Machine Learning in Corporate Financial Decision-Making: A Critical Narrative Review of Predictive Models for Investment, Valuation, and Risk Assessment

Abstract

Machine learning is increasingly applied to support corporate financial decisions under conditions of high dimensionality, nonlinearity, and rapid information change. This critical narrative review examines how predictive models have been used across three core domains: investment decision-making, valuation, and risk assessment. Drawing on a validated corpus of empirical and conceptual studies, the review compares traditional financial models, classical machine learning methods, ensemble techniques, and deep learning architectures, and analyzes how their suitability varies by decision context. The evidence shows that ensemble and deep learning approaches frequently improve predictive performance relative to linear benchmarks when relationships are complex or sequential, yet these gains are conditional on data regime, validation discipline, and integration into actual decision processes. A structural tension persists between predictive accuracy and the transparency, governance, and regulatory requirements that shape organizational trust and adoption, particularly in risk applications. The review consolidates dispersed findings, clarifies the boundary conditions under which machine learning contributes to corporate financial decisions, and identifies the prediction-to-decision gap and explainability constraints as central unresolved issues. Scholarly and practical contributions lie in the comparative synthesis across models and decision domains and in the articulation of evidence-based priorities for more decision-relevant and responsibly governed research and practice.