主题: Early Evidence on the Adoption and Effects of AI Assistants on E-Commerce Platforms
中文题目：从采纳到影响：电商平台AI助手应用的早期证据
时间: 2026年5月18日上午10.00-11.30
地点:管理科研楼801会议室
主讲人: Prof. Zemin (Zachary) Zhong, Associate Professor, Rotman School of Management, University of Toronto
主持人: 彭雪丰
Zemin (Zachary) Zhong is a tenured Associate Professor of Marketing at the Rotman School of Management, University of Toronto. He earned his Ph.D. from UC Berkeley and B.S. from Peking University. His research uses both analytical modeling and econometric models to explore how online platforms and information design influence consumer search, pricing, and market outcomes, with work published in MKSC, MS, and JEBO. He is now serving as Associate Editor of Marketing Science. Zachary’s contributions have been recognized by the ISMS Doctoral Dissertation Award and multiple Service Awards from both MS and MKSC.
照片：
Abstract:
This talk presents two complementary studies examining the integration of Large Language Models (LLMs) into digital platforms. First, I present large-scale descriptive evidence on the adoption and usage patterns of AI assistants within e-commerce ecosystems. Using granular data from Ctrip, China’s largest online travel agency, we analyze "Wendao,” a DeepSeek-based AI assistant integrated into the platform’s shopping interface. Based on a sample of 31 million users observed between April and August 2025, we establish a set of stylized facts regarding who adopts these tools, when they are deployed in the purchase journey, and the specific consumer intents they serve.
Building on these descriptive insights, the second part of the talk focuses on the causal impact of reasoning capabilities in this context. In a large-scale field experiment, we randomly assigned over 500,000 users to an AI assistant powered by either a state-of-the-art reasoning model (DeepSeek-R1) or a comparable non-reasoning model (DeepSeek-V3). Counterintuitively, access to the reasoning assistant reduced hotel bookings by 2.5%. We show that while the reasoning AI provided more suggestions, it discouraged users from browsing outside the chat interface—a "search-narrowing" mechanism associated with longer, more detailed responses. Conversely, while immediate sales declined, the reasoning assistant significantly increased 30-day user retention. These findings highlight a critical trade-off: more advanced AI does not necessarily improve immediate commercial outcomes, and model design choices can have complex, unintended consequences for consumer search behavior.