The purpose of police work and prosecution is crime control.
But what police and prosecutors do from day to day is make arrests and secure verdicts (or guilty pleas) and thus sentences. It seems natural to count those activities and use the counts as performance measures. That, however, turns out to be a mistake. Actual arrests and prosecutions are mostly costs rather than benefits.
[Putting someone away is a benefit when an especially active bad guy gets locked up, preferably for a long time, thus reducing criminal victimization through incapacitation, but the median person who goes to prison isn't actually worth locking up, balancing the costs -- financial and nno-financial -- of keeping him behind bars against the benefits of the crimes he doesn't commit while incarcerated.]
Actual arrests and prosecutions are resource hogs. Threatened arrests and prosecutions are cheap. The key to effective crime control is combining maximally convincing threats with a minimum of actual delivery. That’s not as hard as it sounds: a really convincing threat will virtually eliminate the activity it’s directed at (or crime by the person it’s directed to), thus eliminating the need to deliver on the threat.
This thought is a very difficult one for most police and prosecutors to wrap their heads around, especially since they live in organizational environments that reward activity rather than crime reduction. (The military understands this better: it’s axiomatic that the best war is the one you don’t have to fight because your obviously overwhelming force induces foreign players to do what you want without having to shoot them.) But if we learn to do it right, we can have both less crime and less punishment.
Update: Glen Whitman has related thoughts.
Crime Control
This strikes me as simple and right.