AI Won’t Replace Procurement. It Will Expose Who Was Hiding Behind Process

The tool is only as sharp as the person holding it. Most procurement teams haven’t sharpened either.

Procurement has been here before. Every wave of technology has arrived with the same promise and the same threat: automate the POs, self-service the catalog, let the portal do it. Each time, the function was supposed to shrink. Each time, it didn’t. Not because the technology failed, but because a machine still doesn’t understand leverage. The latest wave, agentic AI, is louder than the last. But the underlying question hasn’t changed. Who actually does the thinking?

The debate gained traction online recently and drew reactions from procurement directors, sourcing leaders, AI consultants, and operators across pharma, energy, automotive, and FMCG. The agreement on the diagnosis was strong. The pushback on where AI actually stalls was sharper.

Exposing the Hiding Place

The most cited reframing came from Mario González, a procurement director at Tier-1 automotive. “AI won’t replace procurement, but it will expose something uncomfortable, how much of the function still hides behind process instead of judgment. The next gap isn’t tool adoption. It’s whether leaders redesign decision rights, capability, and accountability fast enough for AI to matter beyond productivity theater.”

That phrase, “productivity theater,” lands hard. It captures what most AI rollouts in procurement actually look like. Prompts that save individual hours. Dashboards that look impressive. Use cases that never connect to a real decision. The function looks busier, not sharper.

John Cross, an advisor to mining executives, drew the same line. “AI isn’t replacing procurement, it’s just exposing who actually understands leverage and who was hiding behind process. The tech gets sharper, but the edge still comes from the human holding it.”

José Gabriel Tovar Taracena, a strategic procurement manager, sharpened the principle. “AI will not replace procurement. It will replace procurement without strategic judgment. Because negotiation is not just data analysis. It is about reading risk, building trust, aligning the business, and making decisions under uncertainty.”

Where the Wall Actually Sits

The deepest operational critique came from Rabih Suleiman, an SAP Ariba and Jaggaer end-to-end specialist. He accepted the productivity layer worked, then pointed to what doesn’t. “What breaks is when you try to embed AI into the S2P process itself. Three patterns I keep seeing.”

His diagnosis was specific. “Master data. AI amplifies what you feed it: dirty supplier records, inconsistent category trees, and half-maintained contracts. The tool produces confident output on a shaky foundation.”

The second pattern was about ownership. “AI in a P2P flow needs someone who owns the exceptions. That role is often vacant.”

The third was about trust. “Teams either over-trust the output and skip review, or redo everything manually. Both kill the ROI case.”

His conclusion explained the MIT statistic without blaming the technology. “The 95 percent MIT number doesn’t surprise me. It’s rarely the tech. The operating model around it was never built.”

That observation matters because most procurement AI conversations end at the prompt. Suleiman’s framing moves the conversation to where it actually lives: data foundation, exception ownership, and trust calibration.

The Data Confidentiality Layer

Tom Hathaway, a supply chain improvement specialist and CIPS member, raised the constraint most enterprise procurement leaders feel daily. “Data security is a major worry for large procurement operations. Loading data into anything potentially ‘leaky’ in a world of hacks undercuts a duty of confidentiality to suppliers.”

Megane Morel, a procurement manager working across FMCG, pharma, and luxury, made the same point with sectoral nuance. “Not every sector can move at the same pace. Data sensitivity and confidentiality risks are real constraints.”

That constraint matters because most published procurement AI use cases assume the data can be pasted into a public model without consequence. For pharma, defense, energy, and any regulated category, that assumption breaks the moment compliance reviews start. The productivity gain is real. The deployment path is narrower than the prompt list suggests.

Jay Perkins, focused on commercial intelligence platforms, listed the cascade. “Tokens, carbon footprint, AI output accuracy, poor data leading to incorrect AI responses, hallucinations, AI not following guidelines even though they are documented in the agent, legal challenge.”

His punchline summed up the operational reality. “Right place and right use case, but the AI rollout will be won or lost on the data input.”

The Suppliers Are Using AI Too

The most uncomfortable observation came from Oksana Barabash, a procurement and supply chain professional. “We should start recognising how much AI is already being used by suppliers and stakeholders and CPOs in their communication with us and how. Are they actually reading and engaging with our points, or are we effectively speaking to a text generator?”

Her closing line was the sharpest. “Reading the room only works when there is a room, not just an email thread.”

That observation extends the original argument in a direction most procurement leaders have not yet processed. AI is not just a tool the buying side wields. Suppliers are using it to draft proposals, anticipate objections, and structure negotiations. The traditional procurement advantages, asymmetric information, slower response times for the supplier, the ability to surprise, are eroding from both sides simultaneously.

Alex Quek, a senior contracts manager in marine container shipping, captured the trajectory. “I can imagine a day when supplier and procurement are using AI to counter AI. Best AI wins.”

Encoding Beats Prompting

The most forward-looking critique came from Skillsora, a procurement platform. “The 8 prompts work for individual judgment-assist. Next layer for procurement teams: encoding those patterns as vertical agents that run across every contract, RFP, and supplier review automatically. Teams who encode their playbooks pull ahead of teams still copy-pasting prompts one document at a time.”

That point matters operationally. Individual prompting is an entry-level skill. Encoded vertical agents that run procurement playbooks across every artifact, with quality assurance from senior buyers, is the operating model where AI actually scales. Most procurement teams are stuck at the first level because the second level requires the operating-model investment Suleiman flagged.

David Račak, a procurement AI consultant, gave the practical entry point. “I would suggest to put these to AI model within company policy and ask for better prompt. Copilot rules these days if you just build a custom agent for stuff like these.”

Don’t Become the Tool

The closing principle came from Tom Ruello, VP of Sales at SpendHound. “AI provides a tool for procurement to significantly lift its status internally to a strategic thought partner. If you don’t use it, though, you will be eventually replaced. There are some pretty aggressive pitches coming from agentic AI companies out there that they will replace procurement leaders. If they catch a CFO or CEO on the wrong day, you’re in trouble if you’re not leveling up.”

Erick Andrian, a head of supply chain and procurement working on turnarounds, captured the response. “Use AI to kill the administrative ‘blank page syndrome’ so you can spend your energy on what moves the needle, like S&OP governance and TCO architecture. Master the tool to reclaim the time you need to lead.”

Pankaj Tuteja, Head of Operations at Dragon Sourcing, gave the cleanest summary. “The goal isn’t to be an AI expert. It’s to be a procurement expert who knows how to use a sharper tool.”

Takeaways for Procurement Leaders

Three lessons run through the discussion. First, the productivity layer of AI works. Eight well-crafted prompts can save hours per week on contract review, negotiation prep, and drafting. That layer is the entry point, not the destination.

Second, the operating model is where AI actually stalls. Master data quality, exception ownership, and trust calibration determine whether AI moves from individual productivity to embedded process. The 95 percent MIT statistic is an operating-model failure, not a technology failure.

Third, the suppliers are using AI too. The asymmetric information advantage procurement has historically relied on is eroding. Reading the room is harder when half the room is generating text.

Is your procurement function using AI to amplify judgment, or to perform productivity?

Continue the discussion on Chain.NET, the global supply chain community, at www.chain.net. Ask questions, join events, and access exclusive resources.