Organizations are represented as conversational networks of agents, which interactions are modeled by a two-player game of deterrence, i.e. a qualitative game based on the concept of threshold. A congestion control algorithm is then derived from the game Boolean solution set. Last, analysis is extended to the case where the solution set is fuzzyfied. On the whole, the approach enables a mix-management: decentralized when agents can manage their relations successfully by themselves; centralized when the risk of congestion arises.