We study an agency model in which the principal has outcome data under different incentive schemes and aims to design an optimal contract under minimal assumptions about the way the agent responds to incentives. In particular, the principal knows the agent-optimal actions, which are distributions over outcomes, in response to $K$ ``known” contracts but is unaware of other actions available, and importantly, of their costs. The principal seeks a contract that maximizes worst-case profits. The optimal contract is a mixture of the known contracts and the (linear) one that makes the agent residual claimant. Moreover, when $K=1$, the single known contract maximizes the principal’s profit guarantee, whereas with two known contracts, the optimal mixture puts strictly positive weight on one of the known contracts. Our methodology is straightforward to implement, a point that we demonstrate using data from DellaVigna and Pope’s (2018) experimental study of different incentive schemes.