earticle

영어

It is a common challenge for an online company to keep its subscribers engaged with its non-contractual services. Subscribers use such services for free and can stop and leave the company without any notice, sometimes in an alarming rate. The business model of such an online company requires customers remain in the system actively. It is therefore critical for such an online company to be able to identify these at-risk subscribers as early as possible and find the most effective ways of communications to win them back. Finally, after the at-risk members came back, it is necessary to understand whether these won-back members would actually become active members. To tackle these issue, working closely with StumbleUpon - an online company providing such non-contractual service, we developed a holistic approach that combines both observational and experimental data to provide guidance for customer churn management. In particular, our study contains three stages. First, using observed individual level subscriber activity data, we adopted a statistical model to forecast the churn probability of each customer to identify the at-risk ones. Second, we randomly assign these at-risk subscribers into one of the three groups: (1) the no email group, who receives no emails; (2) the regular email group, who receives two personalized e-mails that are sent regularly; (3) the special email group, who receives a single and nonpersonalized email but is specially designed for a strong message. Lastly, using observational behavioral data, we evaluate the activities of the won-back users. Our research contributes to the literature by providing a holistic approach to customer churn management by combining the two well-separated data sources: observational and experimental data.