整合顧客回應與退貨行為之直銷決策支援模型

Abstract

近年來，隨著電子商務技術之進步，零售業也隨之興起透過虛擬通路銷售產品，在這種趨勢之下也帶動了消費者購物習慣的改變，使得網路購物市場規模相當可觀。基於網路交易的特性，這種新興的商業模式除了快速吸引大量的消費者促使企業成長外，也為業者帶來了許多不確定性和不同的支出成本，因此，本研究將分別針對實行直效行銷時顧客名單的篩選以及購物後顧客之退貨情形進行探討。 本研究藉由資料探勘技術進而了解回應行銷活動與具退貨行為之顧客特質，蒐集的變數除一般顧客基本輪廓，另包含於直效行銷領域中廣泛使用的RFM(最近一次消費、消費頻率、消費金額)，以及影響退貨決策的產品種類與付款方式等變數。首先利用羅吉斯迴歸分析篩選出具影響力之相關變數，接著使用類神經網路進行預測模型建構，再以增益圖評估模型表現，並透過決策樹輔助規則呈現。 研究結果發現，顧客是否使用過廠商提供的折價優惠於回應及退貨兩模型中皆具相當影響力，可見與顧客互動的重要，此外，在規則呈現中，顧客之居住地、退貨率、產品偏好與付款方式等也具有相當重要性。且兩模型預測出之顧客名單有所重複，因此，在篩選寄送型錄之顧客名單時，若將退貨情形列入考慮，便可降低對高風險顧客之行銷活動，節省額外作業成本。In recent years, with the advancement of e-commerce technology, retailers have begun selling products through virtual channels. Consumers have also changed their shopping habits under the trend, which increases the online shopping market scale. Based on characteristics of online transactions, the new business model not only fast obtain a large number of consumers to increase business growth, but also brought a lot of uncertainty and various expenditure for the online retailers. This study will probe into customer response and return behavior in direct marketing by data mining techniques. The explanatory variables include general customer profile, RFM (recency, frequency, and monetary value), method of payment and product types, etc. Logistic regression analysis was be used to filter relevant variables, then both two prediction model were constructed by back propagation neural networks. Finally, decision tree algorithm was applied to reveal rules. The result shows that whether customers using discount is a significant factor in both models. Moreover, attributes such as city, return ratio, preferences of payment and product also show importance when applying decision tree algorithms to explain the prediction models. Considering return behavior could avoid deal with high risk customer and decrease extra cost when firms make marketing decisions.