Prescribing Decisions Using AI-Based Prescriptive Analytics: Procedure, Challenge, and Validation Pathways

Prescriptive analytics uses predictions to recommend decisions, but its superiority over the current approach is commonly evaluated on historical data in which only the cost of the implemented decision is observed. When a prescribed decision differs from the historical decision, its actual cost is a counterfactual outcome and is unavailable, so the conventional comparison between predicted and actual costs does not establish superiority. This study formalizes the conditions under which this fundamental challenge arises and develops a four-case classification according to the dimensions of the feature and decision spaces, the existence of an intermediate random variable vector enabling closed-form cost evaluation, and the observability of its components. Each case is assigned a validation pathway: Cases 1 and 2 permit direct prescription or evaluation from sufficient historical observations or fully observable intermediate variables; Case 3 constructs test sets with fully observable intermediate variables through restriction or sampling; Case 4 uses domain-specific explicit rules to approximate the unobservable costs of alternative decisions. The conditions under which a lower evaluated cost supports a conclusion about the actual cost ordering are established. Three freight transport examples illustrate the classification and the validation pathways.