Explaining AI Decisions to Bank Customers: A Systematic Literature Review

The increasing usage of black-box Artificial Intelligence (AI) models in banking has caused a rise in demand for Explainable AI (XAI) methods. Bank customers, an important target audience for XAI methods, are in need of XAI methods that explain decisions made by AI models. This study focuses on reviewing model-agnostic methods which can be applied to the model of any bank. Through a systematic literature review, this study examines studies on the application of model-agnostic XAI methods in credit risk assessment and customer segmentation. The results include showcasing the methods used, categorising them into classes, and indicating their level of globality. The study found that there is some existing literature on applying model-agnostic methods to explain decisions on credit risk assessment and customer segmentation, mostly feature-based.