Abstract
Large Language Models (LLMs) hold great promise for automating telemarketing, yet deploying them in real, revenue-critical conversations remains challenging due to strict reliability requirements. We present AI-Salesman, a reinforcement learning driven dialogue optimization system that builds the full pipeline of training, inference, and evaluation for large-model telemarketing. By optimizing sales scripts through reinforcement learning and grounding the dialogue policy in business objectives, AI-Salesman produces reliable, goal-directed conversations. Deployed in a live business environment, the system increased the core business conversion rate by 10%~20%, demonstrating that carefully aligned LLMs can deliver dependable performance in high-stakes commercial telemarketing.
Qingyu Zhang
Master’s Student in Computer Science
I build LLM agents that hold reliable multi-turn conversations in real-world settings, and I work on making large models more efficient.