报告题目：ML On Time? The Value of Machine Learning Tools on Airlines’Operational Performance

时间：2026年7月1日15：00-17：00

地点：主楼317

报告人：万翔

报告人国籍：美国

报告人政治面貌：群众

报告人职称：教授

报告人工作单位：美国俄亥俄州立大学

报告人简介：

万翔，男，博士，现任美国俄亥俄州立大学费雪商学院 教授，终身教职， 博导， FCOB 杰出教授，商学院博士生项目负责人，俄亥俄州立大学参议员；主要研究方向：供应链管理， 人工智能和机器学习。Decision Sciences Journal 副主编， Journal of Business Logistics 资深主编，Production and Operations Management 编委, Journal of Operations Management 编委，Strategic Management Journal， Manufacturing and Service Operations Management,等十余个国际顶级学术期刊审稿人。在SCI学术期刊发表十多篇学术论文。万教授的科研成果发表在Strategic Management Journal, Manufacturing and Service Operations Management, Production and Operations Management, Journal of Operations Management, Decision Sciences, and Journal of Business Logistics,等国际知名学术期刊。万教授被华尔街日报（Wall Street Journal）报道为12个供应链大数据应用的专家之一 ，其科研论文“Too Much of a Good Thing? The Impact of Product Variety on Operations and Sales Performance”被世界四大会计师事务所之一德勤（Deloitte）引用并讨论。

报告内容简介：

Machine learning (ML) tools have been rapidly applied to many industries in recent years. Their operational impacts, however, are mixed in the literature. While prior research shows that ML techniques can improve predictive accuracy, the improvement in operational performance is not automatic. Still, it depends on how effectively the operations group utilizes these recommendations by ML. Inappropriate use can even degrade performance. In this paper, we investigate the impact of AI-driven machine learning decision-making support tools on airlines' operational performance. This study quantifies the influence of such ML tools and unravels the mechanisms driving performance changes by disentangling the information improvement effect and the misapplication effects associated with ML. Our research focuses on the first AI-driven ML platform in the airline industry. This platform generates predictive signals and operational recommendations using ML models trained on historical data to support operational decision-making. Using the Difference-in-Differences (DID) framework, we find that implementing this ML technology significantly reduced departure delays. To explore the mechanism, we test whether the information-improvement effect is constrained by a misapplication effect associated with ML. The treatment effects of ML tools are investigated across operational conditions (extreme weather and non-hub-to-non-hub flights) and over time to assess tradeoffs between the information-improvement effect and the misapplication effect. Our empirical analyses of ML applications in the airline industry contribute to the literature on human-AI interactions. Our findings offer managerial insights for adopting and refining ML-enabled decision support by highlighting how misapplication can attenuate benefits and why ML implementation should be adapted to different operational conditions.

（承办：管理科学与物流系、科研与学术交流中心）