This capstone presents a proof-of-concept (POC) agentic artificial intelligence (AI) system to transform the sourcing process for a global healthcare Sponsor. The current sourcing pipeline is constrained by fragmented data, reactive renewals, and reliance on individual judgment, leading to inefficiencies and operational risk. Against a baseline of 71 calendar days for an end-to-end sourcing cycle, the proposed solution targets a 35–40% cycle time reduction, bringing the estimated cycle time to 35–49 days. The system is a modular, multi-agent architecture comprising 12 specialized agents across two integrated parts: Part 1 focuses on generating and prioritizing sourcing opportunities before sourcing procedures begin, while Part 2 executes end-to-end Source-to-Contract (S2C) workflows aligned to the Sponsor's Desktop Procedures (DTPs), specifically DTP-01 to DTP-06, for the Information Technology (IT) Infrastructure category. The architecture includes five pre-procedure agents for opportunity generation and prioritization, and seven execution agents for S2C workflow orchestration and decision support. It combines rule-based logic, retrieval-augmented generation, lightweight machine learning (ML), and template-driven drafting to support dynamic sourcing decisions. A human-in-the-loop design ensures oversight while a shared learning-loop architecture captures reviewer edits, overrides, and validation feedback to continuously refine scoring logic, retrieval relevance, and routing decisions over time. Evaluation across simulated scenarios demonstrates an 81.9% AI Reliability Rate and 85.6% Signal Attribution Accuracy for Part 1 under clean data conditions, and a 78% First-Pass Acceptance Rate with 85% Knowledge Base Retrieval Relevance for Part 2. Expected outcomes include prioritized sourcing pipelines, faster sourcing execution, and a scalable foundation for enterprise adoption, with measurable improvements in speed, consistency, transparency, and continuous learning capability.
