The study focuses on addressing the problem of predicting the likelihood of discontinuing banking services using a Probabilistic Graphical Model, leveraging the bnlearn R package as a supporting tool. With the potential support of Structure Learning and the application of causal relationships, this research aims to tackle a sensitive financial issue, something that Neural Networks have yet to achieve high reliability. The primary goal of solving this problem, predicting the likelihood of users discontinuing banking services, is to serve as a foundation for future business strategy development. Probabilistic Graphical Models in general, and Bayesian Networks in particular, have significant potential in addressing this challenge. By employing an exhaustive search approach, we experiment with multiple algorithms supported by Bayesian Networks, including PC-stable, Hill-Climbing, MMHC and two Proposed Probabilistic Graphical Models to obtain results and conclude whether Probabilistic Graphical Models are viable for this problem. The accuracy scores for these models are 0.8178, 0.7988, 0.7968, 0.8278 and 0.8308, respectively. Furthermore, through this model, we were able to assess the features that potentially influence the prediction target using Bayesian Inversion, specifically identifying the likelihood of users discontinuing banking services. While we cannot deny the rapid advancement of Neural Networks, integrating Neural Networks with Bayesian Networks is a promising direction for future research.