Development of a Human–AI Decision Intelligence Framework for Fraud Detection, Risk Analytics, and Operational Resilience in U.S. Digital Banking

Abstract

The rapid expansion of artificial intelligence across U.S. digital banking has created a challenge concerning how automated analytical capability can be combined with human expertise to improve fraud detection, risk analytics, and operational resilience without weakening transparency, accountability, or decision quality. This study aimed to develop and empirically evaluate a Human-AI Decision Intelligence Framework that integrates technological intelligence, professional judgment, governance, and decision effectiveness within digitally intensive banking environments. A quantitative, cross-sectional, case-study-based design was employed using a structured five-point Likert-scale questionnaire administered to professionals representing commercial banks, digital banks, fintech organizations supporting banking, and payment or financial-technology service enterprises. Of 320 questionnaires distributed, 300 were returned and 292 usable responses were retained, producing a 91.3% usable response rate. The principal variables were Human-AI Decision Collaboration, AI-Enabled Fraud Detection Capability, AI-Driven Risk Analytics Capability, Human Oversight, Explainability, and Decision Governance, Fraud Detection Effectiveness, Risk Analytics Effectiveness, and Operational Resilience. Data analysis incorporated descriptive statistics, Cronbach’s alpha reliability testing, Pearson correlation, multiple regression, ANOVA, and diagnostic assessment. AI-Enabled Fraud Detection Capability recorded the highest mean, M = 4.15, SD = 0.57, while Operational Resilience achieved M = 4.10, SD = 0.58. AI-Enabled Fraud Detection Capability strongly predicted Fraud Detection Effectiveness, β = .39, p < .001, and AI-Driven Risk Analytics Capability strongly predicted Risk Analytics Effectiveness, β = .44, p < .001. The integrated model explained 70.9% of the variance in Operational Resilience, R = .842, R² = .709, adjusted R² = .703, F(6, 285) = 115.65, p < .001, with Fraud Detection Effectiveness, β = .28, and Risk Analytics Effectiveness, β = .25, emerging as the strongest direct predictors. All eight hypotheses were supported. The findings imply that resilient digital banking depends on coordinated AI capabilities, human judgment, explainable governance, and effective fraud and risk decisions rather than automation alone.