Artificial Intelligence (AI) Assistant in Online Shopping: A Randomized Field Experiment on a Livestream Selling Platform

Livestream technology enriches consumers’ online shopping experience, enabling streamers to demonstrate products in real time while interacting with a large number of consumers for product sales. However, tension arises between streamers’ constrained service capacity and consumers’ individual service demands on livestream selling platforms. Streamers can only handle a finite number of interactions and inquiries because of time and capacity constraints, whereas consumers expect immediate, tailored responses. In this work, we examine whether and how an artificial intelligence-powered streaming assistant (termed “AI streaming assistant”), which helps consumers with interactive chat-based support for information acquisition and processing, can mitigate this tension in livestream selling. We report a randomized field experiment on a leading livestream selling platform, where the consumers in the treatment group had access to an AI streaming assistant during livestream sessions and the control group did not. Our results reveal that implementing an AI streaming assistant increases sales by 3.00% and reduces the product return rates by 12.55%. Our exploration of plausible mechanisms suggests that access to an AI streaming assistant increases consumers’ perception of intelligent information provision (and, in parallel, interruption), which in turn reduces (and increases) uncertainty in decision making. Overall, the benefits of the AI streaming assistant’s intelligent information provision outweigh its interruptions, subsequently increasing consumers’ purchase intention and decision-making confidence. We also differentiate and explore two distinct modes of human-AI interaction, AI’s proactive and reactive interactions, and our correlational results show that these interaction modes reinforce each other in increasing purchases and reducing product return rates. This study contributes to the literature on human-AI interactions, livestream selling, and product returns in online commerce. Our findings also provide actionable implications for online commerce platforms in designing and implementing AI artifacts.

History: Hsing Kenneth Cheng, Senior Editor; Jingjing Zhang, Associate Editor.

Funding: The work was supported by the National Natural Science Foundation of China [Grants 72201038, 72421001, and 72293561].

Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.0103.