There is increasing evidence that graded return-to-work is an effective tool for the rehabilitation of sick-listed workers. Still, little is known on the optimal timing and level of grading in return-to-work trajectories, as well as the allocation of trajectories across worker types. To fill this gap, we analyze whether the effectiveness of graded return-to-work depends on the starting moment of the trajectory and the initial level of graded work resumption. We use administrative data from a Dutch private workplace reintegration provider. In order to correct for the selection bias inherent to the evaluation of activation strategies, we exploit the discretionary room of the case managers in setting up treatment plans. In correspondence with previous literature we find that graded return-to-work reduces sick spells with eighteen weeks within the first two years after reporting sick. However, the probability of work resumption after two years remains unchanged. Work resumption can be achieved faster when graded return-to-work is started earlier or at a higher rate of initial work resumption. These findings how- ever do not hold for individuals who have problems related to mental health.