Résumé (EN)

In an upstream supply chain dedicated to the mass production of customized products, decentralized management can be an efficient and effective method in a steady state in which stochastic characteristics of customers’ demands remain stable. However, this is possible only if all echelons that precede the final assembly line use periodic replenishment policies that restrain the stockout risk to a low predetermined probability. The safety stocks’ levels are more difficult to define for alternative or optional parts, as well as the components they use, whose demands are weighted sums of random variables, affected by several random factors and organizational constraints. The factors and constraints to consider are not the same for supplied and produced components. The random demand of a component depends on the demand of alternative or optional parts mounted in the final product, through a double transformation involving the bill of materials explosion, which is at the origin of the weighted sum of random variables, and time lags. In the steady state, the knowledge of the probability distribution of that random variable allows for the determination of safety stocks that decouple the management of upstream supply chains. Progressive changes in the steady state require periodic and progressive adaptations of the safety stocks that do not directly depend on the final demand knowledge.