Automated Mediator for Human Negotiation: Pre-Mediation via a Structured LLM Pipeline

The article presents a structured pipeline of LLM modules designed for pre-mediation in human negotiation, which includes specialized components for dialogue, preference prediction, response-level critique, and structured summarization. This approach addresses limitations of traditional monolithic models and demonstrates comparable preparation outcomes to human mediators, achieving a 36% lower RMSE on preference inference tasks and a reduction in excessive affirmation patterns. The findings highlight the potential for scalable, low-effort pre-mediation support using structured LLM pipelines, making it relevant for practitioners looking to enhance negotiation processes with AI.