We provide a framework to implement, evaluate and interpret machine learning (ML) methods and apply this framework to identify nonlinear and interactive patterns in data. Compared to traditional causal inference methods, ML algorithms make fewer a priori assumptions about the functional form of the underlying model that best fits the data—thus researchers can use ML methods to inductively explore novel and robust patterns in data. Once identified, these patterns can be tested using traditional econometric methods. Using employee turnover data, we demonstrate how to implement three specific ML algorithms (decision trees, random forests, and neural networks) to uncover meaningful patterns not captured by our baseline econometric model. We demonstrate visual tools to evaluate machine learning model performance and interpret patterns identified in the predictive model. We also summarize human decisions and common pitfalls in implementing machine learning methods. Importantly, ML analysis should be interpreted as exploratory and inductive and not causal. An online appendix provides code to employ the ML algorithms described in the paper.

Alibaba's Rural Taobao initiative had been launched in 2014 as a public service initiative to increase e-commerce adoption in China’s remote rural areas. In the first two iterations of the initiative, dubbed “1.0” and “2.0,” Alibaba had partnered with local businesses and young returnees to operate village stations. Village stations served as logistical and educational hubs to help villagers make online purchases. Despite significant growth in sales through village stations, the business model was operating at a loss, and competitors JD and Pinduoduo were growing quickly, threatening to challenge Alibaba by claiming the rural e-commerce markets. Bill Wang, the head of the initiative, was tasked with designing and implementing the next model for Alibaba’s future expansion into rural areas. The decision centers around whether Alibaba should copy the successes of competitors JD and Pinduoduo, escalate commitment to its original value proposition (hands-on service to villagers), or embrace an entirely new model. The case highlights the competing pressures of financial performance, CSR and corporate mission, government relations, and long-term competitive positioning that are common to major strategic decisions.