All-or-nothing payments

We develop a general principal–agent framework in which to study optimal incentive schemes where agents are hired to work on multiple heterogeneous and interdependent projects. The incentive schemes can be based on output measures, interpreted as the principal’s payoffs, as well as input measures, regarded as observation of some of the agents’ efforts. We identify that a unifying feature of the optimal incentive schemes, called all-or-nothing payments, arises in three natural scenarios of the general framework: unobservable inputs, verifiable inputs, and observable but unverifiable inputs. Our framework and results embed and generalize several previous studies on multitask principal–agent problems with a limited liability constraint.