This paper presents a data-driven approach to predict the success of bank telemarketing calls. The study focuses on a Portuguese retail bank's data from 2008 to 2013, including 52,944 phone contacts. The goal is to model the success of selling long-term deposits using features known before the call. A semi-automated feature selection process reduced the original 150 features to 22 relevant ones. Four data mining models—logistic regression (LR), decision trees (DT), neural networks (NN), and support vector machines (SVM)—were compared using two metrics: area under the receiver operating characteristic curve (AUC) and area under the lift cumulative curve (ALIFT). The NN model achieved the best results, with an AUC of 0.8 and ALIFT of 0.7. The model was tested using a rolling windows approach, which allowed for more realistic evaluation by simulating real-world conditions. The NN model was found to be effective in identifying the most likely customers to subscribe to a deposit, achieving a 79% success rate by selecting the top half of classified clients. Knowledge extraction techniques, including sensitivity analysis and decision trees, were applied to the NN model, revealing key attributes such as the Euribor rate, call direction, and agent experience. These findings confirm the model's credibility and value for telemarketing campaign managers. The study highlights the importance of feature engineering and the effectiveness of data-driven models in improving the efficiency and success of bank telemarketing campaigns.This paper presents a data-driven approach to predict the success of bank telemarketing calls. The study focuses on a Portuguese retail bank's data from 2008 to 2013, including 52,944 phone contacts. The goal is to model the success of selling long-term deposits using features known before the call. A semi-automated feature selection process reduced the original 150 features to 22 relevant ones. Four data mining models—logistic regression (LR), decision trees (DT), neural networks (NN), and support vector machines (SVM)—were compared using two metrics: area under the receiver operating characteristic curve (AUC) and area under the lift cumulative curve (ALIFT). The NN model achieved the best results, with an AUC of 0.8 and ALIFT of 0.7. The model was tested using a rolling windows approach, which allowed for more realistic evaluation by simulating real-world conditions. The NN model was found to be effective in identifying the most likely customers to subscribe to a deposit, achieving a 79% success rate by selecting the top half of classified clients. Knowledge extraction techniques, including sensitivity analysis and decision trees, were applied to the NN model, revealing key attributes such as the Euribor rate, call direction, and agent experience. These findings confirm the model's credibility and value for telemarketing campaign managers. The study highlights the importance of feature engineering and the effectiveness of data-driven models in improving the efficiency and success of bank telemarketing campaigns.