We study a model where an organization dynamically learns about the effect of managerial behavior on employee well-being. A manager takes an action of unknown harm, which an employee can report for investigation and possible punishment. These reports generate information for both managers and the organization. We show that the welfare effects of changing punishment, the efficacy of investigation technology, the ease of reporting, or the value of a manager-employee match depend on the severity of harm. When the harm is low, increasing these parameters may actually decrease employee welfare, either by motivating more managers to commit harmful actions that employees do not want to report, or inducing managers to opt out of matching with an employee altogether. The optimal punishment for managers can be expressed as a function of the harm, and may be interior. Employee welfare may decrease upon convergence to the steady state equilibrium, and in steady state, harmful actions are seldom punished and no organizational learning occurs.