Budgeting Discretion: Theory and Evidence on Street-Level Decision-Making

A new AI research paper models how human decision-makers like caseworkers should ration their discretion to override rigid policies. The study formalizes discretion as a dynamic allocation problem with a limited 'override budget' over time. It reveals an optimal 'dynamic threshold rule' and a key behavioral invariance: the rate of using discretion depends only on the statistical shape of potential gains, not their scale. The theory was validated using real data from a homelessness services system, showing overrides track operational constraints like workweek timing and housing capacity.