Every direct procurement organisation has them. The handful of people who simply get it. They can look at a supplier, a spec, and a market signal and know — almost instinctively — what the right move is. They've negotiated through shortages, qualified second sources before anyone asked, and spotted the cost creep that didn't show up in a report until it was too late.
These people are the difference between a function that reacts and a function that anticipates. And yet, for all their value, most organisations have never found a way to do the one thing that would matter most: scale them.
That, I'd argue, is the real challenge facing direct procurement today. Not a shortage of data. Not a shortage of tools. A shortage of scaled judgment.
Why the old playbook stopped working
For most of the last two decades, the answer to "how do we get better at procurement" was some version of "give people more visibility." More dashboards. More spend cubes. More supplier scorecards. The logic was reasonable: if we can just see everything, better decisions will follow.
But visibility was only ever half the equation. Direct procurement teams today are not short on information — they're drowning in it. Spend data sits in the ERP. Supplier performance sits in a quality system. Specs sit in engineering. Market intelligence sits in someone's inbox. The context exists, but it's fragmented, and a fragmented picture doesn't produce good decisions. It produces busywork.
Here's the uncomfortable truth I've come to believe: context without reasoning is just noise, and reasoning without human judgment is just risk. More data alone was never going to close the gap. What was missing was the ability to connect the dots and the wisdom to know which dots mattered.
The loop that changes the equation
What's genuinely different now — and what I think will reshape how direct procurement operates over the next few years — is the emergence of a working loop between four things that used to live apart: context, reasoning, human judgment, and learning. Each one is necessary. None is sufficient on its own. Together, they compound.
Context
It starts with bringing the fragments together. The supplier and their history. The spend and where it's trending. The spec and its constraints. The contract, the lead times, the quality record, the alternatives. Not as a static report, but as a living, connected picture of a decision that needs to be made.
This is the unglamorous foundation, and it's where most transformation efforts quietly stall. But it matters, because every layer above it inherits its quality. Reasoning over fragmented context just produces confident nonsense faster.
Reasoning
With context in place, AI can do something genuinely useful: connect signals across sources and propose a next best action. Not "here's a chart, you figure it out," but "this supplier's lead times are slipping while volumes are rising and a qualified alternative exists — here's what I'd consider, and here's why."
This is the layer that has changed most. Reasoning that used to require an experienced buyer's full attention can now be drafted in seconds, across thousands of parts and suppliers at once. It doesn't tire, it doesn't forget the part it qualified eighteen months ago, and it doesn't only look at the categories that happen to be on this quarter's priority list.
Human judgment
And then — deliberately — it stops and hands the decision to a person.
This is the part I want to be most clear about, because it's where a lot of the conversation about AI in procurement goes wrong. The goal is not to remove the expert from the loop. The goal is to put a far better-prepared decision in front of them. Reasoning proposes; people decide. Your experts bring the things the system can't: the relationship with the supplier, the read on where the business is heading, the commercial nuance, the accountability.
Machines are good at consistency and scale. People are good at judgment in the face of ambiguity. A system that respects that division of labour is far more powerful than one that pretends either side can do it all.
The learning loop
The final piece is what ties it together and what makes the whole thing compound rather than plateau. Every decision a human makes — accept, reject, adjust, escalate — feeds back into the system. The reasoning gets sharper. The proposals get more relevant. And critically, the judgment of your best people stops being trapped in their heads and starts becoming a shared organisational asset.
This is how one buyer's hard-won expertise becomes the baseline for the entire team. The instinct that used to walk out the door at 5pm — or worse, at retirement — gets captured, refined, and made available to everyone. Best practice stops being something you write in a policy document that no one reads and starts being something embedded in the way decisions actually get made.
Why direct procurement is where this lands hardest
This matters everywhere, but I believe direct procurement is where it lands with the most force — for the simple reason that the stakes are highest.
In direct, procurement decisions are inseparable from the product, the margin, and the ability to ship. A supplier issue isn't an inconvenience; it can stop a line. A cost movement doesn't just dent a budget; it moves the bill of materials and the gross margin. The decisions are more technical, more interdependent, and less forgiving of error.
That complexity is exactly why scaled judgment is so valuable here. The problems are too numerous and too interconnected for even your best people to cover by hand, and too consequential to leave to automation that doesn't know when to stop and ask. The loop — context, reasoning, judgment, learning — is built for precisely this kind of environment.
What it actually delivers
It's tempting to frame all of this in terms of speed, and yes, teams that build this loop move faster. But speed is the least interesting part of the story.
The real impact is consistency at scale. Best practice applied across every category, not just the ones with a strong buyer attached. Risk spotted earlier, across the whole supply base rather than the squeaky wheels. Decisions that improve over time because the system is learning from every one of them. Experienced people spending their time on the judgment calls that genuinely need them, instead of assembling the context to make those calls in the first place.
And because the learning loop compounds, the advantage doesn't stay flat. A team that's a little better this quarter is meaningfully better a year from now, because every decision in between made the system — and the people using it — sharper.
The choice in front of procurement leaders
I don't think the question for procurement leaders is whether AI will be part of the function. That's settled. The more interesting question is what shape it takes.
You can deploy tools that try to automate the expert away — and inherit the risk that comes with removing judgment from consequential decisions. Or you can build the loop: rich context, capable reasoning, human judgment at the centre, and a learning mechanism that turns every decision into compounding capability.
The teams that choose the second path won't just keep up. They'll turn their best people's judgment into an organisational advantage that grows on its own — and in a function as unforgiving as direct procurement, that's the advantage that will matter most.
Direct procurement has always run on the judgment of a few exceptional people. For the first time, we have a credible way to scale it. The organisations that seize that opportunity will define what good looks like for the next decade.