We develop a double-sided moral hazard model of social entrepreneurship and derive the optimal state subsidy. Then, we analyze the data of an EU-funded training and mentoring program aiming at preparing social entrepreneurs for private financing. Using content analysis techniques, we investigate the 203 applications for the program, the reviewers’ evaluation, and the selection decision. Social enterprises produce private and public benefits, use market and non-market resources, and involve a wide range of stakeholders with different incentives. We examine why different projects can get active financing (financing plus advisory), or only passive financing (financing without advisory), or no financing at all. We identify five relevant selection criteria such as entrepreneurial net present value, entrepreneurial agency cost, advisory net present value, advisory agency cost, and the external effects of the project. Empirical findings are consistent with the theoretical model. Applicants with higher scores in business plan, social impact, and geographical scope were significantly more likely to be selected, especially if their activities required no domain-specific knowledge from the advisors. However, higher agency costs, reflected in too many business lines and early-stage operations, seem to reduce the chances significantly. We formulate a moral hazard model for social entrepreneurship with four simultaneously optimizing players: an entrepreneur, an investor, an advisor, and the state. With the help of our unique database, we get valuable insights into the financing decisions of a profit-seeking investor. Our findings can contribute to the improvement of the design of state-subsidized social entrepreneurship programs.