資料探勘方法應用於目標市場分析之研究──可停電力潛在目標用戶探勘分析實例應用

Abstract

一直以來，企業及組織皆積極努力地尋找可有效擴大市場占有率並增加利潤之可行方法與方向，然侷於有限之資源與資金，改善既有行銷策略並輔以施行具鎖定及開發價值目標顧客群之效的目標行銷乃為各領域產業及研究人員頗重視之探討議題與要點，特別在顧客消費習性快速變遷之現代，更加突顯目標行銷之有效行使與規劃於各類相關決策施行效益與成果之影響力。 雖目前資料探勘已被廣泛地運用於各領域產業之相關研究，且其於目標行銷上之應用績效亦廣受好評，但此類研究少有針對行銷措施已然影響顧客消費行為、造成既有客戶與未消費客戶間缺乏相同消費行為資訊基礎之情境進行研究，故其藉由分析顧客消費習性建立之分析架構所探勘萃取出之消費行為特徵較不盡客觀，有無法確切掌握具開發價值與潛能之目標顧客群的疑慮。鑑此，本研究旨在利用可行之資料探勘技術結合相關目標行銷理論之期許，構建一可基於已受行銷措施影響而改變既有客戶消費行為資訊，關聯歸結出其他較不受影響(靜態)之基本變項，且具產生合理潛在目標顧客比對規則之效的「缺乏相同客戶消費行為資訊基礎之目標行銷分析模式」，以為一可行之目標行銷實行辦法。 考量電力用戶之用電消費行為易受電力公司所推行之各類需求面管理措施影響而改變，造成既有客戶與未消費客戶間缺乏相同消費行為資訊基礎之問題，以及目前尚無運用資料探勘技於輔助電業推行目標行銷之例，本研究乃選定電力產業為實例應用研究之標的，嘗試為個案公司建立起一可行之可停電力措施潛在目標用戶探勘分析流程，期能協助順利改善現階段可停電力參與用戶成長率下降及尖峰降載量無明顯增加等問題。 經實際探勘分析結果發現，本研究所提之潛在目標用戶探勘分析架構確具有效縮小個案公司之目標行銷範圍之效(不計入既有用戶僅有1、2戶之集群，目標用戶群僅占整體探勘母體之21%，特別是各類「最具潛力」目標用戶群之占比更僅為整體探勘母體之0.76%)，且部分提列予產業專家評判之目標顧客群名單的可用性達75%(=90/120)，可證實本研究所建立之潛在目標探戡分析模式極具應用參考價值，且此類「缺乏相同客戶消費資訊基礎之目標行銷分析模式」頗具可行性與未來發展潛力。

With the widespread application and development of information technology, target marketing has become a major trend in marketing. With the assistance of Data Mining techniques, target marketing has been widely practiced in a number of industries. Though Data Mining has been frequently applied in target marketing research, few studies have considered the situation where customers have already been influenced by marketing measures. Hence, there is a lack of information about the consumption behavior of current customers and potential customers. The Electricity Corporation faces this problem quite often, we have chosen the electricity industry as an example for analysis. The target customer analysis framework of the Interruptible Load measure introduced in this research is mainly based on current customer’s consumption records. The goal is to identify some basic variables, which are more static, for generating identification rules and targeting customers, could also be seen as a compensation for general target marketing model. From the analysis result, we find out that the target marketing scope, which excludes the mining results of clusters with only 1 or 2 participants, has reduced the original target population to 21%. Meanwhile, the number of high-potential target users of each Interruptible Load measure has been narrowed even further to only 0.76% of the original target population. Furthermore, our correlation analysis can help find the basic variables that are often ignored by ANOVA. However, because the basic variable related to segmentation variables, it would be better to include this basic variable into the description variables. Thus we can prevent the problem of information dimension shortage from happening. Therefore, the referential value and applicability of this target analysis framework constructed by this research has been proven.