Centralization vs. Decentralization in a Multi-Unit Organization: A Computational Model of a Retail Chain as a Multi-Agent Adaptive System

This paper explores the effect of organizational structure - in terms of the allocation of authority - on the rate of innovation in multi-unit organizations such as retail chains and multi-plant manufacturers. A computational model is developed in which store managers continually search for better practices. In a decentralized organization, a store manager adopts a new practice if it raises her store's profit. Headquarters (HQ) is assumed to observe the new practice and then decides whether to disseminate it to other stores. In a centralized organization, a store manager who generates an idea that would raise her store's profit passes the idea up to HQ for approval. Due to lack of detailed information about stores' markets, HQ decides whether or not to mandate it across the chain on the basis of chain profit. Given that stores are assumed to have heterogenous markets, the obvious virtue to decentralization is that it gives authority to those who have the best information and this allows practices to be tailored to each market. What our analysis reveals, however, is the presence of an implicit cost to decentralization. Allowing stores the freedom to develop very different practices is shown to reduce the amount of inter-store learning; that is, the frequency with which one store's idea is of value to another store. By keeping stores near each other in store practice space, centralization enhances learning spillovers and, in some cases, this results in higher chain profit than is achieved under decentralization. We find that centralization outperforms when stores' markets are not too different, consumer demand is sufficiently sensitive to a store's practices, and markets are changing sufficiently rapidly over time.