HERA-PRO: Prescriptive Rating Optimization Under Explainable Artificial Intelligence

This paper introduces HERA-PRO, a prescriptive extension of the Hierarchical Artificial Intelligence Rating Architecture (HERA, Hierarchical Engine for Rating Assessment) designed to support corporate rating improvement strategies through Explainable Artificial Intelligence (XAI). Building on the predictive framework proposed in Falavigna (2026a) and the explainability layer developed in Falavigna (2026b), the proposed approach moves beyond rating forecasting and focuses on identifying actionable interventions capable of increasing the probability of obtaining a more favorable rating class. The methodology combines hierarchical Random Forest classifiers and counterfactual optimization procedures within a unified decision-support framework. Firms are first classified through a network of sector-specific and rating-specific predictive models. Subsequently, a prescriptive optimization layer is introduced to determine the minimum financial adjustments required to achieve a target rating outcome. The proposed procedure explores alternative counterfactual configurations of firms’ financial indicators and identifies both single-variable and multi-variable adjustment strategies capable of producing rating upgrades. Unlike conventional rating forecasting systems, which primarily focus on prediction accuracy, HERA-PRO provides decision-makers with actionable recommendations by quantifying the minimum changes necessary to cross rating-classification thresholds. The framework therefore shifts the analytical perspective from “What rating will be assigned?” to “What should change to obtain a better rating?”. By integrating prediction, explanation, and prescription within a coherent methodological framework, HERA-PRO contributes to the emerging literature on prescriptive artificial intelligence, counterfactual analytics, and decision-oriented machine learning for corporate credit risk assessment.