資料庫行銷之顧客價值遷徙路徑分析：貝氏多項式模型之運用

Abstract

自電腦資料庫管理系統（DBMS）的技術發展成熟後，企業可輕鬆地紀錄顧客交易資料，但龐大的顧客交易紀錄需要有理論為依據的方法來加以分析研究，才能夠從中獲得有利於企業的決策資訊。資料庫行銷即是透過統計方法，從顧客資料庫之中萃取出個別顧客的攸關資訊，可幫助企業分析顧客生命週期價值、價值遷徙路徑、品牌選擇等，也可以預測具有相同的顧客其購買行為的可能性。

本研究是運用馬可夫鏈理論，以顧客跨期狀態建構馬可夫鏈移轉機率，並透過貝氏多項式（Bayesian Multinomial；BMN）模型，推估出顧客在各種狀態間機率的貝氏統計值，並以對照組樣本進行模型預測力之檢驗。並依據個別顧客的馬可夫鏈移轉機率矩陣推算顧客未來的價值遷徙路徑，並透過逐步回歸法來分析價值遷徙路徑與顧客特質之間是否存在關聯性。本研究所得之結果，可以幫助企業在了解顧客未來的價值遷徙路徑之後，進行後續的行銷策略擬定，進而達到資料庫行銷之目的。

Since the technologies of Database Management System （DBMS） have been well-developed, enterprises could retain and process enormous transaction data easily than before. But when enterprises want to analyze these transaction data, they should study it based on rigorous methodology, therefore they could gain value-added infor-mation from their customer transaction data in order to develop customer-related strate-gies which could make enterprises communicate to their customers directly and fulfill their needs according to heterogeneity. The enterprises can predict customer lifetime value、customer value migration path、brand choice and also probability of different purchase state in the future. The main purpose of this article is to predict customer purchase state given on his/her prior state. Markov Chain Theory helps us figure out the conditional probabilities of each state therefore to build individual customer’s Transition Probability Matrix of Markov Chain and analyze it by using Bayesian Multinomial Model. After that we verify the predictive power of models by using out-sample and try to find out if there is any correlations within the transition paths and demographic variables. The contribution of this study could help enterprises identify customers’ transition paths and develop fol-low-up marketing strategies in order to achieve database marketing.