There is not much research on team collaboration in digital entertainment games, nor is there much evidence for the efficacy of game-based team training or the validity of game-based team assessment. This is a shortcoming because of an increasing pervasiveness of serious games in organizational life, e.g. for operational training, management and leadership. Is it possible to establish marked relationships between psychometric constructs that measure ‘team composition and performance’ and ‘analytics’ that unobtrusively measure gameplay performance? If so, what are the implications for game-based team research and assessment? The authors conducted explorative, quasi-experimental (field) experiments with the multiplayer serious game TeamUp. One field experiment was conducted with 150 police officers as part of task-specific twoday team training. Research data were gathered through pre-game and postgame questionnaires on team constructs such as ‘psychological safety’ and ‘team cohesion’. A large quantity of in-game data was logged to construct indicators like ‘time needed to complete the task’, ‘speak time’ and ‘avoidable mistakes’ to measure team performance. The conclusion of the analysis is that ‘team cohesion’ and ‘psychological safety’ correlate moderately and significantly with in-game performance indicators. Teams with an unequal individual game performance speak the most, while teams with an equally low or equally high individual performance spend significantly less time speaking. The indicative findings support the need to further develop validated analytics and gamebased environments for team research and assessment.