Customer churn is a problem that is visible in almost all SaaS businesses. In its simplest form it is a measure of the number of customers who cancel their subscription after a certain billing period. We were tasked with the goal of identifying if a particular customer will cancel his subscription or continue utilizing the product. If a customer is identified as a potential candidate to cancel his subscription, then customised customer retention programmes can be offered to him by virtue of discounts and other product offers. In most SaaS businesses it is far more efficient to retain existing customers than to acquire new customers, and hence addressing the customer churn problem was critical to their success. To solve this problem, we had to analyse the engagement of the customer on the platform by inspecting various parameters like customer profile, product features used, spending patterns on the website etc.
We started the data acquisition phase of the project by collecting the customer sales records from their MySQL database. Some of the details that we acquired include sign in date, monthly purchases on the platform, frequently product features, duration of activity on the website etc. We then converted this multi feature dataset into appropriate feature vectors to make it amenable to ML training. Predicting if a customer will continue to utilise the products is a classic example of a Classification problem in Supervised Machine Learning. We then studied the performance of various Classification algorithms like k-NN, Linear SVM and AdaBoost on this vectorised data set. It was observed that the Linear SVM classifier performed the best in terms of both accuracy and execution time. We packaged the ML model as a self contained analytic tool that our clients can use on a periodic basis to identify and address the customer churn problem effectively. The analytic tool ingested the MySQL records and produced reports that predict customer churn.
We are able to help our clients to increase their customer retention by nearly 10 % within a short period of 6 months. This project was developed within a 8 week time frame. The Linear SVM based ML model was predicting customer churn at an accuracy of 97%.