The analysis of targeting of cash benefits is typically silent on whether any success is due to encouraging claims from the poor or to the decisions of administrators on the claims they receive. By contrast, the paper models the probabilities of households’ knowledge of a new social assistance scheme, of a claim conditional on knowledge, and of an award conditional on knowledge and claim. It uses household survey data from Uzbekistan where a new social assistance benefit is administered by community organisations. The paper therefore also illustrates problems of design of decentralised social assistance schemes in developing countries.