Organizational change management (OCM) has become unprecedented important for companies to remain their competitive edge in the fast-moving business environment. However, despite the the help of well-designed management strategies and tools, it still lacks quantitative ways to forecast the potential resistance and future consequence that may arise from organizational change initiatives. To this end, in this paper we propose a data-driven approach to predict the impact of organizational change by exploring the large-scale communication data of employees, with the focus on turnover, a critical issue in talent management. Specifically, we first creatively embed the communication network of employees, along with the structure of organizations, into a graph, named Connection Net. Then, we leverage the spectral graph theory to build a time-aware spectral neutral network for predicting the employee turnover following organizational changes. Finally, we evaluate our approach with several classic baseline methods on real-world company data. Experimental results clearly validate the effectiveness of our approach for predicting both exact number of turnovers and turnover rate after organizational changes.