Economic theory predicts that organisations achieve higher levels of productivity when tasks are divided among different subsets of workers. This prediction is based upon the expectation that individuals should perform best when they specialise upon a few tasks. However, in colonies of social insects evidence for a causal link between division of labour and performance is equivocal. To address this issue, we performed a targeted worker removal experiment to disrupt the normal allocation of workers to a cooperative team task - tandem running. During a tandem run a knowledgeable leader communicates the location of a new nest to a follower by physically guiding her there. The targeted removal of prominent leaders significantly reduced tandem performance, whereas removal of prominent followers had no effect. Furthermore, analyses of the experience of both participants in each tandem run revealed that tandem performance was influenced primarily by how consistently the leader acted as a leader when the need arose, but not by the consistency of the follower. Our study shows that performance in ant teams depends largely on whether or not a key role is filled by an experienced individual, and suggests that in animal teams, not all roles are equally important.