Credit-Risk Indicators and Perceived AI-Enabled Early-Warning Effectiveness: A PLS-SEM Study and Conceptual XGBoost–SHAP Architecture

Although small and medium-sized enterprises (SMEs) are essential to economic growth, Saudi Arabia’s Vision 2030 has expanded their access to financing, increasing the need for effective credit-risk management and early financial-distress detection. While artificial intelligence (AI) and machine learning (ML) have shown considerable potential in credit-risk assessment, limited research has examined how banking professionals’ perceived reliance on financial, behavioral, and firm-characteristic information is associated with organizational AI/ML use and perceived early-warning effectiveness in SME lending. This study examines these relationships using survey data from 118 banking professionals in Saudi Arabia and partial least squares structural equation modeling (PLS-SEM). The results show that Perceived Reliance on Financial Indicators (β = 0.215, p = 0.024), Perceived Reliance on Behavioral Indicators (β = 0.266, p = 0.023), and Perceived Reliance on Firm-Characteristic Information (β = 0.242, p = 0.020) are significantly positively associated with Perceived Organizational Use of AI/ML. Among these antecedents, Perceived Reliance on Behavioral Indicators has the largest estimated standardized path coefficient (β = 0.266), although differences among the three coefficients were not formally tested and should not be interpreted as establishing relative superiority. In turn, Perceived Organizational Use of AI/ML is significantly positively associated with Perceived AI-Enabled EWS Effectiveness (β = 0.723, p < 0.001) and partially mediates the relationships between the three credit-risk information domains and perceived early-warning effectiveness. Building on these findings and the relevant machine-learning literature, the study proposes a conceptual XGBoost–SHAP architecture for future borrower-level implementation and validation. No XGBoost model is trained or validated, and the survey findings do not constitute empirical validation of actual SME default or fraud prediction.