Optimizing Financial Ratios with AI: A Dynamic Control Framework for Credit Institutions

This study presents a new approach to the internal control in credit institutions by proposing a closed-loop framework that replaces static financial ratio thresholds with dynamic, adaptive boundaries. Compared to conventional Machine Learning (ML), which only emphasizes forecasting, the proposed approach integrates multivariable regression with multi-objective optimization to convert predictions into a real-time control mechanism. The methodology includes a composite objective function that minimizes deviations from optimal financial ratios while penalizing allocations with high predictive uncertainty. The novelty of the study lies in the dynamic normative boundary, a time-dependent reference that synthesizes historical benchmarks, model-derived optima, and market signals. A comprehensive ratio framework based on asset structure was implemented, and its applications were demonstrated through a case study of Aregak Co. (Armenia). The results identify a capital surplus of 483% above the optimum and demonstrate the framework's capacity to uncover strategic pathways for enhancing Return on Equity (ROE) by up to 14%, all within explicit risk constraints. This study establishes a viable architecture for transforming internal control from a periodic compliance exercise into a continuous process of strategic optimization.