Advances in Business Failure: A Review of Progress and Prediction Methods

ABSTRACT Business failure research has expanded rapidly, yet much of the review literature remains focused on cataloguing methods rather than explaining their economic relevance. This article organizes the literature around four analytical dimensions: predictive versus causal orientation, static versus dynamic modelling, interpretability versus accuracy, and the degree of macroeconomic integration. Combining bibliometric mapping with interpretive synthesis, the paper examines how model development reflects changing data environments, decision requirements, and regulatory constraints. The analysis shows a shift from static, interpretable models based on accounting data toward dynamic, data‐rich, and increasingly AI‐driven approaches. However, improved predictive performance does not automatically translate into economic value. Model usefulness depends on temporal robustness, decision relevance, and compatibility with governance requirements. In this context, explainable artificial intelligence (XAI) provides a means to reconcile predictive accuracy with interpretability and regulatory usability. The review contributes by integrating accounting, finance, econometrics, and machine learning into a unified analytical framework and by clarifying when different modelling approaches are economically meaningful. It shows that evaluation must extend beyond classification accuracy to include economic loss, intervention timing, and macro‐financial conditions. The paper concludes with a research agenda emphasizing external validity, multimodal systems, and decision‐oriented evaluation standards.