Determing Audit Materiality Using AI

The paper estimates the numerical levels of audit materiality thresholds with the sensitivity of market value to key accounting variables in publicly listed corporations. As a response to the perpetual problem of establishing the appropriate quantitative levels of materiality, the study scrutinizes the panels of the Russell 3000 index covering the 2001–2023 time span, excluding the special sector firms. Utilizing Ordinary Least Squares (OLS) regression and advanced machine methods—Random Forest and Gradient Boosting—this research elucidates how book value, revenues, earnings, and total assets influence market capitalization. Log-transformation of the variables is used to address skewness and enable elasticity-based interpretation. The results are that multiple accounting variable models reveal high explanatory power, of which Random Forest outperforms, followed by OLS then Gradient Boosting, in the complex, multi-variable scenarios. OLS, on the other hand, continues to dominate in the simple, mono-variable models. The results are that standard OLS-based materiality levels are wide in accordance with the literature, but machine methods favor lower levels, substantially so, if the machine methods reveal non-linearities, besides when adjusting for multicollinearity. The contribution of the study is twofold: first, it offers the literature the first market-based, empirically derived intervals of audit materiality, and second, it evaluates how traditional OLS and machine-learning models differ in their implications for materiality judgments, highlighting the methodological risks of relying solely on ML techniques in the presence of nonlinearities and multicollinearity. The critical limitation is the exhaustive set of all sectors, uniformly, with no segregations for special sectors.