From Predictive Analytics to Agentic Decision Intelligence: A Governance-Aware Framework for Human–AI Management Systems

Abstract

Organizations are moving beyond dashboards and predictive models toward Artificial Intelligence (AI) systems that can interpret goals, plan multi-step work, invoke digital tools, and initiate actions. Yet management research still lacks a coherent model for deciding when such autonomy creates value and when it creates unacceptable organizational risk. This conceptual article develops the construct of Agentic Decision Intelligence (ADI) to describe governance-bounded AI-enabled decision systems that combine analytical inference, Large Language Model (LLM) reasoning, organizational knowledge retrieval, tool use, and monitored action. Drawing on research in business analytics, human–AI collaboration, algorithmic management, explainability, trust, autonomous agents, and AI governance, the paper develops a five-layer architecture linking organizational sensing, evidence construction, decision reasoning, controlled execution, and learning. It then introduces a decision-rights model that assigns human and machine authority according to consequence severity, environmental uncertainty, reversibility, and evidentiary strength. Four governance mechanisms, provenance, permissioning, escalation, and continuous assurance, are integrated into the architecture so that autonomy becomes conditional rather than binary. The resulting framework explains why agentic systems should not be evaluated only by predictive accuracy or task completion: managerial value depends on the joint quality of decisions, actions, accountability, and organizational learning. The article contributes a vocabulary for distinguishing analytical, generative, and agentic capabilities; a governance-aware architecture for management analytics; six testable propositions; and a staged implementation roadmap. These contributions reposition management analytics from insight production toward accountable decision execution and provide a research agenda for designing human–AI systems that are both operationally useful and institutionally governable.