This paper examines the performance trajectories of Counter Strike: Global Offensive teams after a change in team composition. A large data set from a competitive team-based online game is analyzed using semi-parametric generalized additive mixed-effects models. The data show that teams (five members) over-perform directly after a change, after which performance dips and does not recover. The effect is independent of the size of the change, i.e., of how many members are already known to each other from the previous stage. Moreover, over the average lifespan, team performance declines by about 2.4 percent of the interquartile range of overall team performance. From an applied perspective, the data supports the view that small teams targeting tasks for which performance is easy to evaluate are best newly put together on a case-by-case basis.