The Chilean State delivers essential meal services at schools for low-income students. Junta Nacional de Auxilio Escolar y Becas, the institution in charge of covering 1,300,000 children, leases the meal service to private enterprises. We developed an integer linear programming model to assign the meal contracts, in a process known as combinatorial auctions. The resulting model, which is NP-hard, led to significant improvements in efficiency and also contributed to making the process more transparent. The results are apparent in substantial improvements in quality and coverage of the service, and important savings to the country, which are equivalent to feeding 300,000 children in addition. We developed techniques to solve the combinatorial models and also to analyze and compare multiple scenarios to find robust solutions. For the objective function of this problem, we analyzed several options to consider different kinds of social benefits. In this paper, we describe the problem, the methodology and the results. We also present empirical results based on 6 years of experience. Finally, we discuss the relevance and impact of using operations research in these central issues in developing countries.