Decentralized Learning in a Retail Network

About this event

A prominent explanation for why organizations delegate decision making authority is the informational advantages of a decentralized decision structure. By delegating authority, the principal can exploit agent's private information that is hard to communicate. In addition, even if all the information can be communicated, the principal may still want to decentralize decisions in order to reduce the cost of processing information by distributing the task across multiple agents. The organizational economics literature stresses the role of uncommunicatable private information, but this paper empirically demonstrates that the second factor could be more important by using a proprietary dataset from a retail network selling beverages through vending machines in Japan. This network introduced a point-of-sales system but still delegates part of product assortment decisions to agents hired by third-party operating companies. By comparing the expected revenues under the actual policy and a hypothetical centralized assortment policy that does not use private information but processes the history of sales without cost, I show that the delegation of product assortment decisions in the network cannot be rationalized solely by the existence of private information. It is necessary that the principal expects a substantial cost of processing information when decisions are centralized. It is also shown that the cost of processing information can be the main reason for the decentralized decision making. This result and empirical framework help us to better predict the effects of information-technological innovation on organizational structures.