PROCUREMENT MAGAZINE WHITEPAPER

Artificial intelligence has been present in procurement for years, primarily in analytical form. It has identified savings opportunities, highlighted supplier risks and generated performance dashboards. Yet most AI systems have remained passive. They inform. Humans act.

Agentic AI introduces a different paradigm. Instead of offering suggestions alone, AI agents carry out defined responsibilities.

They sense context, apply logic, execute actions and escalate exceptions when required.

An agent is not simply a feature or a chatbot layered on top of existing systems. It functions more like a digital role. It has access to data, operates under clear instructions, produces auditable outputs and follows predefined escalation paths.

Sabih Rozales, Architect at ORO Labs, describes the shift succinctly: in many organisations, AI now handles reviews autonomously and escalates to humans only when a genuine risk or exception is detected. This humanin-the-loop design does not eliminate oversight. It refocuses it. The agent performs the heavy lifting – collecting and validating information – so that human effort is reserved for highimpact judgment calls.

This model has already produced tangible results. At Novartis, thousands of purchase requisitions once required extensive manual oversight and up to five days of review time. By deploying an AI-powered PR Review Agent, the organisation reduced review times to 16 minutes, improved accuracy by 325 %, and increased processing volume tenfold without adding headcount.

These results illustrate the central promise of agentic AI: not simply efficiency, but operational leverage.

108 March 2026