From Churn Models to Churn Solution

From Churn Models to Churn Solution

DOI: 10.4018/978-1-4666-6288-9.ch008
OnDemand PDF Download:
List Price: $37.50


This chapter explains churn model classification, describes techniques for developing predictive churn models, and describes how to build churn segmentation models, churn time-dependent models, and expert models for churn reduction. Analysts (readers) are shown a holistic picture for churn modeling and presented an analytical method with techniques described as elements that could be used for building a final churn solution depending on current business problems and expected outputs. There are numerous ways for designing final churn models (solutions). The first criteria is to find solutions that will be in line with business needs. The problem is not applying some data mining technique; the problem is in choosing and preparing appropriate data sets. Applied techniques should show holistic solution pictures for churn, which are explainable and understandable for making decisions, which will help in churn understanding and churn mitigation.
Chapter Preview

8.1 Introduction

Classification of churn models is easy task in way of classifying it by functionality in situation of observing specific model for itself, which should solve some specific task in domain of churn. In that case models could be aimed (classified) as churn prediction models, survival models, profiling models, segmentation models etc. In practice it is very rare situation in which only one type of model is convenient for designing final churn solution. It is often solution, which demands chaining various model types depending on specific situation.

Generally speaking appropriate classification of churn model is by their dominant characteristics. A real life churn project uses several methods, but they usually have one central point/task around which whole solution is build. It implies situation where company has need for predictive churn solution, and during solution development process, additional analysis related to churner profiling and customer life period could be done. In general it is still predictive churn model, but it also contains additional values important for decision support. It is not always possible to make classification taking in consider one aspect. Precise classification could be done taking in consider different aspects.

Regarding dominant characteristics basic churn model classification is:

  • Predictive churn models

  • Time dependent churn models (survival, seasonal)

  • Segmentation churn models

  • Profiling churn models

A predictive churn model gives assessment about churn probability in future period of time. They calculate probability that certain subscriber/buyer will commit churn in determined future period of time. These types of models are often widely accepted synonym for churn models.

Time dependent churn models contains time component, which is crucial for the final model construction. In this category belongs a survival model, which calculates survival rates, or models, which primarily analyze temporal subscribers/buyers data and produce time dependent patterns (seasonal regularities, temporal patterns, events related to temporal appearance.) A segmentation churn model has lot in common with regular segmentation models, with one significant characteristic that it makes segmentation on churner data sample. It implies that segmentation churn models try to find mutual characteristics into samples divided by segmentation algorithm. This approach could be extremely useful, because it assume that within churners exists differentiation. Revelation of dominant characteristics within different segments, and existence of segments within churner population could be key for finding solution for churn mitigation. Existence of unique population without possible differentiation also provides useful information for making adequate churn mitigation policy. It could not be determined in advance does segments exists or not, but conscious about their existence could also have influence on further solution development. For example, in case where segments within churner population exist, it is possible to decide to make different predictive churn models for each segment. It depends on analytical aims and project task related to churn.

Profiling churn models on first sight could be wrongly identified as segmentation churn models. They have common points but it is different type of models. Profiling models gives holistic picture of typical churner as member of churner population or as a member of the segment recognized thought segmentation process. Profiling models does not lay on social demographic data only; they also could be constructed by using behavioral and temporal characteristics. Profiling models could provide information how different is for example members of each segments recognized thought segmentation process. It could be trigger for decision about churn mitigation policy, or further analysis processes. Profiling analysis could include variety of data mining techniques.

Regarding determination about churn commitment classification could be done by:

  • Soft churn models

  • Hard churn models

Complete Chapter List

Search this Book: