In Football, AI Sees Better Than in Our Hospitals

Premier League clubs deploy on average twelve high-definition cameras coupled with artificial intelligence video analysis systems per stadium. In NHS hospitals of comparable budget size, fewer than one in five radiology departments has a fully integrated AI diagnostic aid tool, according to NHS England data for 2024. Both technologies rely on the same convolutional neural network architectures. One refines pressing tactics. The other could detect lung cancer at an early stage.

This gap is not a technological accident. It is the product of a logic of incentives that directs progress before engineers have even opened their computers.

The Essentials

  • The global market for AI video analysis in professional sports is growing at an estimated annual rate of 25 to 30%, according to data from the MIT Sloan Sports Analytics Conference, compared to 15 to 18% for AI-assisted imaging in European hospitals.
  • In England, Liverpool FC, Manchester City and Arsenal deployed automated video analysis systems as early as 2019-2020; equivalent adoption in French university hospitals and British NHS Trusts remains partial as of 2025.
  • The main brake on the medical side is not technical but institutional: CE and FDA certifications, medical liability, hospital purchase cycles of three to seven years.
  • If the technological maturity gap is maintained at this rate, sports tools will have accumulated a decade of operational advantage over their medical equivalents by 2030, with a measurable social cost in delayed diagnoses.

Professional Sport as a Regulatory Accelerator

European football adopted AI video analysis at a speed that no regulated sector has been able to replicate. The reason is simple: nobody dies if an algorithm makes a mistake about a passing angle. The absence of vital risk eliminates at once most of the friction that slows adoption in medicine, aviation or energy.

Stats Perform, Signality and Tracab equipped most European first division clubs between 2018 and 2022. These systems track in real time the position of each player twenty-five times per second, calculate hundreds of running and space metrics, and produce automated tactical recommendations. The Bundesliga imposed optical tracking as a standard in all its stadiums as early as 2021. Ligue 1 followed in 2022. La Liga and the Premier League were already operating in advance.

The entry cost collapsed in parallel. A complete AI video tracking system for a second division club costs today between 80,000 and 150,000 euros per season, according to data published by the MIT Sloan Sports Analytics Conference. In 2015, the same level of precision would have required an investment exceeding one million euros and several full-time human analysts. The curve continues to decline.

The Technological Building Blocks Are Identical, the Incentives Are Not

A cardiologist examining an electrocardiogram and a video analyst evaluating the movement of a midfielder mobilize, on the machine side, comparable architectures: neural networks trained to recognize patterns in streams of structured data. Google DeepMind, which developed AlphaFold for protein biology, is also present in sports analysis through its partnerships with the Premier League. Nvidia supplies GPUs to both sectors.

The divergence lies elsewhere. In sports, the purchasing decision belongs to a sporting director or an owner who can sign a contract within a few weeks. In a European public hospital, the same type of software must pass through a Class IIb CE certification (which takes on average eighteen months according to the EMA), an ethics committee, a public tender, negotiations with health insurance on reimbursement, and mandatory training of medical staff before deployment. Each step is justified. Their accumulation produces a structural delay.

Data from the British Journal of Sports Medicine document another phenomenon: clubs have been using their AI analysis tools for injury prevention for several years, by modeling early signals of muscle fatigue and joint overload. These same signals, measured in non-athletic patients in a medical context, would require two to five years of clinical validation to be used as a certified diagnostic aid tool.

What Clubs Have Learned That Hospitals Have Not Yet

The advantage of sports over medicine is not merely regulatory. It is cognitive: clubs have accumulated usage data, corrected errors, and refined models over millions of hours of real game play. This operational experience is a form of capital that hospitals have not yet built for the medical equivalents of the same tools.

Liverpool FC regularly publishes, through its internal research department, methods for evaluating training load based on AI. These publications directly feed into sports medicine research. The British Journal of Sports Medicine devoted a special issue in 2023 to the question of the migration of these tools to clinical practice. The authors noted that algorithms developed to estimate midfielder fatigue could, with minor adaptation, improve the monitoring of patients in cardiac rehabilitation.

Here begins the useful trajectory. Several companies have identified this vector of diffusion and are deliberately exploiting it. Catapult Sports, founded in Australia and whose sensors equip more than 3,000 professional teams worldwide, launched a health division in 2023 targeting rehabilitation centers. Kinexon, official supplier to the NBA and several Bundesliga clubs, is testing its motion sensors in two German university hospitals for post-operative monitoring of orthopedic patients. This is not a secondary application: it is the business model for the next decade for these companies.

Amateur Sports Medicine as a Diffusion Pathway

Between the stadium and the hospital, there exists an intermediate space that industry players have identified as the true terrain for diffusion: amateur and semi-professional sports medicine. National 1 and National 2 clubs in France, regional leagues in Germany, training academies: these structures have doctors and physiotherapists, and a real need for injury prevention tools, without the regulatory constraints of a public health facility.

