管理工程学报
管理工程學報
관리공정학보
Journal of Industrial Engineering and Engineering Management
2015年
2期
149~159
,共null页
齐佳音 马君 肖丽妍 钟永光
齊佳音 馬君 肖麗妍 鐘永光
제가음 마군 초려연 종영광
客户风险 客户终生价值 贝叶斯网络 Beta风险
客戶風險 客戶終生價值 貝葉斯網絡 Beta風險
객호풍험 객호종생개치 패협사망락 Beta풍험
customer risk; customer lifetime value; Bayesian Network; beta risk
客户终生价值(Customer Lifetime Value,CLV)建模中如何考虑客户风险是目前客户关系管理领域所关注而且有待解决的难题。本文旨在探讨如何通过客户风险修正CLV模型。首先将客户风险划分为包括波动风险、衰退风险、流失风险与信用风险,并提出前三者的度量方法;随后采用改良的贝叶斯网络方法,从条件概率的角度描述了风险之间的关系,同时计算客户风险值;最后,借鉴金融领域对资产进行风险修正的方式,通过贝叶斯网络输出的风险得分,计算客户的Beta风险,对传统的CLV模型中的折现率进行风险修正,得到RCLV模型。本文利用河北省某电信公司的中小企业客户的固定电话数据进行模型仿真和验证,说明所建立模型的合理性和应用价值。
客戶終生價值(Customer Lifetime Value,CLV)建模中如何攷慮客戶風險是目前客戶關繫管理領域所關註而且有待解決的難題。本文旨在探討如何通過客戶風險脩正CLV模型。首先將客戶風險劃分為包括波動風險、衰退風險、流失風險與信用風險,併提齣前三者的度量方法;隨後採用改良的貝葉斯網絡方法,從條件概率的角度描述瞭風險之間的關繫,同時計算客戶風險值;最後,藉鑒金融領域對資產進行風險脩正的方式,通過貝葉斯網絡輸齣的風險得分,計算客戶的Beta風險,對傳統的CLV模型中的摺現率進行風險脩正,得到RCLV模型。本文利用河北省某電信公司的中小企業客戶的固定電話數據進行模型倣真和驗證,說明所建立模型的閤理性和應用價值。
객호종생개치(Customer Lifetime Value,CLV)건모중여하고필객호풍험시목전객호관계관리영역소관주이차유대해결적난제。본문지재탐토여하통과객호풍험수정CLV모형。수선장객호풍험화분위포괄파동풍험、쇠퇴풍험、류실풍험여신용풍험,병제출전삼자적도량방법;수후채용개량적패협사망락방법,종조건개솔적각도묘술료풍험지간적관계,동시계산객호풍험치;최후,차감금융영역대자산진행풍험수정적방식,통과패협사망락수출적풍험득분,계산객호적Beta풍험,대전통적CLV모형중적절현솔진행풍험수정,득도RCLV모형。본문이용하북성모전신공사적중소기업객호적고정전화수거진행모형방진화험증,설명소건립모형적합이성화응용개치。
Ever since the mid-late 20th century, in the field of customer relationship management, customer lifetime value (CLV) has been focused and received much attention from scholars and the enterprise managers. Although the idea that customer risk should be considered into the CLV modeling is widely accepted, it is still a question to be further studied that how to reasonably add the customer risk into CLV modeling. In this paper, we aim to explore the method of using customer risk to adjust CLV modeling. Our contributions are as follows: Firstly, we provide several kinds of customer risks that affect CLV and formulate metrics. We develop two selection rules for risks which are the risk is able to actually influence the cash flow and be measured by the data of enterprises. According to the rules, four types of risks are picked up, including the wave risk, the fall risk, the chum risk and the credit risk. The computational methods are provided for first three risks.Today, evaluation index system of customer risk is wildly used and its subjectivitymakes the evaluation results difficult to convince people, so we develop three mathematical models to measure risks objectively. Particularly, we adopt the proportional hazard function which is combined with the objective probability model and the factors of reality for calculating individual probability of chum. Secondly, we research the relationships between four types of customer risks and merge them into one single risk score with an improved Bayes Network. Nowadays, the traditional Bayes Network method uses in customer relationship research used basic datadirectly, so it's incapable of showing the mechanism of data and dealing with time series data.We abandon the traditional Data Mining method and improve the existing Bayes Network method in order to overcome the two weaknesses above. The specific method is calculating the risk through the formulate metrics in part one, making it to be the nodes of Bayes Network and then building the Bayes network of the various types of risk through the training data. Thirdly, we adjust the CLV model with customer risk by learning the idea from Beta risk based on the results of the first two parts. There arc a small number of studies on how to adjust the CLV with customer risk, and the greatest contribution of our paper is that a complete modeling process is provided. We try to adjust CLV with risk in a financial way. We transform the risk score which is the output of Bayes Network into Beta risk score, and finally the risk-adjust customer lifetime value (RCLV) model is set upand adjusts the discount rate in the basic CLV model with the Beta risk. An empirical study is carried out with the operation data from the small and medium-sized telecom company in Hebei province, China. We find that among the CLV top 20 customers, the RCLV of the customer whose CLV ranked 10 ranked 52.It proves that it's of difficultyto distinguishthe risk-high customer through the existing CLVevaluating methods and the RCLV model makes up this.Ultimately, we address that the RCLV model can improve the precise capacity of the enterprise customer management.