Profit-driven pre-processing in B2B customer churn modeling using fairness techniques

Abstract

This paper proposes a novel approach to enhance the profitability of business-to-business (B2B) customer retention campaigns through profit-driven pre-processing techniques, deviating from the traditional focus on in- and post-processing methods. Our study explores the effectiveness of three pre-processing techniques—massaging, reweighing, and resampling—derived from fairness literature. We evaluate these techniques alongside a baseline model and three state-of-the-art in- and post-processing methods using the EMPB and a newly introduced metric, the Area Under the Expected Profit Curve (AUEPC). Our findings demonstrate that reweighing and resampling consistently outperform baselines up to a 49% profit increase. Furthermore, compared to state-of-the-art algorithms, reweighing and resampling methods surpass in-processing techniques and perform favorably against post-processing methods, particularly at optimal customer contact rates. However, post-processing methods are preferred under budget constraints. This study contributes to the current literature by offering a simpler, model-agnostic, and less computationally expensive framework for profit-driven churn modeling in B2B contexts.