From around 2010, sports organisations around the world made the same set of decisions. They opened accounts on Facebook, YouTube, and Instagram. They migrated their fan engagement to platforms they did not own, built audiences in someone else’s garden, and handed over the behavioural data that resulted. By the time most organisations understood what they had given away, the platforms owned the relationship and the data, and reclaiming either was expensive and slow.
That cycle is starting again. The mechanics are different, but the underlying pattern is the same: a new set of technology platforms offering extraordinary reach and capability, and a set of decisions being made right now about how much access to grant them to data, content, and commercial relationships. The difference is that this time, most sports organisations can see it coming.
The question Craig Hepburn posed on a recent Unofficial Partner podcast is the right one: is sport thinking about this differently, or is it going to walk through the same door again?
What the social media decade actually cost
The cost of the social media decade was not immediately obvious, which is part of why it was so effective. Platforms offered reach in exchange for data and engagement, and the reach was real. Fans turned up. Content spread. Clubs got followers in the millions. The problem was structural: none of the value generated by that reach accrued to the club in a durable way.
The data generated by a fan engaging with a club’s content on Instagram belongs to Instagram. The targeting intelligence built from millions of interactions went to Meta’s ad stack, not to the club’s commercial team. The behavioural signals that could have informed hospitality pricing, partnership valuations, or direct subscription products were captured by a third party whose business model depended on keeping them.
Most clubs now have sophisticated CRM programmes and data strategies specifically designed to rebuild the direct relationship that the platform decade eroded. To be fair, there was not really a choice in 2010. Getting on the platforms was commercially rational. Opting out of Facebook when your fans were on Facebook was not a serious option. The extraction was the price of admission, not a failure of judgement. What makes the AI transition different is that there is a genuine choice this time, if you move early enough to exercise it.
The AI transition is moving through the same playbook
The AI layer is being built now, and the access decisions being made in the next two to three years will determine who captures the value from it.
The structural dynamic is familiar. The large foundation models have been trained on publicly available content. They are now looking at the next frontier: proprietary data inside organisations, including sports organisations. That data, covering historical ticketing patterns, fan behaviour, hospitality purchase history, content engagement signals, and partner and sponsor records, is enormously valuable for training and for grounding AI products in specific commercial contexts.
The AI in sports market is growing at close to 30% per year across most independent forecasts, driven increasingly by products built on proprietary sports data. The question is who captures the commercial value: the AI platforms who build products with it, or the sports organisations who generate it in the first place.
When a league does a deal with Google today, a broadcast partnership, a distribution agreement, a data licensing arrangement, what exactly is being exchanged? Hepburn’s framing is useful here: Google does not just have YouTube. It owns the full stack, from its custom TPU chips up through the model to the product. Every piece of content, every search signal, every engagement data point feeds a training and grounding infrastructure that is unique in scale. A league signing a distribution deal is not just signing a distribution deal. It is potentially contributing training data to a system whose primary asset is the intelligence that data creates.
That is not an argument against doing deals with large technology companies. It is an argument for understanding what the deal is, structuring it accordingly, and retaining ownership of what is most valuable.
Building your own infrastructure is not anti-technology. It is the most commercially intelligent move available.
The counter-intuitive point Hepburn makes is this: organisations that build their own data infrastructure, their own APIs, and their own rules around how that data can be accessed do not become more closed. They become more in control of how they open up.
The distinction matters. A club that has built a proper data platform can choose to expose specific data products to AI models, to third-party developers, to agency partners, and to fans who want to build their own applications, on its own terms. It defines what can be accessed, how, at what cost, and under what conditions. It can monitor usage, manage the commercial model, and build new products on top of the same infrastructure.
A club that has not built that infrastructure is dependent on the platforms it works with to set those terms. That is a weaker commercial position, and it becomes weaker over time as the platforms accumulate more capability and the club’s leverage diminishes.
The model Hepburn describes is not hypothetical. Shopify, Bloomberg, Salesforce and Box have all opened their data to AI agents on their own terms, not by handing data to a platform and accepting whatever commercial arrangement follows, but by building infrastructure that allows them to define the rules. The sports world has the same option.
