顧客基礎分析:以一非契約型之線上零售商的顧客購買資訊為例

Abstract

顧客關係管理與一對一行銷不僅是近年來行銷研究領域的熱門議題之一；實務上，許多企業也紛紛致力於深度經營顧客關係，以減少顧客流失率。本研究基於顧客關係管理中的顧客基礎分析(Customer Base Analysis)，利用已知的顧客購買行為(如一定期間內，顧客的購買次數、購買金額、最近一次購買日)等資訊，透過兩個主要的機率模型的結合：BG/NBD模型(Fader et al., 2005 a)與Extended SMC模型(Schmittlein et al., 1994)，加上基於BG/NBD模型的假設，我們推導出原模型並未導出的公式---任一隨機抽取顧客的期望存活率，試圖去預測未來某段期間內，個別與整體顧客的購買次數、平均單次購買金額及顧客存活率。同時，我們將使用一家線上日本動畫及影音VCD零售商店的資料庫，去從事實證分析；並透過單因子多變量分析的驗證，我們發現本模型的三種主要的預測結果，也可以成為良好的區別龐大顧客群的變數，顯示本研究除了可以成為未來顧客一生價值分析(Customer Lifetime Value)的基礎模型外，也同時是企業在落實一對一行銷的良好工具之一。最後，我們將BG/NBD模型原先提供的工作試算表格式，轉換成為一個更有效率、更方便使用的試算表格式，透過新的試算表，在輸入已知的顧客過去購買行為後，我們便可在同一時間、一次得到所有顧客在未來某期間內的預期購買模式。除了可方便模型的使用外，也期待此舉能更模型廣泛被使用並提高模型的價值。

In this research, we combine the BG/NBD (Fader et al., 2005) and the Extended SMC model (Schmittlein et al., 1994) to simultaneously and completely incorporate the past purchase behavior of customers to do some effective forecasts based on customer base analysis. Differed from the “entire” extended SMC model (based on Pareto/NBD), this research preserves and advocates the easy implementing of the BG/NBD and consider the past dollar volume spent by customers simultaneously by adding the Extended SMC model. Hence, our model is more suitable to be a basis for doing further CLV research than the “pure” BG/NBD, which doesn’t consider any “monetary” information of customers. Furthermore, based on the BG/NBD, we derive the equation of expected active probability for a random chosen customer. It could help us to understand the individual active probability and the true customer base of a firm after summing the active probabilities of all customers. We also empirically validate our model by using a database from an online VCD retailer and try to anticipate the possible purchase patterns of customers in the future both individually and collectively. And we also validate our model results through 1-way MANOVA to test and then we have statistical evidence to approve the differentiation capabilities of the key expected values. Finally, we transform the worksheet of the BG/NBD to a more user-friendly form. With this new worksheet, we only should put the basic purchase history of all customers into and then we could get the expected values of interests at one time. It could save a lot of time to implement this model especially when the base of customers is huge. And the other purpose is that we wish it could improve the utility rate of our model.