This paper examines the role of random audits in tax enforcement as an information acquisition tool. Random audits provide an unbiased view of the taxpayer population but are less cost-effective than risk-based audits. We develop a dynamic model in which the Tax Agency learns about the income-generating process from past audits and uses this information to improve future audit targeting. We show that random audits create an informational trade-off: they improve estimation precision by increasing sample representativeness, but reduce the informativeness of risk-based audits due to budget constraints. We characterize conditions under which a share of random audits increase total information and derive a necessary and sufficient condition for when they are revenue-maximizing. Our results provide a theoretical foundation for incorporating random audits into optimal audit design and suggest that their informational value may justify their use as part of an effective enforcement strategy.

When resources are limited, information about the environment becomes critically important. This paper introduces a stylized dynamic model to illustrate how the challenge of audit agents—who may engage in activities such as tax evasion, crime, or other undesirable behaviors—is shaped by the need to gather information for future use. Our analysis reveals that monitors might choose to forgo immediate preventive actions in favor of collecting valuable information. This strategic decision allows them to better identify and target high-risk agents in the future. The extent to which monitors prioritize learning over deterrence depends on how valuable is learning. We discuss two key factors that affect the value of acquiring information: heterogeneity in agents’ characteristics and level of penalties. We also show that an information acquisition rationale arises even in a static model with delegated enforcement.