This paper analyses principal-agent contracts when the risk-averse agent's action generates information that is not directly verifiable but is used to make a risky decision in a formulation more general than previously studied. It focuses on the impact on the decision made and the contract used, establishing a necessary and sufficient condition for the principal to gain by distorting decisions away from what is efficient and conditions under which there is no conflict between incentives to make decisions and to take action. Applications to investing in a risky project and bidding to supply a good or service illustrate those results.