The objective of this memoir is to evaluate the effectiveness of employment assistance programs for recipients with a severely limited capacity for employment under the Social Assistance Program. To our knowledge, there is no causal analysis of the programs’ effect on the number and length of time recipients did not receive social assistance. Our analysis is based on the study of two dependent variables: the cumulative period spent off Social Assistance Program after first time use, and how many times the recipient has exited Social Assistance Program. We use a state-of-the-art method based on automatic learning (a subfield of the artificial intelligence) to evaluate the effects of participation. In the context of the employment integration of people with disabilities, this method has never been used. Our findings show that participation (in employment assistance programs) prolongs time spent out of social assistance and those effects are relatively heterogeneous throughout the population.