Personalisation in sport is mostly a marketing word. What is usually sold as individualised training is an athlete being compared with a population and told where they sit against it. That is a useful starting point and a poor finishing one, because the population never had this athlete’s injury history, this athlete’s exam week, or this athlete’s target race in nine weeks.
Genuine individualisation needs something the sensor cannot supply: a durable model of who the athlete is, what they are training for, and what constrains them. That is what we call the athlete context graph, and building one is more of an organisational discipline than a technical project.
Why a population model is not individual understanding
Population models answer the question what usually happens. Coaching requires the question what is likely for this person, now. The gap between those two is where most automated recommendations quietly fail.
It matters more for some athletes than others. The evidence base underpinning most sports science is thinner for women, thinner again for para athletes, and thin for the majority of events outside sprints and the marathon. An athlete outside the well-studied middle is being compared with a population that barely contains them, and a system that does not say so is misleading its user.
The layers of an athlete context graph
Five layers, in descending order of how often they change.
- Identity and objective. Event, role, competitive standard, the target that defines this season, and what the athlete themselves says success looks like.
- History. Training completed rather than planned, injury and illness episodes, previous responses to load blocks, and what has worked before.
- Availability and constraint. Work, study, travel, caring responsibilities, medical restrictions, and the parts of life that determine whether a plan is realistic.
- Schedule. Competition calendar, selection windows, and the fixed points everything else bends around.
- Signal. The fast-moving daily data: load, sleep, wellness, and whatever the wearables report.
Most platforms are excellent at the fifth layer and almost silent on the first four. That is the wrong way round, because the top layers are what give the bottom one meaning.
Keep four kinds of statement apart
A context graph is only trustworthy if it never blurs these four.
- Measured fact. The session was completed at this time, for this duration.
- Model estimate. The system infers a training strain of a given value, using assumptions it should disclose.
- Athlete report. The athlete says their legs felt heavy. This is data, not noise, and it is not a measurement.
- Coach interpretation. The coach believes this reflects the travel rather than the training.
When these four are displayed identically, a plausible guess acquires the authority of a stopwatch. Keeping them visually distinct is the single cheapest integrity improvement available to any performance system.
How context changes what the same number means
Take two 800 metre runners in the same training group. Both complete an identical session and both record the same internal load value.
The first is 12 weeks into an uninterrupted block, sleeping well, with no competition for two months. The second returned six weeks ago from a stress response, has an important trial in 11 days, and is in the middle of assessments. The number is the same. The decision is not remotely the same, and no readiness score that lacks the top four layers can tell them apart.
This is the practical case for context. It is not about knowing more. It is about knowing enough to stop treating identical measurements as identical situations.
Athlete access, correction and consent
If a system holds a model of an athlete, the athlete should be able to see it, question it and correct it. This is partly a legal position, since access, rectification and erasure are existing rights under general data protection law rather than concessions. It is also a performance position: athletes who cannot interrogate a system will not act on it, and athletes who can will improve its accuracy for free.
More context also means more responsibility. Collect only what serves a stated performance purpose, and define access, retention, correction and deletion before deployment rather than after a complaint. A context graph assembled without those rules is a liability wearing the costume of an asset.
Where the evidence stops
Richer context improves relevance. It does not confer predictive power, and it is worth being blunt about that because the temptation runs the other way. Individual injury prediction remains weak on the published evidence, and adding life context to a model that could not predict does not turn it into one that can. Adaptive plans and personalised workouts from consumer platforms are product features whose derivations are largely undisclosed, and treating them as prescriptions rather than suggestions is a category error that a well-built context layer should make harder, not easier.
The question to take into your next performance meeting
For your three most closely managed athletes, could you produce today, in under five minutes, the four layers above the daily signal, and would the athlete agree with what it says about them?
Frequently asked questions
What is an athlete context graph?
A longitudinal model of an athlete that connects stable information such as objectives, history, constraints and competition schedule to fast-moving daily signals such as load, sleep and wellness. Its purpose is to give the daily numbers meaning, so that the same measured value in two athletes is not treated as the same situation.
Why is personalisation in training more than comparing to averages?
Because an average describes what usually happens in a population, and coaching decisions concern a specific person on a specific day. The gap widens for athletes underrepresented in the research, including women, para athletes and competitors in less-studied events, where population comparison can be actively misleading.
Should athletes be able to see the data held about them?
Yes, and in most jurisdictions they already have the right to. Access, rectification and erasure exist in general data protection law. Beyond compliance, visibility is what makes the model more accurate, because the athlete is the only person who can correct what it has wrong.
Does more athlete context improve injury prediction?
Not on current evidence. Individual injury prediction performs poorly in published testing, and additional context does not repair a model that lacks predictive validity. Context improves the relevance and fairness of a decision, which is a different and more achievable claim.
What data should a performance programme not collect?
Anything without a stated performance purpose, anything the athlete has not been told about in plain language, and anything the programme cannot commit to retaining, correcting and deleting under a defined rule. If nobody can name why a field exists, it should not exist.
About this series
This is the third of seven articles on the movement from isolated measurement to connected, coach-led decision support in high-performance sport, following the Athleet AI cycle of observe, understand, decide, deliver and learn. It builds on the coach as system designer and leads into how video and physical data converge. The governance framework underneath the series is set out in The Responsible Performance System.
Every club, athlete and scenario used as an illustration in this article is desensitised. Product capabilities are attributed to their vendors and should not be read as independently validated performance benefits. Nothing here constitutes legal or medical advice, and the material is presented to support discussion and further review.