We introduce learning by doing in a dynamic contest. Contestants compete in an early round and can use the experience gained to reduce effort cost in a subsequent contest. A contest designer can decide how much of the prize mass to distribute in the early contest and how much to leave for the later one in order to maximize total efforts. We show how this division affects effort at each stage, and present conditions that characterize the optimal split. The results are indicative of the fact that the designer weakly prefers to leave most of the prize mass for the second contest to reap the gains from the learning by doing effect.