There is also a direct parallel with intellectual property control. Sport has always policed rights contracts carefully: who has access, to what content, in which markets, under what conditions. The AI infrastructure question is the same conversation, applied to a new layer of the commercial stack.
What building the infrastructure actually involves
This is where the abstract argument needs to connect to something practical.
The foundation is a data architecture that consolidates the club’s or venue’s proprietary data in a form that can be used, accessed, and governed. This is not primarily a technology project. It is a decisions project: what data do we have, what is it worth commercially, and what access are we prepared to give to whom under what conditions.
On top of that foundation, the priority is connecting that data to AI intelligence through an API architecture that allows products and services to be built without exposing the raw data. The club owns the data. The AI model, whether that is Claude, GPT, Gemini, or an open-source model, provides the intelligence. The product is built at the intersection. The club retains control of what the model knows, how it is used, and what the commercial arrangement looks like.
Why how you access AI matters as much as whether you use it
This is a practical point that most commercial teams have not registered, and it matters significantly for organisations with sensitive data.
If someone on your team pastes a hospitality client list or a draft partnership agreement into the consumer version of ChatGPT, that data has entered OpenAI’s product environment and may be used to improve the model. The same applies to free and personal subscription tiers across most frontier models.
When an organisation accesses the same underlying model through a commercial API, paying for tokens and building its own application on top, the terms of service explicitly exclude that data from training. Your inputs are used to generate the response, then discarded. They are not fed back into the model.
So the choice of how you access AI is not just a technical or commercial decision. It is a data governance decision. For a Premier League club or a national stadium working with sensitive commercial data, it is one worth making deliberately and communicating clearly to the team.
The agent framework layer
The third element is the agent framework layer: the infrastructure that allows AI agents, both internal and eventually external, to act on behalf of the organisation in commercial contexts. Hepburn’s framing points toward this most directly. The fans who will engage with sport in five years will increasingly have agents acting on their behalf. Those agents will need somewhere to connect to. The organisations that have built the right infrastructure will be the ones those agents engage with productively. The organisations that have not will be dependent on intermediaries.
The operating model shift this requires
None of this is possible without a shift in how elite sports organisations think about their own role. Hepburn’s argument is direct: you are already in a technology business, because every business is accessed to some degree through a technological platform. That is accelerating, not reversing.
This does not mean every club needs to become a software company. It means thinking about technology infrastructure the way commercial directors already think about stadium assets: as something that generates long-term value, requires active investment, and cannot be left to depreciate without cost.
The clubs that made that shift early in the data analytics era, building internal data science capability, investing in proprietary modelling, treating commercial data as an asset rather than a byproduct, have a measurable commercial advantage today. The same dynamic applies at the AI layer, with a shorter window before the gap between early movers and late movers becomes hard to close.
The organisations that navigate this well will do three things. They will build the data infrastructure that gives them control over their most valuable proprietary assets. They will invest in the technical capability, internally or through a close partner, to connect that data to AI intelligence in a commercially intelligent way. And they will make deliberate, documented decisions about access: who can use their data, in what products, under what commercial arrangement.
The choice is available now, but the window is finite
The good news is that the window has not closed. The AI infrastructure layer is being built now. The data decisions are being made now. The cost of building something robust, agent frameworks, data APIs, the governance layer around them, is lower than it has ever been. Unlike the social media era, where the capability genuinely required platform scale to access, the AI tools required to build this infrastructure are available to organisations of all sizes.
At Earl , the infrastructure and control question is at the centre of how we work with clubs and venues. The AI Visibility Diagnostic gives a clear picture of the current state: where your commercial operation sits today in terms of AI visibility and infrastructure, what the gap is costing, and what the priority steps are. But the diagnostic is a starting point. The more substantive work is helping organisations build the data architecture, the API layer, and the commercial governance framework that gives them genuine control over their position as the AI layer matures.
Sport came into the social media era with extraordinary assets: passionate audiences, premium content, scarce commercial inventory, and decades of proprietary data. It gave most of the value generated from those assets to the platforms it worked with. The AI era offers the same assets and the same choice. The difference is that this time, most of the organisations involved have already lived through the consequences of getting it wrong.
That is not a comfortable observation. But it is a useful one. The pattern is visible, the tools to respond to it exist, and the decisions that will determine the outcome are being made now.