在研究顧客關係管理中，如何分辨及維繫舊有的高價值顧客是很重要的課題，對 於企業而言，該課题則是生存的關鍵。高價值顧客已被視為是企業的珍貴資產，因此 如何針對高價值顧客對於企業未來的營收貢獻進行預測，則相當於預測企業的資產消 長。本研究以RFM模型作為基礎，分別結合貝氏隨機模式及ARIMA進行顧客行為 預測，再比較兩種模式的預測效度。模型驗證的資料為某化工貿易公司的的企業對企 業交易資料，包含157個企業顧客共1，928筆交易資料。在預測效度比較方面，分別 比較預測正確率及偏誤程度。研究結果顯示，以RFM模型結合ARIMA模式的預測 效度較RFM模型結合貝氏隨機模式佳，研究並發現區隔狀態的定義會影響預測的效 度。分析結果也顯示RFM模型結合ARIMA模式較傳統貝氏隨機模式具有更高的準 確度。經過K-mean演算後，結果發現高價值顧客數目僅佔顧客總數的18.59%，但卻 佔交易總額的76.68%及交易頻率的66.55%，非常符合20-80法則所描述的「20%的顧 客帶來80%的獲利」。此外我們也發現顧客狀態空間定義上的盲點。

In studying customer relationship management, how to find the high- value customers and retain them is a very important topic; this issue is also the key for a company for survive. High-value customers have been considered to be a company’s precious property，Therefore, forecasting the high-value customers1 revenue contribution will help forecasting the company's property change. In this study, we combine the RPM model with Bayesian stochastic model and ARIMA, respectively, to construct two models for forecasting customer state. The forecasting accuracy of these two models is tested by the 1,928 transactions records from the 157 business customers of a chemicals trading company. The results indicate that the RFM model with ARIMA performs better than the RFM model with Bayesian stochastic model. The present study also find the deficiency of traditional customer state definition.