the data business it always was.
AI can generate film, music, video games, immersive experiences. Is sport immune to that? Is the unpredictability of the result the last asset a machine cannot manufacture?
Sport isn’t immune, and it doesn’t need to be. I was in Cannes last week, and the mood in the creative industries has turned. The fear has given way to a clearer view: generative AI is a new tool in the kit, very good for prototyping, production and pushing what’s possible, and it doesn’t remove the human part, the taste and the judgement. Every creative field has always taken on new tools. Sport will take on this one too, across production, content and the fan experience, and it’ll be better for it.
But there’s a hard line, and it’s at the result. The moment nobody scripted, the goal, the comeback, the outcome you didn’t see coming, can’t be synthesised. The instant you generate it, it stops being sport and becomes a game or a fiction. Nobody cares about a result they know was written in advance. So yes, the uncertain outcome is the one thing a machine can’t manufacture, and that’s exactly why it becomes more valuable as everything around it becomes abundant and synthetic.
What follows matters more than the observation. If the real result is the scarce thing, the industry’s job is to protect it, its authenticity and its scarcity, not dilute it. Generative tools should sit around the live event, making it richer to produce and experience: personal replays, immersive recreations, fans making their own content from real archives and statistics. That’s a good new layer of entertainment, and it’s welcome. The discipline is knowing the difference between what deepens the meaning and what’s just a gimmick. Sport is full of technologies that felt exciting for a season and faded because they added novelty, not meaning. The test doesn’t change: what do people care about, and what still holds quality and truth. Protect the real thing at the centre, let AI make everything around it richer, and be clear about which is which.
Is there anything you have refused to delegate to AI?
The thinking. That’s the line, and I hold it on purpose, because it’s subtle and it matters more than anything else in how you work with these tools.
The failure most people are drifting towards isn’t dramatic. You throw everything into the model, let it produce the answer, and then trust it wholesale. Do that often enough and the relationship turns around. The AI is no longer a capability you direct, it’s leading you, and you’ve stopped noticing. It’s easy to end up there, because the models are good enough to make it comfortable.
Have you used AI to respond to these questions? To what extent?
Yes. Completely, and on purpose.
You sent a list of questions in a document, to be typed into. Reading them, I could see the hours of typing ahead. So I did the thing I’ve argued for across every answer here. I gave the questions to Claude and asked it to interview me: to put each one to me in turn, let me talk through my thinking, and then synthesise my own thoughts, logic and beliefs into clean, quotable answers. Every idea and every position in this document is mine. What the machine did was take them from a rough spoken answer to something sharp enough to read.
And it was a better process, not only a faster one. Going question by question, the model held the context of everything I’d already said, so each answer could build on the last and connect across the whole set. It knows how I think, so it could push and shape while I did this in the gaps between the fifty other things I’m building. The document would have given you worse answers and taken far longer.
So this is the argument, done rather than described. I didn’t tell you I work with AI as a thought and production partner. I did it, in front of you, to answer your own question. The thinking stayed mine. The synthesis, the structure, the artefact you’re reading, that was the machine, directed by me. That’s what I mean when I say the value isn’t in the model but in what you do with it.
And if you want, next time we go one better. We build you an agent that runs the interview itself, so the next speaker just talks, and it does the rest.