From Risk Scores to Decision Value: An Evidence-Governed Business Analytics Framework Under Capacity Constraints

Organizations increasingly use machine learning to identify customers, transactions, suppliers, and projects that require attention. Yet a high predicted risk does not establish that an intervention will help, that its benefit exceeds its cost, or that the underlying evidence permits action. This paper develops an analytical framework that connects prediction, intervention value, evidence verification, and limited managerial capacity. The framework combines temporally governed data, calibrated risk estimates, bounded intervention effects, constrained optimization, and selective human review. Its central object is a decision record linking each proposed action to the information available at the decision time, the assumptions behind its expected value, and the authority required for execution. Analytical propositions establish when risk ranking agrees with value ranking, how uncertainty changes the intervention threshold, and when additional review has positive expected value. A fully specified six-case worked example demonstrates the distinction. With capacity for two interventions, risk ranking produces an assumed expected net value of −1 unit, whereas value-based selection produces 54 units. A common uncertainty allowance of 0.10 reduces the optimized lower-bound value to 34 units. These quantities are arithmetic consequences of stated assumptions, not estimates from an organizational experiment. The contribution is an auditable decision formulation and a prospective evaluation protocol that separates predictive performance from intervention effectiveness. The framework offers testable design principles for business analytics while identifying the causal, temporal, and organizational evidence required before deployment.