[2L4-GS-5c] Agents

1:45 PM - 2:00 PM JST(4:45 AM - 5:00 AM UTC)

[2L4-GS-5c-02] Concept and Technical Validation of a Sales Support System Based on Generative AI Multi-Agent Systems

〇Yoji Kawano¹, Satoshi Suga² (1. MACNICA, Inc., 2. Kansai University)

Keywords:

LLM Multi-Agent Systems,Sales Negotiation Simulation,Behavior Change

With rapid advances in large language models (LLMs), AI agents are increasingly expected to support business operations. We present a technical validation of an LLM-based multi-agent system that emulates customer–salesperson negotiation to support enterprise sales. The system parameterizes customer organizational constraints and psychological states and simulates iterative consensus-building dialogues with a sales agent. Using real product information, we evaluate whether rubric-based scoring of utterances enables interpretable Bayesian updates of a latent customer conviction modeled as a Beta distribution, and we visualize conviction trajectories under an ROI-focused sales strategy. Across trials, rubric-scored utterances yielded consistent, interpretable posterior updates, often showing a monotonically increasing posterior mean. We discuss challenges and future directions for internal deployment, including sales-strategy hypothesis testing and early warning of lost-deal risks.