We study the optimal incentives provision in non-stationary, long-term projects. A project’s progress is modeled by a Brownian motion that must hit a target level for completion. An agent has CARA preferences, can privately save and borrow, and exerts hidden effort to increase the process’s drift. Thanks to the absence of wealth effects, the usual two-dimensional state space (progress and continuation utility) collapses to a single state variable—project progress—thereby allowing a direct comparison with the first best.

We show that moral hazard depresses effort below the planner’s level, raises the threshold at which the principal cancels the project, and, if the principal endogenizes the project’s scope, leads to a less ambitious project target than under full information. A key feature of the contract is the possibility of “resets”: the principal can abandon the current approach at some low progress level and restart the project from a fresh approach. We show that relative to the first best, the principal resets too early when the reset point lies close to completion, and too late when it is far from completion. 

In an extension, we introduce unobservable risk-taking, which boosts the drift but raises the probability of a disaster that terminates the project. The optimal contract may tolerate risk-taking at early stages—where the agent is shielded from severe penalties upon disaster—but discourages risk-taking when progress is near completion. By balancing effort incentives against moral hazard in risk choice, the resulting effort profile exhibits a discontinuity at the boundary where risk-taking is no longer permitted.