Look at what procurement technology companies are actually building. Raising and processing POs. Three-way matching and invoice queries. Chasing approvals. Supplier onboarding paperwork. RFQ admin. Contract admin. Tail spend buying. Savings reports. Every one of them is a task procurement should never have been known for in the first place.
Tom Mills named the uncomfortable consequence recently. By letting the category define itself this way, the profession has effectively broadcast the message: that is what procurement did, AI is here, we can be replaced.
His counter-argument is that procurement was never meant to be a processing function. It was meant to fight for the business, help it take innovative risks safely, be the sensible one in the room when things got hard. Spend strategists. Value architects. Which leads to the question nobody in the vendor pitch is asking. Once AI takes the admin, what is left, and what capability is your team actually building?
The responses came from buyers, suppliers, founders and former CPOs. What emerged was broad agreement on the diagnosis and one serious disagreement about whether AI can even do the work it is being sold for.
What Actually Went, and What Stayed
The most valuable contribution came from someone whose entire job was on the list. Jenson G Sam, a supply chain analyst working RFQ and sourcing for a US client, described the split with precision.
“RFQ admin was my whole job, so I will take that one. AI did take a real piece of it off me. From a part number alone it could tell me the manufacturer, pull specifications out of datasheets, and check the number against several distributor listings at once.”
Then the part that matters. “What it never touched was the judgment. Whether the cheaper sea freight option was right depended on whether anything was actually waiting on the part. Why a price had moved since the last quote was a question I had to go back and ask a human. The admin went. The judgment stayed where it was.”
That is the cleanest first-hand account of the boundary anyone has offered. Lookup, extraction and cross-checking are gone. Knowing whether a lead time matters this week, and finding out why a price moved, remain conversations.
Cengiz Gezer, who has spent 25 years placing global buyers inside Chinese factories, confirmed it from the other side of the transaction. “On the supplier side, the work that saves a buyer money is what no system captures: whether the quoted material grade matches the mill certificate, or whether the factory’s real capacity is already booked by someone else. Those calls take judgment and a conversation with the floor, and they decide the margin long after the PO is raised.”
Two different continents, two different roles, the same conclusion. The margin is decided by information that does not exist in any system.
The Dissent: AI Is Weakest Exactly Where Procurement Lives
Mike Wynn, a partner at Advanced Purchasing Dynamics, offered the sharpest challenge in the thread, and it cuts at the premise rather than the ambition.
“AI is unreliable when judgement is required, information needed spans multiple systems, data is incomplete or messy, and exceptions are common. I’d say that’s a good description of most procurement activities.”
That list deserves to be printed and taped to the wall of every procurement transformation steering committee. The four conditions under which AI underperforms are the four defining conditions of procurement work. Which means the admin automation being sold may deliver less than the business case assumes, because even PO processing and three-way matching run into exceptions, missing master data and cross-system lookups constantly.
The implication is awkward for both camps. The optimists assume the admin disappears cleanly and frees capacity. The pessimists assume the function gets automated away. Wynn’s reading suggests a messier third outcome: partial automation that handles the clean cases and dumps every exception on a team that has been cut on the assumption the work was gone.
Steam Engines and Horses
Arjen Van Berkum, an entrepreneur in contract and supplier management, supplied the analogy that frames the whole problem.
“AI is a steam engine and the work we do today is replacing the horse with a steam engine. What we are not doing is opening the new opportunities. Which is a shame.”
Mike Lander, a former group CEO and procurement director, said the same thing in business terms. “Unfortunately the easiest thing to point AI at is all the tactical stuff. We know real business results come from system level re-design focused on improved outcomes, not making the existing processes more efficient.”
Robin Say, a product leader with two decades in fintech, closed the loop. “Making an inefficient process faster doesn’t fix the underlying problem. The real value comes from rethinking how decisions get made, where value is created and what the team should focus on once the admin is gone.”
The steam engine point is worth sitting with. Replacing a horse with an engine gets you a faster cart. The economic transformation came from redesigning everything around what an engine made possible. Procurement is currently buying faster carts and calling it transformation.
Delegate, Don’t Just Automate
Max Francks, a senior solution strategist, proposed the most practical distinction in the discussion.
“The key to using AI well is to delegate to it, not just automate with it. Automation runs a fixed process end to end. Delegation is how you’d hand work to a capable team member: you give clear context and a goal, let them do the groundwork, then review what comes back before it goes anywhere.”
His point about where the saved hours go is the one most transformation plans get wrong. “The hours freed from PO admin, RFQ prep and savings reports shouldn’t just disappear into a cost target. They should go into reviewing and challenging the output, and into building relationships with stakeholders and suppliers. Agents can do the analysis behind a category strategy, but a person still has to have the conversations that turn it into something the business acts on.”
Esther Ho, an AI operations architect with fifteen years in supply chain, named the capabilities that have to be built deliberately. “Procurement teams need to deliberately build the skills that were previously crowded out: stakeholder influence, supplier strategy, commercial judgment, risk sensing, and cross-functional decision-making. Removing the admin only creates value if that capacity is redirected.”
The Perception Problem Nobody Has Solved
Matthias Svetic, a negotiation advisor, raised the obstacle that sits outside procurement’s control. “The biggest challenge might be actually changing how stakeholders perceive procurement. Even with better technology, old expectations might keep teams focused on transactional work.”
Denis Seliun, a sales and purchase manager, explained how the industry created the problem. “Tech companies started by automating POs, approvals, and three-way matching simply because transactional tasks were the easiest to turn into software, but that accidentally cemented the outdated stereotype of procurement as a back-office gatekeeper.”
Clarice Camacho, a global energy procurement and risk leader, set the standard for what replaces it. “Real procurement requires commercial understanding, experience, sound judgement and the ability to align strategies with business objectives, manage complexity and anticipate future risks. AI can support the function but it cannot replace practical experience.”
And one addition to the original list, from Chandrika Konathala, a procurement leader with EPC and energy project experience: “You missed the endless loop of status updates. It is the most daunting task I’d happily hand over to AI any given day.”
Takeaways for Procurement Leaders
Three lessons run through the discussion. First, test the automation case against Wynn’s four conditions. If the process involves judgement, multiple systems, messy data or frequent exceptions, model partial automation plus exception handling, not clean replacement.
Second, delegate rather than automate. Give context and a goal, review the output, then act. Automation that runs end to end without review is where confident errors enter the business.
Third, decide where the freed hours go before you free them. If the capacity is not explicitly redirected into supplier strategy, stakeholder influence and commercial judgement, it will be absorbed into a cost target and nothing about the function will change.
If AI took every item on that admin list from your team tomorrow, what would they do on Monday?
GSCC members access events, replays, reports, learning tracks, AI tools and the prompt library. See the full benefits at https://gscc.co/benefits and the event calendar at https://www.chain.net/c/events/.
Continue the discussion on Chain.NET, the global supply chain community, at https://www.chain.net/c/ask.