Optimization of credit processes in retail banking

Abstract

Abstract. Machine learning credit scoring technology is becoming an essential part of retail banking's process for predicting risks associated with customer loans. However, technology that improves prediction accuracy is not always an indicator of improved operational efficiency or stability of the lending portfolio. Using Sherstiuk Retail Credit Process Optimization Framework (SRCPOF) as an example, the author of the study demonstrates the incorporation of ML-based credit risk evaluation into retail banking operational workflows. The SRCPOF framework is based exclusively on contemporary peer-reviewed articles regarding credit risk assessment, use of alternative data, explainable artificial intelligence (XAI), and process analytics. The framework used balances the precision of credit risk prediction with the calibrated risk assessment (decision) thresholds and the orderly integration (slicing?) of the operational workflows. The author proposes that decision thresholds should be considered as one of the managerial optimization variables involved in balancing anticipated losses and operational load. SRCPOF uses the combination of ensemble learning, feature optimization, alternative/exogenous data, explainability, process routing, and workflows to convert credit scoring from an evaluative tool into a process optimization system. It is also proven that prediction (risk analytics) embedded in branch-level decision systems improves early risk detection, reduces the overall approval cycle time in credit decision process, enhances risk control and governance transparency, and strengthens (or improves) the portfolio. The study primarily aims at integrating credit risk modeling with business process management. It also strives to provide banking institutions a fully scalable framework that integrates business process optimization and risk management.