Consider a business model based on per day an individual did not recidivate, pro-rated by a prediction model for initial likelihood of recidivism, and severity of crime.

This would perhaps ascribe a number for level of difficulty, and an initial estimate for how hard the case is.

Keep in mind that a convicted offender on the streets may suffer from a potential for revenge and retaliation, and thus is possibly safer in jail, however horrible jail may be. This would augment a recidivism predictor further, if you cannot isolate them from their victims, should victims be bitter about the perception of your service lacking punitive action.