【作者】 郭磊；

【导师】 刘杰；

【作者基本信息】 复旦大学 ， 管理科学与工程， 2013， 博士

【摘要】 在“以客户为中心”经营理念的时代背景下,客户价值计算作为差异化服务的理论基础,受到了业界和学术界的广泛关注。本文以数据挖掘技术、Bayes统计、RFM和Pareto/NBD等理论和模型为基础,从客户价值构成要素、客户行为特征、客户行为预测等几个研究视角切入,使用文献综述、调查研究、数据挖掘建模等规范的研究方法,对客户价值的动态衡量问题进行了深入研究,并给出了企业改善客户管理活动的具体策略。本文介绍了客户价值的定义,阐述了客户价值与信息技术应用、科学决策、精准营销等研究领域之间的密切联系,指出相关研究具有重要的理论价值和实践意义。在此基础上,对以往学者们在客户价值领域取得的研究成果和管理实践进行了综述,指出企业管理活动在客户价值衡量方面有着迫切需求,需要建立起有效的客户价值动态衡量模型。首先,本文利用因子分析法对客户价值的构成要素进行了探索性分析。研究发现,所有因素中,客户的“最近购买时间”对客户价值的影响比重超过了50%,是动态衡量客户价值的关键。在此基础上,本文首次从理论上证明了“最近购买时间”具有递减特性,即随着最近购买时间间隔的增加,客户的生存概率逐渐变小。这一研究发现可以帮助企业识别不同类型客户的生存概率,从而分析出客户价值的差异所在。其次,在动态变化的市场环境中,利用传统模型拟合出的客户生存概率很难适用于企业的未来需求,针对这一问题,本文应用Bayes统计中的后验估计法构建了客户生存概率模型,利用历史数据获取客户的先验知识,结合当前发生的信息,可以动态地对各参数进行后验估计。通过本文提出的模型,可以求得每一个客户在各个时期的生存概率曲线,从而实现客户价值的动态衡量。区别于传统客户价值计算方法要求客户首次购买行为的起始时间一致,本文模型可以选取任意时间点展开生存概率测算,扩展了客户价值理论在实践中的适用范围。再次,本文利用时间序列模型预测客户未来每次的购买量,弥补了传统方法使用均值进行预测的不足。研究发现,本文提出的研究模型不仅是对传统客户购买预测理论的完善与丰富,在企业实际应用中可以大大改善客户购买量的预测质量和精度,从而提高客户价值计算的精确性。最后,本文在识别客户价值构成要素、衡量客户生存概率和预测客户购买量的基础上,增加价格和成本等重要客户价值要素,构建了企业客户价值动态衡量模型。与传统客户价值计算方法进行比较,本文模型在客户生存概率、客户购买量两个方面的预测精度都有了较大改善。同时,较之于传统方法只能进行同类型客户价值的计算,应用本文模型可以分析个体客户价值的具体情况,实现了理论和应用两个层面的创新。利用企业客户价值的动态衡量模型能够帮助企业准确分析客户的价值构成,为客户的精细化管理提供数据决策支持,从而提高企业的管理能力与核心竞争力。更多还原

【Abstract】 The computation of customer value, as the theoretical basis of differentiated service, has attracted widespread attention from enterprisers and scholars at the time of "customer-centric" business philosophy. On the basis of data mining technology, Bayes, RFM and Pareto/NBD and using research methods such as literature review, investigation and data mining modeling, this dissertation has done in-depth research on dynamic measurement of customer value and provided a specific strategy for enterprise to improve customer management activities. This dissertation was done from the perspective of elements of the customer value, customer behavior characteristics, customer behavior prediction.This dissertation introduces the definition of customer value, expounds the relationship between customer value and information technology application, scientific decision-making, precision marketing research, and emphasizes the theoretical and practical importance of related research. Then, this dissertation makes a review of the existing studies and management practices on customer value, and points out an urgent need of customer value measurement in the enterprise management activities, which means effective dynamic measure model of customer value is to be set up.Firstly, this dissertation does an explanatory analysis about the components of customer value using factor analysis and finds the factor, customers’"recency ", is the key of dynamic measuring of customer value, whose influence on customer value has amounted to more than 50%. Furthermore, this paper is the first to prove that the survival probability of customers decreases with recency. This finding can help enterprises identify different types of customers’survival probabilities so as to analyze the differences of customer value.Secondly, this paper constructs the customer survival probability model using the posterior estimate method of Bayes statistic in order to solve the problem that customer survival probability fitted out from traditional model has difficulty to meet enterprises’future demands in the changing market. Using historical data to get customers’prior information, this method can combine current information to make dynamic posterior estimate of various parameters. This model can obtain every customer’s curve of survival probability in each period in order to achieve dynamic customer value measurement. What’s more, this model can do survival probability estimates at any time point, expanding the application of the customer value theory in practice. Instead, traditional customer value computation requires the starting time of the customer first purchase behavior to be identical.Thirdly, this dissertation uses the time series model to predict customer’s future purchases each time, which overcomes the shortcoming of the traditional method using the mean forecast. We find the model proposed in this dissertation not only enriches the traditional customer purchase prediction theory but can greatly improve the prediction quality and precision of customer purchases in enterprise’s application, so that the precision of customer value calculation is improved.Finally, after identifying the elements of customer value, measuring customer survival probability and predicting customer purchases, this dissertation adds other important customer value elements which are prices and costs. Thus the dynamic measurement model of enterprise customer value is constructed. Comparing with the traditional customer value calculation method, this model has greatly improved the prediction accuracy of customer survival probability and customer purchase. At the same time, to be different from the traditional method only to calculate the same type of customer value, this model can analyze the specific circumstances of the individual customer value, which makes the innovation in both theory and application. The enterprise customer dynamic value measurement model can help enterprises accurately analyze customer value constitution, providing data support for the fine management of customers, so as to improve the management and core competence of enterprises.更多还原