A meta-analysis of the effects of AI agent implementation on customer-level outcomes
该元分析整合95篇文献、468个效应量，发现客户对AI代理的平均反应不如对人类，这种厌恶源于偏见，但效果受设计、信息、情境和国家因素调节。
Firms increasingly use AI agents, such as recommendation algorithms and chatbots, to enhance customer value, yet research documents mixed effects on customer outcomes. To address and clarify these heterogeneous findings, we conduct a meta -analysis of 468 effect sizes reported in 95 articles with 82,751 participants examining AI agent implementation across both substitution contexts (AI replacing humans) and adoption contexts (AI introduced into previously non-AI processes). Results indicate that customers, on average, respond less favorably to AI agent implementation. Drawing on implicit prejudice theory, we attribute this aversion to a prejudicial bias that arises when AI systems intrude into domains traditionally managed by humans or manual systems, evoking a perception of threat and mistrust. However, the direction and magnitude of this effect depend on AI design, information, contextual, and cross-country factors. Together, these findings advance understanding of when and why customers resist or embrace AI agents, offering actionable guidance for managers.