Churn Rate Analysis and Marketing Campaigns Adjustment Based on Machine Learning

Modern technologies allow consumers to compare the variety of offers from various suppliers. Customers can change the supplier in a few clicks, which adversely affects company’s performance. This forces companies to implement churn management and customer retention programs. The aim of this work is to formulate retention strategy in a large Russian e-commerce company using churn prediction model results. The object of the study is customer churn in e-commerce companies, and the subject is the retention strategy formulation. The research objectives include: - Description of the current state of the e-commerce market in Russia. - Defining characteristics of the e-commerce market. - Customer churn problem formulation. - Overview of the methods to predict customer churn. - Overview of the customer retention strategies in marketing. - Е-commerce company case description. - Data search and preprocessing. - Predictive model creation and forecasting. - Interpretation of forecasting results. - Retention strategy formulation based on the forecast results. - Overview of the study findings. The significance of this work is explained by the fact that, firstly, small amount of literature on the topic considers the e-commerce sector, and secondly, the existing literature on the topic focuses mostly on the accurate forecasting models development, which often turn out to be heavy, complex and difficult to interpret (black box problem), while interpretability and low resource intensity are important for business. Thirdly, while creating predictive model, researchers rarely connect model results with the retention strategy. Thus, the present work will try to close these gaps. To achieve this goal, the most popular churn prediction models are compared: logistic regression, random forest, and several implementations of gradient boosting. The forecast results then used to determine factors that affect churn and to develop a customer retention program. As a result, a forecast of customer churn was developed using XGBoost algorithm and it helped to determine the portrait of a churn client. Such a customer has been using company's services for a long time, but he makes rare one-time and expensive purchases. Such churn is called partial, when the client's relationship with the company is volatile. To encourage customers to buy more often, it is proposed to modify the loyalty program, set up a newsletter to notify customers about new products, offer bonuses or discounts to customers when launching new services, and explore new channels of communication with a young audience.