企業為了提高營收獲利,須經由顧客資料庫分析與預測顧客交易行為,以及判斷忠實顧客與流失顧客,使未來的行銷策略與活動經費能達到維持顧客與開發顧客的投資效益,並進而使企業的獲利增加。但是目前企業尚未從學術界的研究中,獲得實用性的顧客交易行為預測方法,其主要原因之一在於目前學術界所發展的預測模型都太複雜不易使用,且其預測效果並未明顯的比簡單RFM分析及迴歸預測結果要好。本研究以某量販店的會員交易資料為例,使用Fader et al.於2005年所發表宣稱預測準確度佳且較容易使用的BG/NBD預測模型,藉由預測模型建立過程並進行顧客交易行為預測分析,探討預測模型對企業管理者的實用性。實證結果顯示,BG/NBD預測模型建立僅需微軟Excel套裝軟體,且樣本資料的處理彙整也相當容易,經由模型建立過程,對顧客交易行為的解釋能力也大為增加。因此,本研究驗證BG/NBD預測模型對企業管理者確實具有實用性。在某些情況下顧客交易資料可能為有限的樣本,例如企業規模較小資料獲得成本太高、天災造成資料庫損毀等等,本研究探討在這種有限的樣本情況下,同樣以微軟Excel套裝軟體建立“finite mixture model”預測模型的預測可行性。同時以企業管理者比較偏好的目標顧客管理,進行目標顧客的分類顧客交易行為預測與銷售預測,探討預測模型對目標顧客的預測能力。並以迴歸預測分析比較BG/NBD預測模型與“finite mixture model”預測模型的預測能力。實證結果顯示,與企業常用的迴歸預測方法比較,“finite mixture model”預測模型可以提供全體顧客樣本較準確的交易行為預測及90%以上的分組目標顧客樣本準確的交易行為預測。 In the current market environment, managers need to predict the purchase behaviour of their customers. However, there is a frustration among academics insofar as the stochastic customer base analysis models have not found their way into wide managerial application. It mainly dut to the time and money costs associated with implementing complex stochastic models in managerial practice. The family of NBD models and their extensions (Pareto/NBD and beta-geometric /NBD) represent a promising approach. In this paper, we use hypermarket transaction data to build up beta-geometric /NBD model and to predict the purchase behaviour of customers. We also compare the predictive performance of BG/NBD model and regression models in terms of repeat purchase levels and active status. In the condition of limit transaction information is avalible, we present that the “finite mixture model” can also has good predictive performance.This paper concludes that the family of NBD models is worth to use by managers in terms of it ease to implementing and better explanation capability of customers’ purchase behaviour, compare to regression models.