Optimizing campaigns with uplift modeling: an empirical study in turkish financial sector

Today, artificial intelligence (AI) not only facilitates our daily lives but also serves as a critical tool in academic settings and sectors such as finance and manufacturing. In Türkiye, with the increasing competition in the finance sector, AI is actively used in marketing. While approaches such as churn modeling help reach the right customer, new methods are being developed for more complex problems such as reaching the right customer through the right channel and by the right method. One of these methods is Uplift Modeling, which calculates the net difference between the customer's behavior under the presence and absence of a treatment. The impact of uplift modeling can be elevated when integrated with machine learning (ML). A literature review has highlighted the benefits of ML and Uplift Modeling, and particularly in Türkiye, there is a noticeable lack of direct research in this area. This study aims to calculate the uplift with ML using real data in the Turkish banking sector and contribute to the literature by predicting the customer behavior in advance, to measure the success of ML compared to the traditional methods. The methodology involves identifying the target customer group, extracting features for this group and cleaning the data, selecting the correct ML and uplift modeling approaches using metrics such as Uplift Curve and Qini Coefficient, training selected models and analyzing the results. Accordingly, six different approaches such as S-Learner, Uplift Trees have been tested using Random Forests and XGBoost for credit card sales. As a result, although they need smaller sample of customers, Modified Outcome and T-Learner methods result in at least three times better customer selection compared to the traditional methods, thus highlighting the potential efficiency of uplift modeling in future research.