How Much Should Procurement Actually Let AI Do?

At SOAR26, we put AI autonomy up for debate. The most interesting takeaway wasn't which side won—it was where the line between human judgment and machine execution should actually sit.

AI can recommend a supplier. But should it select one?

It can prepare a negotiation. But should it conduct one?

It can identify the next best action. But should it take that action without waiting for someone to approve it?

These questions are becoming increasingly real as procurement moves from AI that assists people to AI that can act on their behalf.

So at SOAR26, we put them up for debate.

We asked two procurement experts—Joël Collin-Demers of Pure Procurement and Chris Sawchuk of The Hackett Group—to debate two deliberately provocative motions around AI autonomy and negotiation.

There was an important catch: we assigned the sides. They were asked to play devil's advocate and make the strongest possible argument for positions that didn't necessarily reflect their personal views. Only after the debate did we ask them to drop their assigned personas and tell us what they actually believed.

We won't spoil who moved the room.

But underneath the arguments were several questions every procurement team experimenting with agentic AI will need to answer.

1. A Human in the Loop Isn't Automatically Governance

It feels safer to require a person to approve an AI-generated decision.

But what if that person isn't really making a decision?

Procurement already has plenty of workflows where someone reviews an approval, sees nothing obviously wrong, and clicks a button. Adding the same step after an AI recommendation doesn't necessarily create better governance.

It can simply create a human rubber stamp.

One of the strongest ideas raised during the debate was that accountability doesn't have to mean manually approving every transaction. Organizations can instead govern the policies, thresholds, business rules, and exception paths that determine what AI is allowed to do.

That changes the role of procurement.

Instead of asking, Who needs to approve every action?

We can start asking, Who owns the rules under which those actions happen?

2. Risk and Reversibility May Matter More Than “Human vs. AI”

The question of whether AI should recommend or act sounds binary.

In practice, it probably isn't.

Consider two decisions.

An AI agent identifies additional suppliers for an event and sends outreach based on established criteria.

An AI agent awards a strategic supplier agreement with significant business continuity implications.

Both involve AI taking action. But the consequences are dramatically different.

The framework that emerged at the end of the session was risk and reversibility: if an action is clearly defined, low-risk, and easy to reverse, there may be little value in requiring a human checkpoint. As decisions become higher-risk, harder to reverse, or more ambiguous, human involvement becomes more important.

That gives procurement a much more useful way to think about autonomy.

Not: Do we trust AI?

But: What level of autonomy is appropriate for this decision?

3. The Biggest Opportunity May Be Work Humans Aren't Doing Today

This was one of the most interesting ideas from the negotiation debate.

When we compare AI with humans, we tend to imagine the machine replacing an experienced procurement professional.

But that's not always the real comparison.

Procurement has finite capacity. There are transactions, suppliers, categories, and commercial terms that simply don't receive meaningful negotiation today.

So in some cases, the choice isn't:

AI negotiation vs. human negotiation.

It's:

AI negotiation vs. no negotiation.

That creates a very different business case.

AI doesn't necessarily need to outperform your best strategic negotiator to create value. It may create enormous value simply by extending structured commercial discipline to areas procurement doesn't have the capacity to reach today. The debate specifically raised opportunities around routine transactions and terms that might otherwise go unnegotiated.

The goal isn't necessarily replacing the top of the organization.

It may be raising the baseline.

4. Not Everything That Can Be Optimized Should Be Optimized the Same Way

Machines are very good at optimizing toward defined objectives.

Procurement's challenge is that the objective isn't always obvious.

Price matters. So do supply continuity, capacity, innovation, speed, risk, and supplier relationships.

And sometimes those priorities change in the middle of a negotiation.

One of the most important distinctions raised on stage was between commercial optimization and business optimization.

An AI system can be given increasingly sophisticated rules, context, and data. But procurement still needs to determine what outcome the business is actually trying to optimize for.

That's where autonomy becomes as much an operating-model question as a technology question.

Before asking AI to execute a decision, organizations need to be able to articulate the policies, objectives, trade-offs, and boundaries that define a good one.

5. Human Judgment Becomes More Valuable When It's Used Selectively

When the debate personas came off, the conversation became much less about choosing a side.

There was acknowledgement that AI can appropriately take over parts of sourcing and negotiation—particularly when activities are routine, bounded, or don't create significant strategic commitments. Supplier discovery, outreach, RFX administration, and less complex negotiations were among the examples discussed.

But there was also recognition that no system will perfectly capture every variable.

Eventually something changes.

A parameter is missing. The context is different. The relationship matters in a way the model didn't anticipate. Someone needs to recognize that the situation no longer fits neatly inside the rules.

That doesn't make human involvement a failure of automation.

It makes where we use human judgment more intentional.

And that may ultimately be the bigger opportunity with agentic procurement.

Not humans doing everything.

Not AI doing everything.

But machines handling more of what is definable and scalable, while people spend their time on the decisions where judgment actually changes the outcome.

So, Should AI Assist or Act?

We're not going to settle the debate here.

That's what the wigs, robes, procurement experts, and full SOAR26 session are for.

Watch Assist or Act: Industry Experts Live Debate on demand to hear both sides make their case—and see whether they managed to change the audience's minds.

Watch the SOAR26 debate →

One registration gives you access to the complete SOAR26 on-demand session library.