This paper studies the design of monitoring/audit policies in dynamic settings. Firm quality

is private information and evolves stochastically via a Markov process with transitions depending

on the rm's unobservable eort. The rm benets from having a reputation for quality. A

principal designs a monitoring policy that allows him to learn the rm's quality by conducting

costly reviews. Monitoring plays two roles. First, it plays an incentive role, because when the

principal discloses the outcome of inspections to the public, it aects the rm's reputation.

Second, information is directly valuable when the principal's payo is convex in reputation, for

example because it allows consumers to make better choices. Our main result is a characterization

of the optimal monitoring policy that induces full eort. The policy is surprisingly simple.

It is either deterministic, whereby the date of the next inspection is pre-announced, or random,

with a constant hazard rate of next inspection. We discuss how the type of optimal monitoring

depends on history, the cost of monitoring, the persistence of quality, and the cost of eort. We

also consider how the optimal monitoring policy is aected by the presence of exogenous news,

showing how the evolution of the hazard of random monitoring depends on whether the absence

of news is perceived to be good or bad news about quality.