[4N1-GS-10c] AI application

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[4N1-GS-10c-05] Development of a Conversational Agent for Investment Advisory with a Focus on Natural Dialogue

〇Mizuho Toyama¹, Kaito Toda¹, Yuji Akatsuka¹, Yasuhiro Sekine¹, Temma Saito², Satoshi Matsuse², Hane Kanazawa², Tomohito Yamada², Yousuke Motohashi² (1. Daiwa Securities Group Inc., 2. NEC Corporation)

Keywords:

Agent,Dialogue System,Industrial Applications,Finance

The purpose of this study is to develop a conversational agent that achieves both strict regulatory compliance and natural dialogue in investment advisory services. We propose a multi-agent architecture composed of conversation analysis agents, a dynamic routing agent, specialized task agents, and an output adjustment agent, which enforces strict flow control based on predefined rules and procedures while simultaneously enabling flexible and empathetic responses that adapt to user reactions and inquiries. We evaluated the prototype with users including securities-firm employees against two single-prompt baselines: a “flexible” LLM and a “strict” LLM. The agent improved psychological safety, empathy, question naturalness, and situational adaptability relative to both baselines. It strengthened flow control versus the flexible baseline but remained slightly below the strict baseline. The flexible baseline achieved the highest conversational naturalness and reader adaptation, indicating a trade-off between expressive naturalness and tighter control. Additional challenges include long-context understanding and latency.