Computational physics

A researcher has an observable Y of interest and a correlated quantity Z whose expected value is known exactly. How can Z be used as a control variate?