A financial director had fourteen days to recommend a catering partner for a major sports venue contract. Four international suppliers. More than 1,200 pages of proposals. The kind of deadline where you read faster rather than better.

One supplier had proposed 40% fewer staff than every competitor. The reason was written into the document in plain language: the client would provide the sales staff. Not flagged. Not highlighted. Just sitting in the assumptions, in a stack of material that one person didn't have time to read properly. Annual cost if nobody caught it: £200,000.

This is the sort of thing AI is supposed to find. So why didn't it?

According to The Hackett Group's 2025 Key Issues Study, 49% of procurement teams ran generative AI pilots in 2024. Only 4% reached meaningful deployment. The usual explanations (data quality, integration, change management) are real but insufficient. They don't explain why pilots keep stalling at the same point: the moment the tool encounters knowledge that was never written down.

The OECD put it well in their "Governing with Artificial Intelligence" report. Procurement managers aren't sceptical of AI because they don't understand it. They're sceptical because they understand their own work. Negotiation judgment. Category knowledge. The instinct for when a staffing ratio feels light or a growth projection feels optimistic. That expertise is tacit, refined over years, never formalised, because it never needed to be.

If you're a procurement professional, this should be reassuring. Your expertise is the thing AI can't replicate. You were right to be wary.

But here's the harder question. If that intelligence stays in your head, undocumented, unstructured, untransferable, it isn't protected. It's invisible. Not just to AI, but to your own organisation. The knowledge you think of as your competitive advantage is actually your ceiling, because it can't be scaled, shared, or amplified by anything.

In the sports venue case, the analysis worked because the context came first. Before anyone touched a proposal, we built the reference layer: industry benchmarks for staffing ratios, post-pandemic growth rates for comparable venues, standard food cost percentages, realistic revenue projections. We call this mapping the contours of what an organisation actually knows, and it's the foundation of our methodology.

None of that context was in the proposals. All of it determined whether the proposals made sense.

With that foundation in place, three things surfaced immediately. The £200,000 staffing gap, quantifiable the moment you knew the industry benchmark. Growth projections of 12-15% from two suppliers, in a sector where the post-pandemic norm is 5-7%. And food cost percentages ranging from 28% to 35%, calculated on entirely different bases, not comparable until standardised against a common framework.

The financial director had a board-ready recommendation, with full audit trail, within the deadline.

The gap in procurement AI isn't between human judgment and machine capability. It's between organisations that have made their best people's knowledge explicit, and organisations that haven't. The former can use AI to genuinely dangerous effect. The latter will keep running pilots that stall the moment the tool needs to understand what it's looking at.

Your expertise isn't what AI is replacing. It's what AI requires. The question is whether you're going to leave it locked in your head.