A surrogate-based explainability framework with semantic abstraction for human-centered supply chain intelligence

As supply chain networks grow in structural and computational complexity, organisations increasingly rely on artificial intelligence (AI)-enabled optimisation systems to generate prescriptive recommendations. When these systems are embedded within supply chain digital twin environments – where real-time scenario evaluation and rapid disruption response are expected – the gap between algorithmic output and human comprehension becomes a barrier to effective, accountable decision-making. This paper proposes a surrogate-based explainability framework with semantic abstraction that enables structured, anticipatory characterisation of how an optimisation system’s decision logic shifts across operating conditions. The framework is model-agnostic with respect to the optimisation layer and operates in three stages. A surrogate-based explanatory layer approximates the optimisation response surface and produces feature-level attributions using Shapley-based decomposition. A semantic abstraction layer aggregates these attributions into operational concepts computing normalised concept relevance, scenario variability, and concept interaction metrics that bridge numerical explanation and managerial reasoning. Evaluated on a supply chain case study comprising 7’500 instances solved by a mixed-integer routing-assignment-scheduling model under baseline and supply-disrupted conditions, the framework reveals that demand and transport concepts form a stable explanatory hierarchy robust to disruption, while processing constraints – though not directly perturbed – nearly double their relative explanatory weight, emerging as the primary differentiator among feasible solutions under stress. The study identifies a distinction between operational importance and explanatory salience, demonstrating that a disrupted parameter can decrease in explanatory weight when uniform perturbation compresses cross-instance variance. By providing decision-makers with structured, scenario-comparable explanatory profiles prior to operations, the framework supports the anticipatory preparedness that contemporary supply chain resilience demands – advancing transparent and accountable prescriptive analytics for human-centred decision-making.