Using Negotiation and Large Language Models in Human – To Software Agent Negotiations

Large Language Models (LLMs) offer unprecedented opportunities for researchers and practitioners to enhance the performance of business representatives, including software agents acting on behalf of businesses. One promising application is employing LLMs to negotiate deals with potential customers. In this article, an approach combining negotiation models at the back end with the capabilities of LLMs to generate textual offers at the front end in machine-human nego-tiations is proposed. A prototype agent application based on a phone plan sales case is presented. An experiment involving human subjects tested the performance of the LLM-powered negotiation agents against a version without LLM. The results show that two versions of LLM-enhanced soft-ware agents achieved more beneficial agreements for the agents (the sellers) as compared to agents without LLM. Furthermore, this gain in objective outcomes by agents did not result in the worsening of the subjective assessments of negotiations.