AI Moved the Centre of Gravity. Procurement Is Standing Where It Used to Be.

A point from a recent ERP-industry video has stuck with me. The argument: the ERP has always been the centre of gravity in the enterprise tech stack, and AI is shifting that centre. The result is confusion about architecture, integration and data. Everything you need to make AI work is being exposed as a weakness by AI itself.

I agree with all of it. I would go one step further. Nowhere is this more visible than in procurement, because procurement lives at the exact point where the old centre and the new one pull against each other.

What the old centre of gravity looked like

For most of us, the procurement stack was built outward from the ERP. The vendor master lived there. The purchase order lived there. The GL coding, the approval matrix, the three-way match. The source-to-pay suite sat on top, the sourcing tool sat on top of that, and contract management was bolted on somewhere to the side, usually as a repository nobody fully trusted.

That arrangement had one great virtue. Everyone knew where the truth was. If the PO said it, it happened. If it wasn't in the ERP, it didn't exist.

Where the gravity is moving

AI does not care where the transaction is recorded. It cares where the context is.

And in procurement, the context never lived in the ERP. It lived in:

• Contracts, mostly PDFs, in shared drives, with the key terms read once at signature and never again.

• Supplier conversations, in email threads and meeting notes.

• Category knowledge, in the heads of the two people who have run the category for six years.

• Risk and onboarding evidence, in questionnaires, certificates and due diligence packs spread across four systems.

The moment you ask an AI tool a real procurement question ("which of our suppliers can raise prices above CPI next year?", "where are we exposed if this vendor fails?"), it has to reach into all of that. The PO table is almost irrelevant to the answer.

So the centre of gravity moves from the system of record to the layer that holds the unstructured knowledge and the model that reasons over it. That is the shift.

What AI exposes

This is the uncomfortable part, and it is where the confusion comes from. AI does not create the weaknesses. It removes the places they used to hide.

• The vendor master. Duplicates, dormant records, three spellings of the same supplier. A human learned to live with it. A model cannot tell they are the same company.

• Spend classification. Half the spend sits in "Other" or "Miscellaneous services". Your AI spend analysis is only as good as that taxonomy, and most taxonomies were never maintained.

• Contracts. If you cannot say where the signed version of a contract is, no retrieval system will find it for you.

• Approval logic. Much of it lives in policy documents and informal habit, not in the workflow. An agent cannot follow a rule nobody wrote down.

• Ownership. Who owns supplier data: procurement, finance, IT, risk? In most organisations the honest answer is "nobody, jointly".

None of this is new. Every procurement leader has known about it for years. The difference is that it used to be a data quality project that got deferred every budget cycle. Now it is the thing blocking the AI business case the board is asking about.

Procurement has a second problem: it is buying the chaos

Procurement is not only living on the fault line. It is also the function signing the contracts that make it worse. Every business unit now wants its own AI tool. Intake orchestration layers on top of the S2P suite, on top of the ERP. A contract AI point solution. A supplier risk AI. A spend analytics AI. The ERP vendor's own AI add-on. Each comes with its own copy of your data, its own model, its own view of the truth.

That is how you end up with five AI answers to the same question and no way to say which one is right. The centre of gravity has not moved to one place. It has fragmented into five.

Where I think the centre should land

My position will not surprise anyone who has read my earlier pieces. The new centre of gravity should be a layer you own: your data, your retrieval, your model, running in your environment. Not because on-premise is fashionable. Because it answers the three questions the confusion is really about:

1. Architecture. One knowledge layer that every tool reads from, instead of every tool building its own copy.

2. Integration. The ERP stays the system of record for transactions. The owned AI layer becomes the system of context. Clear boundary, clear contract between them.

3. Data. Contracts, supplier records and category knowledge never leave your walls, which in a regulated environment removes a whole class of argument before it starts.

What to do in the next 12 months

• Fix the vendor master first. It is the single highest-leverage cleanup for any AI use case in procurement.

• Put contracts in one place, with the metadata extracted. Parties, term, renewal, price mechanism, liability cap. This is the foundation of almost every useful procurement AI question.

• Refresh the spend taxonomy. Kill "Other". If more than a small slice of spend is unclassified, your analytics AI is guessing.

• Write the approval rules down. If a new hire could not follow them from the document alone, neither can an agent.

• Audit your AI contracts as a portfolio, not one by one. Map which tools hold copies of which data. Overlap is where the confusion and the cost both live.

• Decide who owns the context layer. If you do not, a vendor will decide for you.

The ERP is not going away. It will keep recording what happened. But the question of where the enterprise thinks is being answered right now, mostly by accident. Procurement has more at stake in that answer than any other function and more influence over it than it realises, because it signs the contracts.