Intelligence Layering in Accounting Information Systems: A Theory of Progressive AI Embedding in Financial Reporting

Abstract

Artificial intelligence (AI) is increasingly embedded in accounting information systems (AIS), yet much of the emerging literature conceptualizes AI either as a collection of discrete technologies or as an organizational innovation whose adoption can be explained through established technology-acceptance and technology-adoption models. Such approaches provide limited theoretical explanation of what occurs between the decision to adopt AI and the emergence of an AI-enabled accounting information system. This paper develops intelligence layering as a theoretical construct that explains how AI capabilities are progressively embedded within established management information systems (MIS) and AIS architectures. Intelligence layering is defined as the progressive embedding, integration and orchestration of AI-enabled capabilities into an existing information-system architecture through which the system acquires enhanced capacities for automation, learning, language interpretation, prediction and adaptive decision support while remaining subject to human, organizational and governance controls. Drawing on the Technology Acceptance Model (TAM), Technology–Organization–Environment (TOE) framework, Resource-Based View (RBV), and socio-technical systems theory, the paper distinguishes adoption antecedents from the post-adoption mechanism through which AI changes information-processing capability. A multidimensional model is proposed comprising infrastructural compatibility, data readiness, integration depth, automation intelligence, learning intelligence, language intelligence, predictive intelligence, human interpretive integration, and governance integration. Nine propositions specify relationships among enabling conditions, intelligence layering, professional interpretation, governance and AI-enabled financial-reporting capability. A seven-level maturity model is proposed, ranging from conventional rule-based AIS to adaptive intelligent AIS. The paper also differentiates intelligence layering from a recently published multi-layered framework for AI in financial reporting by positioning layering not as an organizing taxonomy of AI themes but as a theoretical and architectural mechanism of post-adoption integration. Finally, the paper proposes an empirical operationalization strategy and research agenda. The contribution is to provide AIS scholarship with a middle-range theoretical mechanism linking technology adoption and resource conditions to progressive transformation of accounting information-processing capability.