Previous research shows that Benefit-Cost Analysis (BCA) is seldom done in Sweden, and that the results e.g., in Norway and the Netherlands do not influence the ultimate policy choice. We explain why bureaucrats may choose (not) to do a BCA with cognitive- and search costs coupled with career concerns. Given the initial policy chosen by an agenda setter, bureaucrats who stay working at an agency have policy preferences close to the initial policy; those with reservation wages above a threshold quit and therefore do not influence policy. The bureaucrats’ preferences converge to the initial policy level over time. A BCA reveals the inefficiency of the initial policy and the bureaucrats consequently have no incentive to do one, except when the policy is restricted by a binding governmental budget constraint.