Analysis of Customer Purchase Frequency and Profitability in Retail Pharmacy Stores

本研究主要探討藥妝零售產業提升預測顧客行為的模型與方法，並以RFM模型為基礎進行延伸。RFM模型在行銷領域中是廣泛被使用的模型，具有良好預測和分群顧客的能力，本研究在此模型中加入了兩項新指標：集中度 (C) 和 廣度 (B)，並針對顧客的「交易頻率」和「交易利潤」進行分析，藉此找出優於RFM的指標組合。首先將RFM、C、B共五項指標進行排列組合，並以迴歸分析驗證新增的兩項指標能顯著提升模型解釋能力，接著將RFM指標組合及RFMCB指標組合分別作為機器學習方法的解釋變數以預測顧客行為。對顧客交易頻率而言，C和B兩項指標的加入能顯著提升其預測能力，對顧客交易利潤而言，新指標的加入，平均而言對於預測精準度有所提升，但在部分資料中會使誤差值增加以致整體誤差的最大值有所提升。This research proposes modeling techniques to better predict customer behaviors in the retail industry. Extending the widely-adopted RFM model in marketing, we introduce two new metrics – clumpiness (C) and breadth (B). Using more than two million transaction records from over 100 retail pharmacy stores in Taiwan, we fit a set of regression models, in which we assess the explanatory power of different combinations of RFMCB for customer purchase frequency and profitability. Our analysis shows that the RFM model is significantly inferior to models with C and/or B, suggesting that C and B are indeed promising metrics. In the next stage, we will apply machine learning methods to incorporate C and B into predictive models and assess their out-of-sample prediction performance. On Average, RFMCB outperforms RFM in predicting Frequency & Profit. However, there are some cases where RFMCB leads to larger prediction error