Uses and impact

What can we use AI for?

If you believe the sales pitches, it’s everything. But of course, that’s not true.

For starters, as set out in the What is AI? section, AI is not one thing. It covers a wide range of technologies and research fields. Whilst Deep Learning can help with the analysis of medical scans and detect cancerous growths, and optimisation systems can provide you with the best route to the airport, what most organisations are considering now are generative and agentic AI technologies. Products like ChatGPT, Co-Pilot, Claude and Gemini and their associated agentic platforms.

AI could be adopted in organisations through various approaches. In some cases bespoke AI tools might be developed in-house, while others may be procured from external vendors, and in many cases, organisations deploy off-the-shelf publicly available models. Whilst we acknowledge that some membership organisations may well be data mature and infrastructurally sound enough to be considering developing their own AI systems, for this knowledge base, we will focus on the adoption and deployment of existing AI models, specifically generative and agentic AI systems and tools.

In practical terms, these technologies serve two distinct but connected business functions – generating content and carrying out tasks. While the promised potential is significant, it is important to note that the long-term evidence for these benefits is still yet to emerge. Much of the current data comes from early adopters or controlled pilots and the sustainability of claimed efficiency and productivity gains over years of operation are still to be proven.

Generative AI can act as a tool for tasks that require creating content from scratch or synthesising large amounts of information. This could include:

Then Agentic AI models are promoted as being able to take action on your behalf. Agentic AI systems don’t need prompts but are instead given a goal and are left to execute the necessary steps to achieve that goal across different systems. They can be used for:

It all sounds great right? But they all do come with some major caveats. Adopting these tools requires a fundamental shift in governance and risk management. With Agentic AI taking actions rather than just making suggestions, organisations need to adopt human-in-the-loop or human-in-control models. You must establish clear guardrails, define approval thresholds for high-stakes decisions and implement continuous monitoring. Agents could achieve a goal but in an unintended, unethical and costly way.

While early reports suggest procession accelerations and efficiency gains of high percentages, organisations need to approach these claims with caution. The lack of a long-term evidence base means that hidden costs – such as energy consumption, the expense of needing to review and correct AI errors at scale, the consequences of de-skilling (the erosion of human expertise through the use of AI) and cognitive offloading (delegating cognitive tasks to AI) – are not fully understood. Successful adoption is a misnomer as from adoption you can only claim short term success while long-term success will depend on treating the implementation journey as an ongoing process of review and monitoring through rigorous assessment of impacts and outcomes.

A key question to keep in mind when considering efficiencies and productivity gains, is who benefits and at whose cost? If implementing these technologies helps one team work quicker but adds more work to another team, for example through checking and reviewing AI outputs, then these gains may be cancelled out? If using AI cuts the time to write a report in half but doubles the time needed to review it, are you really gaining anything?

What about specifically in sport?

The hyperbole around AI in sports is very real. Claims of enabling transformation at a “blistering pace”, “revolutionising” the sector, “unleashing” innovation… all the usual overblown hype is evident. But is there truth beyond the headlines?

The use of data driven technologies in sports is not new. Many of us have seen Moneyball which highlighted how statistical analysis could identify undervalued players and optimise team strategy long before the term "artificial intelligence" entered the mainstream lexicon. The current integration of generative and agentic AI is less a revolution and more an evolution of this existing framework, shifting from retrospective reporting to real-time processing and automated execution. Instead of relying solely on historical spreadsheets, teams can now use AI to process live feeds from player tracking sensors, video, and biometric monitors, turning raw data into immediate tactical suggestions or operational decisions.

In practical terms, generative AI can assist coaches by summarising large volumes of opponent footage into tactical briefs, identifying patterns that might take human analysts significantly longer to find. It could also automate the creation of content for fans, such as customised highlight reels. Agentic AI can handle logistical tasks, such as coordinating travel and equipment maintenance.

However, these tools also enable practices that raise ethical concerns. Dynamic pricing algorithms, for instance, can adjust ticket costs in real-time based on demand and opponent strength, potentially making attendance unaffordable for traditional supporters. Similarly, biometric data collected for player safety can be repurposed to assess long-term health risks during contract negotiations, creating a power imbalance where clubs possess detailed health insights that players may not fully control or understand.

The deployment of these technologies also presents strategic and integrity challenges. The "black box" nature of some AI models means that the reasoning behind a specific tactical suggestion or pricing change is not always transparent, which can erode trust among coaching staff and fans.

There is also the risk of algorithmic bias, where systems trained on historical data might perpetuate existing inequalities in scouting or recruitment. The use of AI to inform decision making has often been justified as enabling “objective” or “value-free” decision making based purely on data and analytics. However, the reality is that if this data contains biases or inaccuracies or is not up to date, the decisions it informs will be likely to be biased or inaccurate. In this way, far from being value-free or neutral, AI systems can in fact entrench and conceal existing biases. If the use of AI is not transparent, and the ways this has informed decisions are not fully explained it can be difficult for an impacted individual to challenge or appeal decisions. This could have very real implications, for example if an athlete wasn’t selected for competition based on predictions made by an AI.

Overreliance on AI systems risks deskilling critical sporting roles, where human intuition and experience are gradually eroded by automated insights. Coaches may lose the ability to read game dynamics independently if they depend solely on algorithmic tactical suggestions, while scouts might overlook intangible traits like resilience or leadership that data fails to capture. Similarly, medical staff could become overly reliant on AI injury predictions, potentially dulling their clinical judgment and hands-on assessment skills. Ultimately, if AI becomes the primary decision-maker, the workforce risks becoming mere validators of machine output, losing the deep expertise required to act when technology fails or provides ambiguous guidance.

Consequently, the value of AI in sport depends less on the technology itself and more on the governance surrounding it. Organisations that establish clear ethical boundaries - ensuring AI supports rather than overrides human judgment regarding player welfare and fan accessibility - are better positioned to avoid the reputational and operational risks associated with unchecked automation.

Shadow AI

Shadow AI refers to situations where employees are using unauthorised AI tools to process data or to carry out tasks without the visibility or control or remit of their organisation’s policies which can lead to potential security blind spots and risks. This can lead to serious data breaches e.g. a staff member putting someone’s personal data into ChatGPT or feeding confidential client data into Claude when their organisation doesn’t allow the use of/or has no policy for the use of these tools.

To mitigate the risks of shadow AI it is important that organisations have clear policies and guidance on acceptable use of AI tools, and that these are communicated to staff members in accessible formats.