Analytics of Third-Party Risk in U.S. Supply Chains: AI for Proactive Risk Scoring and Mitigation
Juliana Twum-Ampofoah *
Management Information System, Bowie State University, United States of America.
Samuel Olawole Akande
Industrial Technology, University of Central Missouri, United States of America.
Daniel Kofi Yeboah
College of Professional Studies, Northeastern University, Roux Institute, United States of America.
Attah Nnaemeka Melford
Department of Cybersecurity, Iowa State University, United States of America.
*Author to whom correspondence should be addressed.
Abstract
U.S. supply chains face escalating third-party risks across cyber, financial, operational, and ESG domains as multi-tier networks, data heterogeneity, and regulatory scrutiny expand. This study aims to synthesise how AI/ML models and data pipelines produce decision-useful vendor risk scores and to link analytics with governance for resilience and compliance. To ensure a comprehensive review, a systematic search was conducted, encompassing peer-reviewed, U.S.-anchored studies published between 2015 and 2025. These studies were retrieved from prominent databases, including Scopus, Web of Science, IEEE Xplore, ACM Digital Library, PubMed, and ScienceDirect. The selection process involved a combination of backward and forward snowballing, resulting in the identification of 15 cases that met the predetermined criteria. A review of the included papers on distress prediction reveals that text-augmented models exhibited superior performance in terms of discrimination metrics, such as AUC and PR, when compared to ratio-only baselines. In the context of inbound logistics, ETA models demonstrated a reduction in mean absolute error (MAE) and root mean square error (RMSE) when operating in the presence of disruptions. This finding aligns with the observations reported in production-adjacent cases. The review employed a combination of backward and forward snowballing, with a standardised extraction template utilised for models, data sources, metrics, and governance artefacts. However, reporting is heterogeneous: calibration is often absent, multi-tier visibility remains sparse, and drift monitoring/explainability are inconsistently implemented, with coverage bias risks for small vendors. Future research should establish open, U.S.-anchored benchmarks; standardise reporting and calibration; adopt causal and survival designs for cost-of-error and hazard; operationalise human-in-the-loop controls, provenance scoring, and drift-aware MLOps for ERP/IoT and OSINT pipelines at production scale.
Keywords: Third-Party Risk Management (TPRM), supply chain analytics, machine learning (AI), explainable AI (XAI), model risk governance