Asian Journal of Economics, Business and Accounting

Artificial intelligence has moved from specialised analytical support to a strategic infrastructure for banking, yet evidence of transformation remains more uneven than the prevailing innovation narrative suggests. This critical narrative review examines how artificial intelligence is changing bank operations, customer interfaces, credit allocation, fraud control, organisational capabilities, competitive structure and prudential governance across advanced and emerging economies. Live literature searches covered publications from 1 January 2000 to 31 May 2026, supplemented by necessary foundational material and authoritative regulatory sources. The synthesis distinguishes improvements in prediction from changes in decision rights, processes and business models. Evidence is strongest for narrowly specified applications, including credit-risk estimation, transaction monitoring, fraud detection and selected service tasks. Even in these domains, gains depend on data quality, class imbalance, concept drift, implementation architecture and the cost of false decisions. Evidence that artificial intelligence reliably raises bank-level productivity or profitability is newer and contradictory: recent studies report positive effects in some Chinese samples, whereas United States patent evidence indicates improved asset quality alongside higher short-run operating costs and weaker profitability. Customer-facing and generative systems extend transformation beyond prediction, but their benefits are constrained by hallucination, privacy, cyber-security, explainability and accountability requirements. Alternative data may widen access to formal finance, yet can also reproduce exclusion through proxy discrimination and opaque segmentation. At system level, common vendors, correlated models and automated responses may create concentration and procyclical feedback. The review argues that artificial intelligence should be treated as a socio-technical and institutional transformation rather than a stand-alone technology investment. Durable value requires complementary changes in data governance, workforce design, model-risk management, procurement, consumer redress and supervisory capacity. Future research should prioritise causal, longitudinal and cross-country designs, distributional outcomes, production-scale model monitoring and measurable links between governance controls and financial performance.