The "Black Box" Dilemma in AI-Driven Management Accounting: Implications For Corporate Governance and Managerial Accountability

Abstract

Artificial intelligence (AI) is being used today to address various management accounting functions, including strategic cost tracking, capital budgeting and performance forecasting, by the integration of its unprecedented computational power to financial decision-making. Deep learning models, however, suffer from the "black box" effect which makes them seriously challenged for accountability of managers and corporate governance. In this paper, we draw together what is learnt from recent empirical research and case analysis and explore how the non-interpretability of models impacts strategic decisions using AI for resource allocation when those decisions are heavily reliant on AI output. Based on research on financial misstatement detection, capital budgeting applications, and public sector accountability, we illustrate how a lack of transparency in AI systems affects auditability, poses accountability risks, and leaves a burden of accountability which traditional practices are not sensitive to. The study introduces a hybrid approach “Explainable AI (XAI) plus deterministic governance plus human-in-the-loop”. The results suggest that while AI helps to improve predictive accuracy and analytical productivity, it will require organizations to create strong data governance frameworks, deploy AI interpretability tools like SHAP Value Plots, and keep advisors at the helm of key decisions to ensure accountability. The study will be of interest to anyone involved in powering AI's emergence in management accounting and provides real-world guidance on algorithmic efficiency and responsible accountability.