Risk Analysis of Artificial Intelligence Deployment in Administrative Processes Using the Integrated FMEA-DEMATEL Approach: A Case Study of Bank Sepah, Mazandaran Province

Abstract

Today, the deployment of artificial intelligence (AI) in organizational administrative processes, particularly within the banking industry, has emerged as one of the most significant manifestations of digital transformation. Despite its advantages, including increased speed, accuracy, productivity, and enhanced decision-making, the implementation of AI in administrative processes is associated with numerous risks across data, technical, security, organizational, and governance dimensions. Therefore, the present study aimed to analyze and prioritize the risks associated with AI deployment in administrative processes using the integrated FMEA-DEMATEL approach in Bank Sepah of Mazandaran Province. In terms of purpose, this study is applied research, and in terms of methodology, it adopts a descriptive-analytical approach. Research data were collected through a literature review and expert questionnaires. The study population consisted of experts, managers, and specialists familiar with information technology, risk management, and banking administrative processes, who were selected through purposive sampling. In the first stage, 20 major risks related to AI deployment were identified and evaluated using the Failure Mode and Effects Analysis (FMEA) method based on three criteria: severity, occurrence probability, and detectability. The identified risks were subsequently ranked according to their Risk Priority Numbers (RPNs). The results of this stage indicated that the risks of “information leakage and confidentiality breaches,” “lack of model transparency and interpretability,” and “model errors and inaccurate predictions” received the highest priority rankings. In the second stage, the top ten risks were analyzed using the Decision-Making Trial and Evaluation Laboratory (DEMATEL) method to examine their causal relationships and mutual influences. The findings revealed that “absence of an AI governance framework,” “poor data quality,” and “bias in training data” were among the most significant causal and root risks. In contrast, risks such as “information leakage and confidentiality breaches” and “cyber and adversarial attacks on AI models” were primarily consequence-oriented and effect-related risks. The results indicate that effective management of AI-related risks in banks requires simultaneous attention to both critical risks and their underlying root causes at the governance, data, and model design levels. The findings of this study can serve as a foundation for designing AI governance mechanisms, strengthening information security, and improving decision-making processes within banking organizations.