We review the principal–agent multi‐tasking literature and discuss the relevance of this literature to the implementation of performance measurement in public organizations. Arguably, the most important lesson from the literature is that performance measurement may elicit dysfunctional and unintended responses, also known as gaming responses. We propose an evolutionary model of how organizations manage performance measures when gaming is revealed over time. The model stresses the dynamic nature of the performance‐measure selection process—a feature that has been overlooked in the literature. We present evidence from a job training programme that is consistent with the model and we discuss implications for the selection of performance measures.