Global supply chains remain fragile with geopolitical tensions, pandemic disruption, port congestion, and climate shocks. Conventional supplier scorecards are sluggish, passive, and rarely audit-worthy, opening up areas of blindness in risk identification and decision making. This paper provides a machine learning-ready supply analysis methodology and incorporates high-quality data governance. Autonomous supplier assessment refers to an automated judgment system that proposes calibrated probabilities and prescriptive steps of the judgment, or actions, namely: block, review, or allow, by implementing policy-as-code, a constraint by compliance requirements. The strategy unites three types of evidence: tabular data, such as lead-time volatility, OTIF performance, and defect rates; unstructured evidence, including audit reports, certificates, and contracts; and network-based features that capture the length of the tier and the risk of the community. Data are processed by entity resolution, normalization, and temporal cross-validation and leakage-safe labeling. Governance processes such as data contracts, lineage, quality SLAs, and decision logs provide accountability and audit-readiness. Trust and adoption are further increased through counterfactual explanations and human-in-the-loop triage. Experiments show that it is better for early warning of risks in delivery, quality, and compliance by combining the use of tabular and text features and graphs. The calibrated ranking strategies are more effective than the static thresholds in a limited review capacity because they can detect more adverse issues without dealing with false positives. The results reinforce that stewardship practices do not create overhead but enable resilient, transparent, and explainable autonomy. The work collectively gives methodological contributions and a business roadmap for implementing trustworthy AI in procurement.
Bonthu et al. (Sun,) studied this question.