From prediction to execution: a narrative review of artificial intelligence tactical models in badminton, player physiology and coaching use

Machine learning models predict badminton strokes, but the practical value of tactical recommendations depends on whether players can execute them. Player capacity can change within matches and across successive tournament matches. Based on structured searches of eight bibliographic sources, this narrative review synthesizes tactical modelling and badminton physiology to examine the evidence needed for executable recommendations and coaching use. We distinguish five types of claim (description, prediction, prescription, adoption and outcome), each requiring a different form of validation. Most tactical studies focus on prediction or the development of coach-facing interfaces. Amongst tactical studies with assessable inputs, 30% included match-context variables, such as score, but no explicit measure of player state. Very few used explicit physiological variables, and none reported fitting or validating models using measured player state during match play. Beyond developer-led evaluations, our search identified no study in which practitioners received model-derived recommendations and an outcome was recorded. Physiological studies document metabolic demand, neuromuscular fatigue, perceptual changes under load and cumulative effects across successive matches. Several report maintained performance, so whether fatigue alters feasible or optimal tactics remains a testable hypothesis. We distinguish candidate inputs for use during play, principally video-derived estimates of player-specific movement load, from variables suited to research and post-match monitoring. We also propose a reporting checklist separating information studies can report now from requirements that need new data. This framework specifies the evidence needed to evaluate whether recommendations are executable, whether coaches adopt them and whether measurable outcomes follow.