Designing an AI-Based Financial Advisor for Distressed Firms: A Decision Support Framework for Actionable and Accounting-Consistent Algorithmic Recourse

IEEE Access. 2026. Vol. 14. P. 20084–20099.

Elizaveta Lashkevich, Zelenkov Y.

While machine learning models have achieved high accuracy in predicting firm financial failure (FFF), they often function as “closed boxes” that fail to provide actionable guidance for decision-makers. Existing counterfactual explanation methods typically operate in the space of financial ratios (FR), neglecting fundamental accounting identities and implementation costs, thereby producing recommendations that are theoretically valid but operationally infeasible. This paper proposes a novel decision support framework that bridges the gap between predictive accuracy and managerial actionability. We introduce an end-to-end architecture that generates algorithmic recourse in atomic accounting-data (AD) space, ensuring that all recommendations strictly adhere to accounting constraints and business rules. The framework incorporates a cost-aware optimization engine that trades off default risk reduction against monetary implementation costs, presenting the decision-maker with a Pareto-efficient set of strategic alternatives. We validate the system using a dataset of 211,275 Finnish SMEs. A classification model with high discriminative power (Gini 97%, Geometric mean 94%) effectively separates positive and negative instances. A typical decrease in the probability of default along the Pareto boundary is 40-50%. The results demonstrate that the proposed framework generates diverse, accounting-consistent, and economically viable recovery strategies—ranging from tactical inventory adjustments to strategic asset reconfigurations—thereby transforming passive risk prediction into active, intelligent decision support.

Keywords: системы поддержки принятия решений, bankruptcy prediction, decision support systems, финансовые проблемы, предсказание банкротства, multi-objective optimisation, counterfactual explanations, firm financial failure, algorithmic recourse, actionable AI, алгоритмические рекомендации, контрфактуальные объяснения, действенный ИИ