臺鐵會員制再深化──利用資料探勘技術訂定忠誠計畫規則

MINING TRA’S TRANSACTION DATA FOR LOYALTY PROGRAM RULES

積點制的會員回饋計畫，或所謂的忠誠計畫，已廣泛地被許多行業用來維繫與顧客間的關係，甚至作為刺激顧客消費與公司獲利的工具。由於會員系統需要資源的持續投入，因此有必要針對會員的加入進行篩選，甚至分級提供不同程度的服務，以激勵及鎖定會員的消費。本研究利用臺鐵局電話及網路訂票的巨量資料庫，以RFM及延伸變數進行集群分析與建立決策樹模型，並輔以忠誠度矩陣加以評估，結果發現利用三個月訂票資料所建置的顧客價值分類準則即可達到94%的預測準確率，同時本研究並運用前述準則篩選出高價值旅客及具潛力的旅客作為臺鐵局邀請加入會員系統的對象，也建議藉由降低積點兌換門檻、提供多樣化的獎品以及採取層級式的會員架構等措施來強化會員積點制度的功效。

Point-based customer loyalty program has been extensively adopted in many industries to maintain customer relationships, even to stimulate repeat purchases from customers and to obtain more profits for companies. Due to the need to continuously invest resources in loyalty programs, companies should only allow profitable customers to join the programs. This study evaluates the ticket reservation data of Taiwan Railways Administration with RFM and extended variables using clustering and decision tree techniques and loyalty matrix concepts to identify customer values. Through this research, we are able to provide 94% classification accuracy on our decision tree model employing three-month ticket reservation data. Also, high-value and potential high-value customers are identified via the classification rules for member-recruiting. In the end, lowering the thresholds of redeeming points, offering diversified rewards and using tier membership structure are suggested to enhance the functions of the loyalty program.