以資料採礦之羅吉斯回歸方法建置信用評分模型之研究 -以某電信公司為例

Abstract

自民國86年開放電信民營化以來，競爭態勢日趨嚴峻不言可喻，電信經營受到前所未有的極大挑戰。不論是從市場競爭面來看，有不斷加入的網路電信營運商、低價行銷策略，如網內免費吃到飽服務、以及智慧型手機所帶動的新創OTT(Over the Top)服務，如Line & Face Book等；抑或從法規面來看，影響巨大深遠者，如96年配合電信法2波六年連續調降網外行動資費，降幅達26.9% (14.0% + 12.9%)，光是99年至101年，營運商已流失202億元營收、另一波調降中間接續費管制也從102年起，持續調降中，這都將導致電信營運商營收及獲利受到巨額侵蝕。要如何維持電信應運商經營績效，實為一值得研究課題。 本研究係以一主要行動通信網路營運商為案例，研究該營運商如何藉由資料採礦 (Data Mining) 技術的協助，結合自建資料倉儲系統(Data Warehouse)資料，篩選出六大類共16個有效變數，以羅吉斯回歸 (LR: Logistic Regression) 方法建置預測用戶欠費違約之模型與信用評分，並注入ECE (Excel Customer Experience)精神，設計出「矩陣式差異化」之催收策略管理，強化風險管理能力，以提升經營績效外，亦能滿足該營運商專注創造用戶美好經驗與服務品質企業文化，達到用戶、員工及公司三贏局面。 經實證分析結果發現，以該16變數所建置Logistic Regression預測模型&信用評分有效性甚高，有著極佳之預測力，在配合公司策略及可容忍風險能耐下，運用「矩陣式差異化」所設計出之催收策略管理的確大幅度提升經營績效，增加該營運商競爭力。擷取具體直接可量化數據，第一年業已提升經營績效約$74.1M，包含因應收帳款累計回收率上升，可減少呆帳損失超過約$40M、因好壞用戶區辨能力強，適切的催收策略管理，提升委任年催收效益至少$7.4M，及每年降低客戶服務成本及客戶取得成本約$26.7M。

It is going without saying that the competition situation is getting severer and severer, and telecom operator faces unprecedented challenge when Taiwan government deregulated Taiwan telecom market in 1997. From market aspect, there is always new entrants, and low-price market strategy, such as all-you-can-eat, and boomed over-the-top service produced by smart devices, such as Line & Face Book… etc., is whipping out the market in recent years; or from regulation aspect, based on Taiwan Telecommunication Act, the price cut for user’s rate plan mandated by NCC has made telecom operators lose at least of NT$20.2 billion for the second-three-year ( 2010~2012), and has reached total 26.9% price-cut percentage for the past six-year, also another price cut for middle price lasting 4 years is on from 2013 with total cut percentage of 46%. The impact from above will eventually result in eroding telecom operator profitability heavily, thus, how the telecom operator makes strive is a worthiness class to study. This study is taking a mobile telecom operator as an example, study how the telecom operator to filter out the 16 distinguished variables with data mining technique after taking advantage from own data warehouse, and how to predict the possibility of default and build the credit score model by logistic regression algorithm . With the high model validity, how the new differential and strategic dunning management designed by matrix of PD & ARPB creates three-way-win for subscribers satisfied by touching experience, for employees satisfied by good effectiveness in work and for company satisfied by good operational outcome and company image while ECE (Excel Customer Experience) spirit implanted. As a result of empirical analysis, the credit score model has very good validity with reliable predicting performance. To apply the new differential and strategic dunning management, abiding by company strategy and acceptable risk tolerance, has indeed generated prosperous operational outcome and has raised company competitiveness dramatically. Taking some concrete figures, in terms of operational outcome for the first year, total operational benefit is around NT$71.4M, including from less bad debts loss of NT$40M owing to better recovery rate of accounts receivable, from less commission expenditure of NT$7.4M owing to better collection performance, and from acquisition cost and customer support expense savings of NT$26.7M owing to lower churn rate and customers complaints.