Most existing game analysis attempts do not consider situational context information sufficiently for the assessments of tactical behaviour. The purpose of this case study was to apply a new analysis approach, so far only used for 1-versus-1 analyses in soccer, to small-sided games and to identify and analyse therewith the efficiency of tactical patterns in 3-versus-2 game play. Therefore, 279 games, performed by an elite youth soccer team, were recorded by video and a position tracking system. The data was processed by the new approach, including unsupervised machine learning routines and artificial intelligence methods. Two crucial moments were used to analyse the occurring tactical patterns for the specific game: T0 (time instant of the shot) and T1 (time instant of the last action before the shot). For both moments typical tactical patterns, considering the spatio-temporal context of the situations were found. By combining the patterns of both moments, hidden tactical strategies were identified and their efficiencies calculated. The most significant result is that offence attempts have the highest scoring probability when played via the wings near to the goal line and the final pass, before the shot, is given against the playing direction into the centre of the field.