We extend the standard evaluation framework to allow for interactions between individuals within segmented markets. An individual's outcome depends not only on the assigned treatment status but also on (features of) the distribution of the assigned treatments in his market. To evaluate how the distribution of treatments within a market causally affects the average effect within the market, averaged over the full population, we develop an identification and estimation method in two steps. The first one focuses on the distribution of the treatment within markets and between individuals and the second step addresses the distribution of the treatment between markets. We apply our method to data on training programs for unemployed workers in France. We use a rich administrative register of unemployment and training spells as well as the information on local labor demand that is used by unemployment agencies to allocate training programs. The results show that the average treatment effect on the employment rate causally decreases with respect to the proportion treated in the market. Our analysis accounts for unobserved heterogeneity between markets (using the longitudinal dimension of the data) and, in a robustness check, between individuals.