6.1 Combining AI, wearable and video data
Figure 6.1 shows the flow. Each source passes through a quality check by its data owner before it enters the player profile. From the profile come two things: a short report to the coach, and a decision taken by a person.
Three rules for combining:
- Keep the sources separate in the profile. A GPS distance, an AI score and a tag count are different kinds of evidence. Show them side by side; do not add them into one "overall score".
- Check each source on its own terms. Signal quality for wearables (section 4.2), data completeness for AI (section 3.2), tag agreement for video (section 5.2).
- Do not mix systems. Numbers from different brands or systems are not comparable (section 4.3). The same applies to AI scores from different model versions and to tags from different definitions.
D2.3's Lithuanian cases (Cases 10-12) describe profiles that combine video, monitoring and emerging AI tools. They also note that some of this integration was still in development at the time of writing. Distinguish what you use now from what you plan to use.
6.2 Reading a performance dashboard
Most systems offer a dashboard. Use it in this order:
- Read the data quality indicators first. Missing sessions, poor signal, low completeness. If quality is poor, stop here for that player.
- Look at the trend. One high number after one session means little; four weeks in the same direction is worth acting on.
- Find out how each "score" is built. Suppliers' composite scores (load scores, readiness scores, talent scores) are formulas you did not write. Ask for the formula before you report the number (Torres-Ronda et al., 2022).
- Turn it into one message. A few key numbers per player and one clear sentence, soon after the session (Torres-Ronda et al., 2022). The dashboard serves the staff; the sentence serves the coach and the player.
- Treat every number as an estimate with error (Torres-Ronda et al., 2022).
6.3 Individual baselines and appropriate reference groups
The most useful comparison for a developing player is with their own earlier data. Build a baseline for each player over the first weeks of a season, and read later values against it. D2.3 (Cases 6 and 7) found that tracking change per player, rather than team averages, served development decisions better.
Group comparisons still have a place, but choose the group with care:
- Same age group, same system, same season. A group that differs in any of these is not a fair reference.
- Account for age and maturity. In youth players, read physical data in relation to age, growth stage and fitness (Torres-Ronda et al., 2022). For AI outputs, check by birth quarter and maturity status (Hill et al., 2020).
- Use position groups for tactical video, not the whole squad.
A baseline cannot tell you who will be injured (Bahr, 2016) or who will be selected (Altmann et al., 2024). It shows whether this player is changing, and in which direction.