Development of Automated Negotiation Models for Suppliers Using Reinforcement Learning

During a negotiating process in the supply chain, the negotiating parties try to maximize their interests while presenting a reasonable proposal that the counterpart may accept. Human negotiators, however, can have difficulties in making the optimal negotiation proposal or decision due to various limitations related to time, cost, and information processing capabilities. To deal with these issues, this study proposes artificial intelligence-based models to automate the negotiation process for a supplier in the supply chain. Specifically, we focus on the sell-side perspective as this area has not been studied comprehensively. To deal with the insufficient amount of data, we generate data on quantity (Q), price (P), and delivery lead time (D) by analyzing the correlation among these three variables from the available real transaction data. Then, more than 23,000 negotiation episodes between a buyer and a seller are simulated with correlated Q, P, and D. For learning the optimal negotiating strategy through simulated episodes, we apply the reinforcement learning models based on the Q-learning algorithm. We present two negotiation models: (i) a P negotiation model dependent on Q and (ii) a P negotiation model dependent on D. Our results show that the proposed structure of the automated negotiation model with correlated Q, P, and D can be applied to various negotiation environments by helping sell-side agents.