Customer – Software Agent Negotiations Using Large Language Model: An Experimental Study

Abstract

Large Language Models (LLMs) present remarkable opportunities for researchers and professionals to improve the effectiveness of software agents acting on companies' behalf. A particularly promising application is using LLMs to negotiate deals with potential customers. This paper proposes integrating negotiation models with LLM capabilities to generate textual offers in machine-human negotiations. It builds on an assumption that LLMs demonstrate emotional intelligence in their interactions with humans, positively influencing negotiation outcomes and human perceptions. The work introduces a prototype agent application based on a phone plan sales scenario. An experiment with human participants tested the performance of the LLM-powered negotiation agent against a version without LLM. The results indicate that the LLM-enhanced software agent reached agreements with better prices.