Adaptive AI-Driven Supplier Risk Scoring System in Procurement

Thesis Abstract

The study addresses the pervasive risk management gaps in supplier networks within large-scale procurement by developing an adaptive, AI-driven supplier risk scoring system that dynamically assesses and updates risk profiles in real time. The central aim is to design, validate, and operationalize a decision-support tool that integrates machine learning, network analytics, and domain-specific risk factors to enhance procurement resilience and supplier collaboration. Specific objectives include (1) identifying a comprehensive set of quantitative and qualitative risk indicators across financial, operational, supplier diversity, geopolitical, and cyber-security domains; (2) developing a hybrid predictive model that updates risk scores using incremental learning techniques and feedback from procurement outcomes; (3) evaluating model performance against traditional weighted-scorecard approaches using historical procurement data; (4) analyzing the impact of the AI-driven scoring system on decision lead times, supplier selection accuracy, and disruption incidence; and (5) formulating governance and ethical guidelines for AI-enabled procurement risk management. The study adopts a pragmatic mixed-methods design anchored in the resource-based view and transaction cost theory to justify the strategic value of risk intelligence in supply networks and to interpret model outputs for procurement professionals. The population comprises procurement professionals, supplier managers, and risk analysts within multinational manufacturing firms. A purposive sampling frame identifies 12 organizations with diverse supplier bases, and within each organization a stratified sample of 5–7 procurement teams (n ? 72 individuals) is surveyed and interviewed to triangulate quantitative findings with qualitative insights. Data collection combines (a) archival procurement data spanning five fiscal years (approximately 1.2 million transaction records, supplier performance data, and disruption logs) and (b) primary data from a structured 36-item survey measuring perceived risk dimensions, decision quality, and system usability, complemented by semi-structured interviews with 20 key informants. Instruments are validated through content validity by risk management scholars and pilot-tested with a subset of 12 respondents. The analytical framework employs (i) regression analysis and machine learning techniques, including gradient boosting and recurrent neural networks, to build and adapt the risk scoring model; (ii) time-series analysis to examine score stability and sensitivity to external shocks; (iii) network analytics to capture inter-supplier dependencies and cascading risk effects; and (iv) thematic analysis of interview transcripts to elucidate practitioner interpretations and governance concerns. Model performance is benchmarked against traditional scorecards using metrics such as AUC-ROC, F1-score, mean absolute error, and decision accuracy in supplier selection across disruption scenarios. Expected findings indicate that the adaptive AI framework achieves superior predictive accuracy and faster update cycles (real-time to hourly) compared with static methods, reduces undisrupted supply days by 12–18%, and improves procurement decision speed by 15–20% without compromising risk controls. The study anticipates identifying key drivers of model performance, including feature engineering of financial health signals, supplier resilience indicators, and event-driven reweighting schemes, as well as potential biases and ethical considerations in automated decisions. The contribution to knowledge encompasses (a) methodological advancement in integrating adaptive machine learning with network-based risk analytics in procurement; (b) empirical evidence on the effectiveness of continuous learning systems in supplier risk management under varying macroeconomic conditions; and (c) a practical, governance-oriented framework outlining data stewardship, transparency, and human-in-the-loop requirements for AI-enabled procurement. The main conclusion posits that an adaptive AI-driven supplier risk scoring system substantially enhances procurement resilience by delivering timely, context-aware risk assessments that support proactive supplier development and contract strategy. Recommendations include deploying modular AI components for incremental integration, establishing AI governance boards to oversee model drift and fairness, investing in cyber and data privacy measures for supplier data, and pursuing longitudinal studies to assess long-term impact on supply chain performance and sustainability metrics.