Artificial Intelligence, Emotional Labor, and Service Operations
通过随机田野实验发现，AI服务机器人比人类更准确地表达情绪，但只有情绪与任务情境匹配时（如对轻度逾期客户表达积极情绪、对反复逾期客户表达消极情绪），AI才能比人类多收回49%–94%的债务；否则AI表现更差。
Problem definition: Emotional labor is increasingly demanded in service operations, placing tremendous psychological strain on human employees and posing challenges to scalability and sustainability. Our study scrutinizes whether artificial intelligence (AI) service bots may tackle these challenges by examining how and when AI’s engagement in emotional labor enhances economic performance in service operations. Methodology/results: We provide causal evidence from a pair of randomized field experiments conducted in partnership with a firm for loan collection service. Results suggest that, compared with human employees, undisclosed AI service agents display the required emotions (both positive and negative) more accurately. However, AI’s advantage of higher emotion display accuracy does not always guarantee better economic performance. For AI to collect more payments from borrowers than human workers, the displayed emotion must be contextually appropriate. Specifically, AI substantially outperforms human workers in debt collection by 49%–94% when the emotion display instructions are suitable for the collection task (i.e., displaying positive emotions to borrowers with minor delinquency but negative emotions to borrowers with repeated delays). However, when the displayed emotions are unsuitable, AI backfires and performs worse than human employees because of its unwavering adherence to inappropriate emotional display instructions. Further, AI’s performance advantages over human agents are amplified when the suitable emotions involve negative (versus positive) valence. We also leverage the machine learning method causal forest to explore heterogeneous treatment effects across customer segments. Managerial implications: Our research suggests that operations managers should deploy AI to reduce frontline employee emotional burnout, develop explicit emotional labor guidelines for AI, and use negative-emotion AI strategically to boost compliance and efficiency. It is also important to identify emotion-intense operational tasks, target AI when it has clear advantages, and set up continuous monitoring and quality control for AI emotional performance. Funding: Z. Fang acknowledges support received from the National Natural Science Foundation of China [Grant 71925003] and the Double First-Class Initiative of Sichuan University. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2024.1459 .