Decision Support System for CRM based on an Individual Level Model

在客户关系管理(Customer Relationship Management，CRM)中，经常使用特殊规则来判断客户在非合同环境中是否活跃。例如，如果客户超过三个月没有购买，则被视为退出。然而，对于两次购买时间较长的客户，这三个月的时间框架将不适用。因此，在评估客户流失时，重要的是要考虑客户的异质性。尽管这一问题得到了Schmittlein等人的承认。(1987)，他在近20年前提出了Pareto/NBD的“计算你的客户”的框架，今天的营销需要更多的个人层面的分析。本研究提出了一个提出的模型，通过估计个人特定的参数来捕捉客户的异质性，同时在Pareto/NBD模型(泊松购买过程和无记忆退出过程)中保持对个人行为的理论上的合理假设。该模型不仅放松了这两个行为过程相互独立的假设，而且还为客户关系管理提供了有用的输出，如特定于客户的生命周期和保留率，这是在其他情况下无法获得的。将其预测性能与基准Pareto/NBD模型进行了比较。应用于扫描仪面板数据的模型扩展表明，新近频率(RF)数据与客户行为和人口统计数据相结合，可以为直接营销问题提供重要的见解，例如长寿客户是否花费更多和更有利可图。

In customer relationship management (CRM), ad hoc rules are often employed to judge whether customers are active in a "non-contractual" setting. For example, a customer is considered to have dropped out if he or she has not made purchase for over three months. However, for customers with a long interpurchase time, this three-month time frame would not apply. Hence, when assessing customer attrition, it is important to account for customer heterogeneity. Although this issue was recognized by Schmittlein et al. (1987), who proposed the Pareto/NBD "counting your customers" framework almost 20 years ago, today's marketing demands a more individual level analysis.This research presents a proposed model that captures customer heterogeneity through estimation of individual-specific parameters, while maintaining theoretically sound assumptions of individual behavior in a Pareto/NBD model (a Poisson purchase process and a memoryless dropout process). The model not only relaxes the assumption of independence of the two behavioral processes, it also provides useful outputs for CRM, such as a customer-specific lifetime and retention rate, which could not have been obtained otherwise. Its predictive performance is compared against the benchmark Pareto/NBD model. The model extension, as applied to scanner panel data, demonstrates that recency-frequency (RF) data, in conjunction with customer behavior and demographics, can provide important insights into direct marketing issues, such as whether long-life customers spend more and are more profitable.