In multiple-task agency setups it is commonly accepted that wage incentives
must be weaker when the agent's performance on some of the activities is d
ifficult to measure. This article shows that stronger incentives can be res
tored through a scheme of selective audits in which the appraisal of less t
angible activities is contingent on observing high performance levels in th
e more visible tasks. This scheme would make the efforts expended on the va
rious tasks complementary rather than substitutes in the agent's utility fu
nction. It is optimal under plausible assumptions concerning the monitoring
technology (separability of the multivariate likelihood function) and the
agent's risk behavior (absolute prudence larger than three times absolute r
isk aversion).