Managing Churn to Maximize Profits

56 Pages Posted: 9 May 2017

See all articles by Aurelie Lemmens

Aurelie Lemmens

Tilburg University

Sunil Gupta

Harvard Business School

Date Written: May 8, 2017

Abstract

Customer defection threatens many industries. To tackle defection, companies are developing sophisticated churn management strategies that involve two steps: estimating the churn propensity of each customer and their response to retention actions, and based on it, optimizing the retention campaign by offering incentives to the subset of customers who is expected to generate the highest lift in profit. While this approach focuses on the profitability of retention actions in its second step, rather than solely on churn, the estimation step essentially aims at maximizing the correct classification of churners and non-churners, not their expected profit lift.

In this paper, we show that using a profit-based loss function in the first stage can significantly improve the overall profitability of retention campaigns. Effectively, it ensures that high-profit customers’ churn is predicted more accurately than that of low-profit customers. In our empirical application, we find that using this new loss function can lead, on average, to a 62% increase in profit lift with no additional implementation cost. For a company like Verizon Wireless, this would translate into millions of US dollars of additional profits per retention campaign.

Keywords: Churn Prediction, Loss Function, Profit Lift, Stochastic Gradient Boosting.

Suggested Citation

Lemmens, Aurélie and Gupta, Sunil, Managing Churn to Maximize Profits (May 8, 2017). Available at SSRN: https://ssrn.com/abstract=2964906 or http://dx.doi.org/10.2139/ssrn.2964906

Aurélie Lemmens (Contact Author)

Tilburg University ( email )

Tilburg, 5000 LE
Netherlands

Sunil Gupta

Harvard Business School ( email )

Soldiers Field Road
Morgan 270C
Boston, MA 02163
United States

Register to save articles to
your library

Register

Paper statistics

Downloads
252
rank
116,114
Abstract Views
914
PlumX Metrics