EXPLAINABLE ARTIFICIAL INTELLIGENCE AND MANAGERIAL DECISION QUALITY: TRUST, HUMAN OVERSIGHT, AND ACCOUNTABILITY IN HYBRID DECISION SYSTEMS

Abstract

Artificial intelligence is increasingly used to inform recruitment, credit, procurement, risk assessment, resource allocation, and strategic planning. The managerial usefulness of these systems depends not only on predictive performance but also on whether decision makers can understand, appropriately trust, and critically appraise their recommendations. This conceptual integrative review investigates how explainable artificial intelligence (XAI) affects managerial decision quality in hybrid human–ai systems. Building on work related to explainability, trust in automation, algorithmic aversion, and organisational decision structures, it separates explanation quality from explanation availability and distinguishes calibrated trust from generalised confidence. The paper contends that explanations can enhance decision quality by facilitating comprehension, enabling error detection, supporting justification, and promoting organisational learning. At the same time, explanations may also lead to cognitive overload, automation bias, superficial rationalisation, or misplaced confidence when they are technically complex, unstable, or disconnected from managerial tasks. Human oversight is therefore treated as an active organisational capability that involves authority, competence, time, and responsibility, rather than as mere nominal presence of a person in the decision loop. Five propositions link explanation quality, task expertise, decision stakes, calibrated trust, and accountability to decision quality. The paper develops a staged model of responsible managerial decision support and proposes a 2 × 2 experimental design comparing explanation conditions with human-review conditions. It advances management research by reframing XAI from a technical attribute to a governance mechanism and by showing that the goal is not maximal trust in AI but rather appropriate reliance under conditions of uncertainty and contestability.