Designing a Model for Customer Behavior Analysis in Banking Using Artificial Intelligence

Abstract

This study aimed to develop and validate a model for analyzing and predicting bank customers’ behavior using artificial intelligence. This research was basic-applied in terms of purpose, sequential exploratory mixed-methods in terms of data type, cross-sectional in terms of data collection time, and based on content analysis and meta-synthesis in the qualitative phase and descriptive-survey procedures in the quantitative phase. In the qualitative phase, the statistical population included all studies related to customer behavior in banking systems and the factors and indicators affecting its analysis and prediction, published in valid databases between 2000 and 2025. After screening through the PRISMA procedure, 84 studies were selected for final analysis. Extracted concepts were classified into components and broader dimensions. In the quantitative phase, combined data from banking customers, along with the views of banking experts and customer-relationship specialists, were used. A researcher-made questionnaire derived from the qualitative phase was applied to assess the model components. Customer behavior prediction was performed using decision tree, support vector machine, and deep learning based on Transformer architecture. The qualitative findings led to the identification of 49 concepts, 22 components, and 12 main dimensions. The identified components included individual characteristics, financial status, responsiveness, reliability, assurance, digital experience, in-person experience, transparency, security, service satisfaction, value satisfaction, offer characteristics, communications, campaign response, brand value, transaction patterns, consumption patterns, channel preference, cooperation continuity, brand advocacy, reduced use, and offer acceptance. In the quantitative phase, the decision tree model showed the strongest predictive performance, with an R² of 0.8327, MAE of 0.0421, and RMSE of 0.0532. The support vector machine model ranked second, followed by the deep learning model, although all three algorithms demonstrated acceptable predictive capability. The proposed model demonstrated that integrating demographic, financial, behavioral, experiential, transactional, and marketing-related data can provide a valid basis for analyzing and predicting bank customers’ behavior. The findings indicate that artificial intelligence algorithms, particularly decision trees, can support banks in identifying hidden behavioral patterns, personalizing services, reducing customer churn, improving campaign effectiveness, and strengthening managerial decision-making in customer relationship management.