Data-Driven Demand Forecast for Online-to-Offline Operations : An Adaptive Hierarchical Incremental Approach

Online-to-offline (O2O) refers to a new type of e-commerce that combines online order acquisition and offline on-demand order fulfillment service. The daily demand for O2O store is affected by both online and offline factors. Given the highly dynamic online operation and offline environment, the effect of latent factors may change over time. Therefore, forecast at the aggregation level may be subject to real-time information loss and structural changes, and generate less accurate forecasts. In this study, we propose an adaptive hierarchical incremental forecasting (AHIF) approach to forecast daily O2O store demand, which integrates incremental method to handle structural change and hierarchical process to catch the valuable real-time information. A Data-driven O2O Demand Forecast Application based on the AHIF approach is implemented in one of the largest O2O platforms in China in an actual business environment. The proposed AHIF approach can significantly improve forecasting accuracy compared with conventional forecasting methods, as shown from the result of the case study. By tracing and quantifying the contribution of performance improvement by different characteristics of the proposed algorithm, this study also provides valuable insights for future algorithm development and operational improvement concerning O2O operations management