Churn prediction plays a critical role in the success of cloud service providers, enabling them to identify potential churners in advance and prevent customer attrition. In cloud computing scenarios, enterprise customers constitute a significant portion of the cloud services market. They contribute substantially more revenue and exhibit much more stable subscription patterns compared to individual consumers. Therefore, it is essential and feasible to perform churn prediction and user retention, focusing on enterprise customers. This paper is grounded in actual enterprise customer churn prediction scenario in Huawei Cloud Computing Technologies Co., one of the largest cloud service providers in China. The data collected for churning behavior analysis consists of three parts: order data, webpage browsing data, and monitoring metric data. Building upon the data, we propose an effective churn prediction system that provides a comprehensive solution for enterprise customer churn prediction in cloud services. The system is based on a data-driven churn predictor, CloudChurn, which utilizes specially designed structures for handling distinct feature subset. It integrates these sub-models in a boosting-like manner, effectively addressing the data imbalance problem and culminating in enhanced predictive performance. The effectiveness of CloudChurn is demonstrated through both offline and online evaluation. Since its deployment on the ModelArts platform in June 2023, it has consistently achieved a 78.3% accuracy in recognizing potential churners, contributing to revenue retention at the million-level. This highlights CloudChurn’s potential as a valuable tool for cloud service providers to address customer churn and optimize their service offerings. Codes are publicly available at https://github.com/YHYHYHYHYHY/CloudChurn/tree/main .