Empirical validation of an activity-based optimization system

We present an optimization model for steel manufacturing, whose objective function depends on activity-based costing (ABC). The constraints describe the capacity and the material balance, the product-dependent yield and the electrical balance, for all the activities of the plant. For the model to be reliable, one should validate that it accurately expresses the objective function of the user, and that it defines with precision the constraints that determine the feasible solutions. We validate the cost estimation by empirically testing how accurately it can predict future expenditures. We validate the constraints by comparing the polyhedron they generate with the production possibility set obtained with data envelopment analysis (DEA). We show the model implemented and validated in a Chilean steel manufacturer. The tests prove the model reliable enough for defining the product mix, and for evaluating investments and sourcing alternatives.