面向采购专业人士的代理人工智能手册

How to Move from Hype to Action and Results AI is Moving Fast, AndSo Must Procurement The last three years have redrawn the boundaries ofwhat’s possible in procurement. Generative AI camefirst, showing that machines could now write, sum-marize and converse in ways that felt natural. Butthe value was narrow. Writing supplier emails fasterdidn’t make the sourcing cycle shorter. Then came AI agents. These tools promisedtask-level automation. An agent could complete anRFQ, update a supplier record or respond to routinequeries. Helpful, yes. But still reactive — dependenton instructions, limited in scope and easily derailedby exceptions. “90% of CPO surveyrespondents say they haveconsidered or are consideringusing AI agents to optimizetheir procurement operationsin the next 6 to 12 months.”1 What we’re seeing now is qualitatively different. Agentic AI systems don’t just wait for tasks. They in-terpret goals, build plans, weigh trade-offs, and thenact. They can understand a category strategy, detectwhen it’s no longer working, and recommend a shift,with supporting data. Unlike earlier tools, these systems don’t just com-plete tasks. They orchestrate decisions across sys-tems and teams, working toward defined outcomeswith minimal handholding. This is not just anotherevolution in AI capabilities. It’s a shift in how enter-prise systems operate. What Changed? Three fundamental developments enabled this shift,with each one accelerating the next. Foundation modelsbecame operational1 While early LLMs were impressive in language gen-eration, they were weak in enterprise reasoning. Thatchanged as models began ingesting multimodal data— from invoices and contracts to market indicesand supplier risk reports. Today’s models can parsethe fine print in a force majeure clause, recognizepatterns in invoice anomalies or extract supplier riskfrom external financial data. 2AI agents gainedautonomy Earlier agents were designed to follow workflows.They could complete predefined steps but lackedadaptability. Today’s agentic systems can pursue agoal, such as “identify cost-saving opportunities inindirect categories over the next quarter,” and worktoward it by exploring data, orchestrating workflows,and engaging users only when needed. 3Procurement’s role becamebroader and more exposed Procurement teams face a wide range of challenges— from tariffs and supply disruptions to ESG man-dates and shifting regulations. They must navigatecomplexity across global markets while collaboratingmore closely with finance, legal and sustainabilityteams. Static playbooks and reactive processes areno longer enough. Agentic AI arrives at a time when procurement needssystems that can adapt in real time and take the leadwhen conditions shift. From Automation toAutonomy Most procurement technology today still runson fixed rules and triggers. A system detectsthat a purchase order (PO) exceeds $50,000and routes it for managerial approval. Athree-bid policy kicks in automatically whensourcing a new supplier. These workflows arehelpful, but they’re rigid. They only handle sit-uations they’ve been explicitly programmedto recognize. Agentic AI introduces a planning layer. In-stead of waiting for inputs, it defines objec-tives, evaluates options, and acts — withmemory and feedback loops. Here’s how the evolution looks across three generations of procurement systems: In short, conventional systems follow a script. AI agents follow instructions. Agentic AI creates its ownstrategy based on your goals and your data and continuously improves it. “70% of Asia/Pacific organizations expectagenticAI to disrupt business models within the next 18months.”2 What Does This Meanfor Procurement? Agentic AI is not theoretical. In someorganizations, it is already coordinatingsourcing events, assessing supplier risk, andoptimizing award strategies, with minimal manualinvolvement. What makes it different is not just the capabilityto complete tasks, but the way it orchestratesdecisions across systems. Let’s explore how this works in practice, usingGEP’s agentic AI framework as a reference. Use Case 1: Autonomous Sourcingand Negotiation In traditional sourcing, processes diverge depending on the value andcomplexity of the purchase. High-volume, low-value buys are oftenrouted through catalogs or quick quotes. High-value strategic sourcing requires stakeholder alignment, supplierevaluation and detailed negotiation. Most systems can handle one orthe other — rarely both. Agentic AI can manage both, contextually. In Quick Quote Automation, agents use real-time requisition datato recommend a sourcing path. They identify suppliers based onperformance, pricing, and transaction history as well as initiate bidding,negotiate using AI-driven counteroffers, and issue award decisions— all the while syncing with ERP systems and ensuring compliancepolicies are enforced. In Strategic Sourcing, agents move upstream. They ingest policy rules,category strategy documents, m