Selling mode choice and AI streamer adoption in a live-streaming supply chain with consumer returns

Artificial intelligence (AI)-enabled live-streaming is rapidly transforming online retailing, while also introducing significant operational challenges related to post-purchase dissatisfaction and consumer distrust toward AI streamers. We develop a Stackelberg game to analyze a dual-channel supply chain in which a manufacturer sells through an offline channel and a platform-based live-streaming channel operated under either an agency mode or a reseller mode. The model formalises the key operational trade-offs in streamer selection: the human streamer typically provides stronger selling influence but may induce impulsive purchasing and higher product returns, while the AI streamer mitigates return risk but suffers a trust deficit that can be alleviated by investing in intelligence levels. Our analysis reveals that the optimal streamer choice is contingent upon the product’s post-purchase expectation mismatch probability and the human streamer’s unit sales cost. Specifically, the human streamer is preferred for low-return categories to maximise conversion, while the AI streamer is optimal for high-return products to curb impulsive buying. The reseller mode emerges as a Pareto-dominant strategy for managing products characterised by high return risks. Two extensions, which consider a positive post-purchase expectation mismatch probability and channel advantage heterogeneity, confirm the robustness of our main insights.