Moral hazard is the workhorse model of financial incentive provision in economics, with applications ranging from pay-for-performance in employer-employee relationships to the design of insurance premiums. While an extensive theoretical literature has delivered important qualitative insights about the properties that real-world incentive schemes ought to have, the practical implementability of these prescriptions is limited by the model's demanding assumption that the employer knows both the employee's performance capabilities and also how strongly he responds to financial incentives. Our paper “Data-Driven Contract Design'' develops a novel methodology for using real-world performance data to design incentive schemes that work well even in situations in which the employer has no prior knowledge of the employee's ability beyond what's reflected in the data. In our main result, we show that robustly optimal compensation plans augment the employer's existing compensation structure with equity bonus payments for good performance.