Decentralization for multidivision enterprises

The paper deals with the organization of decision making for multidivision enterprises. If decisions can be represented by linear programming models with division sharing resources, an organization is proposed with one division or a combination of these setting resource prices, while the remaining ones determine quantities. The role of each division is determined by the numerical coefficient values as well as the models’ structure. This approach is related to, but quite different from, the Dantzig-Wolfe decomposition principle. Interactions between divisions are like ordinary commercial transactions with one party setting the price and the other the quantity traded. The usefulness of this organizational structure depends on its stability for expected variations of model data, which can be determined by parametric variations and simulations. These concepts are applied to a model by R.M. Burton and B. Obel to decide whether the M-form or the U-form organization is preferable. For central values of the model data the M-form turns out to be appropriate, and remains valid for large variations of the coefficients. A comparison with Dantzig-Wolfe decomposition indicates that the proposed approach has significant advantages.