奧客客群之偵測與探索：以某台灣染整業為例

Detecting and exploring the cluster of difficult customers: a case of dyeing industry in Taiwan

以RFM變數進行顧客區隔（Customer Segmentation）是一種最廣為人知的手法，其基本流程是先各別對單變數切等分、找出各變數之群落邊界（Cluster Boundaries）、再建立資料立方體（Data Cube）。然而，過去典型的方法皆是主觀地決定各變數該切幾等分，並未提及其緣由，亦無法因應領域的不同而做適當的調整，故激起本研究提出一套適應性的程序。本研究採用k-means演算法搭配間隙統計量（Gap Statistic）來達成自動決定各變數分群數與群落邊界之願景，希冀得到更客觀的顧客區隔結果，該方法已成功地被運用在台灣某一染整業中，其能偵測出客訴行為較活耀的客群（稱奧客群），並釐清客訴成員之客訴原因以及布匹幅寬、品牌與顏色之前三大客訴類型。

Using RFM variables to segment customers is one of the most well-known methods. The basic procedure is to cut the data into bins for each variable, to find the cluster boundaries of each variable and then to establish the data cube. Conventional methods, however, select the fixed number of bins subjectively without mentioning about the reasons which encourages us to propose an adaptive procedure. This study associates k-means and gap statistic to achieve the goal of auto-selecting number of bins and cluster boundaries. This method is successfully applied to a dyeing industry in Taiwan in which the cluster of difficult customers is finally detected. Those difficult customers have complained over the defective fabric actively. Based on their complaint history, we try to clarify the sources of customer complaints and to find out the top three complained width, brand and color.