Reversibility-Aware AI for Supply Chain Decision-Making: A Risk-Adaptive Framework for Human–AI Control

Abstract

Background: As artificial intelligence (AI) systems move from supply chain decision support towards operational execution, governance must distinguish technical capability from permissible decision authority. This study develops a reversibility-aware framework for allocating human–AI control according to decision risk, recoverability, uncertainty and consequence propagation. Methods: A design science research-inspired approach was used to operationalise four constructs on normalised scales and integrate them into an equally weighted Decision Control Score (DCS), supplemented by governance overrides. Five control modes, from HC-0 (human-controlled) to HC-4 (monitored AI execution), were evaluated using six author-constructed supply chain scenarios. The framework was compared with illustrative fixed HC-3 and risk-only baselines, followed by reversibility and weighting sensitivity analyses. Results: The framework assigned HC-4 to routine inventory replenishment, HC-3 to forecast adjustment and purchase-order quantity modification, HC-2 to logistics rerouting, HC-1 to supplier selection or replacement, and HC-0 to critical supply interruption response. Its recommendations differed from the risk-only baseline in two of six scenarios. When the other constructs were held constant, varying reversibility changed the logistics-rerouting control mode. Across four alternative weighting configurations and six scenarios, 23 of 24 classifications matched the equal-weight results (95.8%). Conclusion: The analytical evaluation illustrates how recovery feasibility can affect rule-based control allocation. Because scenario inputs, weights, thresholds and safeguards were stipulated rather than empirically calibrated, these findings do not establish operational effectiveness or predictive validity. Independent expert assessment and field validation are required before implementation.