Work Towards Using Micro-services to Build a Data Pipeline for Machine Learning Applications: A Case Study in Predicting Customer Churn

Abstract

Customer churn describes the situation where a customer stops using a certain supplier or vendor and moves to a competitor. Customer churn is an expensive problem for companies because obtaining new customers is more costly than retaining existing customers. If sufficient quantities of data is available, then neural networks can be used to make a reasonable prediction of when customers are going to churn. If this prediction happens early enough, before the customer does in fact churn, then the company can take preemptive steps in an effort to retain a customer who is showing danger of churning. However, a neural network can not stand in isolation. Data needs to be obtained. The data must be manipulated into an appropriate format. Neural networks need to be trained and tested. All this needs to be done before the neural network can make any predictions. This paper describes work in using micro-services to build this pipeline of data. The pipeline is initiated with data from the company's sales database. The pipeline is terminated with messages to the company's outbound customer contact center so that human-to-human calls can be made in order to persuade the customer not to churn.