Prevent Biometrics, an American startup, has been commercializing since 2021 a connected mouthguard that measures the impact of head injuries and transmits data to a doctor via an application. The tool was born in sports and is used in American high schools. In Europe, equivalents are being developed for rugby and football. The boundary between sports tool and medical device becomes blurred, which is precisely the zone where regulation will have to make a decision.

This is not the first time a consumer or sports technology has outpaced institutional medicine. The pulse oximeter was commercialized for sports and the general public well before being integrated into hospital protocols. The connected blood pressure monitor followed the same trajectory. In both cases, medicine ultimately adopted what mass consumption had validated. The question is whether the same pattern repeats for AI analysis tools, and at what pace.

This points to a broader dynamic that the journal has had the opportunity to examine in other sectors: the speed of diffusion of an innovation depends less on its technical quality than on the institutions that receive it. Professional football is a light, fast institution financially incentivized to experiment. The public hospital is a heavy, slow institution incentivized not to be wrong. Both logics have their own rationality.

The Technological Maturity Gap and Its Social Cost

If one projects the current development curves, AI analysis tools in sports will have accumulated a decade of real operational experience over their medical equivalents by 2030. This advance translates concretely: models trained on more data, interfaces refined by millions of users, known errors that have been corrected.

The question is not abstract. A tool for early detection of musculoskeletal anomalies that could have been deployed in hospitals in 2025 might be deployed in 2030 or 2032. In that interval, some diagnoses arrive later than they might otherwise have. The evaluation of this social cost is difficult to quantify precisely, but researchers from King’s College London and Inserm are working on models to estimate diagnostic delays linked to AI medical device certification delays. Initial results, expected for 2025-2026, should feed into political debate on overhauling the CE procedure for medical software.

This debate is underway. The European AI Office, created as part of the European regulation on AI that came into effect in 2024, is responsible for coordinating national approaches to AI medical devices. The European Commission is examining a revision of the medical devices regulation to accelerate certification timelines for AI diagnostic software considered low-risk. Industry is pushing hard. Patient associations are divided: some want acceleration, others fear that haste will reproduce the errors of early certifications of connected medical devices, several of which have been withdrawn from the market after serious incidents.

The question of who captures the gains from technological progress and according to what logic of incentives is at the heart of this debate. Innovation economists, from Daron Acemoglu to Philippe Aghion, have shown that the direction of technical progress is not neutral: it follows price signals and market structures. Professional football sends strong and rapid purchasing signals. Public health systems send slow and constrained signals. Engineers and venture capital investors respond to this difference in signal, not to the difference in social utility.

What the Trajectory Says About the Diffusion of Progress

The case of AI video analysis in sports is not isolated. It belongs to a family of dynamics where lightly regulated sectors serve as experimental terrain for technologies that eventually, with delay, reach sectors of high social utility. Drones were massively developed for the military and leisure before being used for crop surveillance or inspection of power lines. GPS chips equipped cars and phones long before rural ambulances.

What is new with AI is the speed of the improvement cycle. A computer vision model trained on ten million images of football gameplay improves much faster than a model trained on a certified radiological dataset, smaller and slower to constitute. The maturity gap does not mechanically close over time: it risks widening if the conditions for producing training data remain so different.

Several initiatives seek to correct this asymmetry. The European EUCAIM program (European Cancer Imaging Initiative), launched in 2023, aims to build an infrastructure of medical imaging data on a continental scale to train diagnostic AI models on volumes comparable to those available in other sectors. More than 70 European hospitals participate. The stated objective is to provide by 2027 a training database of more than 100,000 annotated cases for several types of cancer. This is a structural response to a structural problem: if data is missing, build the data.

Professional sports medicine itself is beginning to play a bridging role. Doctors who work with Premier League or Bundesliga clubs have a dual culture: they know the AI analysis tools of sports and medical protocols. Several of them publish in medical journals on the transposition of these tools. This community of hybrid practitioners could accelerate technological migration where institutional processes are stuck.

The real question posed by this gap is not whether medical AI will catch up with sports AI. It will, as pulse oximeters and GPS did before it. The question is whether European health systems can design accelerated certification pathways for low-risk tools, while preserving rigor where it is essential, or whether they will let another decade pass before what improves the pressing tactics of a Bundesliga club contributes to detecting an anterior cruciate ligament tear in a teenage amateur club player, or a lung mass in a 55-year-old patient in general practice.

Sources

  1. MIT Sloan Sports Analytics Conference — https://www.sloansportsconference.com/
  2. British Journal of Sports Medicine — special issue on the migration of AI analysis tools to clinical practice (2023) — British Medical Journal Group, no stable URL verified
  3. NHS England — data on adoption of diagnostic AI in radiology departments, 2024 report — NHS England, no stable URL verified
  4. European Cancer Imaging Initiative (EUCAIM) — Horizon Europe program, Cordis — no stable URL verified
  5. European regulation on artificial intelligence (AI Act), Official Journal of the European Union, 2024 — https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689
  6. Kinexon Sports & Industry — data on deployments in professional sports and tests in university hospitals — kinexon.com, no stable report URL verified
  7. Catapult Sports — annual report and health division — catapultsports.com, no stable report URL verified