New Template for Retail Customer Churn Prediction

This post is authored by Xinwei Xue, Senior Data Scientist at Microsoft.

We are pleased to announce the availability of our newest template, for retail customer churn prediction. This is the latest addition to our existing set of templates which include online fraud detection, retail forecasting, text classification and predictive maintenance.

The term customer churn, of course, refers to the loss of customers for a given business. It is supremely important for any customer-centric business be it banking, telecom, retail or any other to be able to identify, a priori, which customers are likely to churn, and then take appropriate actions to retain such customers and keep their business.

This template is based on Azure ML, which is now a core part of our state-of-the-art Cortana Analytics Suite. The template provides a data schema and a solution consisting of pre-configured ML modules and custom scripts in the Execute Python Script Module for addressing this problem.

This template formulates the churn prediction problem as a binary classification problem, i.e. classifying customers as either churners or non-churners. It uses two data sources from retail: customer demographic data and customer activity (i.e. transaction) data. The template allows the user to define the churn condition, and based on that it labels the data for training, followed by feature engineering, model training and evaluation, and finally it publishes a web service.

The template has 4 steps as shown in the workflow below, each one a separate experiment. Additional details on each step are available from the Cortana Analytics Gallery here.

Although the template was designed with retail companies and retail stores in mind, the concepts and technology encapsulated in this template can be readily applied to other industries as well. We hope you’ll give it a spin, and share your comments and thoughts below.

Xinwei
Click here to subscribe to this blog.