For heads of procurement · Hamburg and northern Germany
In procurement, whoever has the numbers first decides.
Artificial intelligence reaches you along two routes: as a tool that takes over document-heavy checking, and as a cost position that arises in IT and lands on your cost centre. Both belong in the same pair of hands — yours. This is how to put it in order, without anyone selling you a licence.
What is in it for you
Before finance asks
You know the number before it is requested
Consumption-based AI cost is currently growing unsupervised in many companies. Whoever can name three figures in the autumn is in a different position from someone who needs a week to understand the bill.
- Spend, transactions, cost per transaction
- Limits set before the first trial runs
- Chargeback to the department causing it
At the table
A documented consumption profile, not an estimate
Volume tiers, minimum commitments, expiring credits, price-change notice: all of it negotiates differently once you can evidence three months of real demand.
- No commitment in the first quarter
- Four clauses to check before bundling
- Keep the ability to switch
Along your own chain
Meeting discount windows instead of explaining them
The discount period runs from receipt of the invoice, not from approval. Every day in the queue comes off it, and the loss is never posted. Two per cent at fourteen instead of thirty days is roughly forty-six per cent a year.
- Throughput time per stage, measured not estimated
- Blanket orders made manageable
- Automatic approval where the three states agree
The one that is urgent
Cost control
The bill nobody ordered
A licence has a ceiling. Consumption-based AI does not. It arises per transaction, without a signature, through a cloud agreement or a credit card held by IT — and nobody sees it until the month closes. That is not an IT problem. It is a procurement problem parked in IT.
More decision material
Ordered the way it gets invoiced
An invoice cannot be approved automatically because it does not match the order — and it does not match because the order was never placed the way the supplier later bills. The break is at the beginning of the chain, not at the end of it.
Process analysis9 min 03
What AI actually does in procurement
Reading documents, comparing them against a reference, and flagging what looks wrong. Everything else you have been sold is either an interface on top of that, or a promise nobody has kept yet.
Fundamentals6 min 04
Article 4 quietly landed on your desk
Since February 2025 the EU AI Act has obliged every company deploying an AI system to take appropriate measures on AI literacy. Since July 2026 no particular level of competence has to be guaranteed any more — the duty to take measures remains. If your buyers use Copilot, that includes you.
Regulation5 min 05
Five questions before you buy the tool
A procurement department that cannot answer these five questions about its own process will not get a good outcome from any vendor, however capable the software.
Buying advice6 min
Where I am usually called in
Full description →Situation one
Checking costs more than the errors
Invoices arrive on paper or as PDFs, are verified by hand against contracts and measurements, and corrected in most cases. Nobody can say what the checking itself costs.
- Documented process baseline
- Effort measured per step, not estimated
- A ranked list you can put in front of a board
Situation two
A tool was bought, nothing changed
The software works as advertised. The process around it did not move, so the same people do the same work, now with an extra system in the way.
- Honest assessment of what the tool can carry
- The process redesign that was skipped
- Adoption work with the people affected
Situation three
Your buyers already use AI
Copilot and ChatGPT are in daily use, unsupervised, often on supplier data. Nobody has written down what is allowed, and Article 4 of the AI Act has been in force since 2025.
- Training that changes how people work
- Written record of the qualification
- Ground rules agreed with legal and works council
The position
Process first. Technology is the easy part.
I did not arrive at artificial intelligence through artificial intelligence. I arrived through a decade of sitting next to people while they worked, writing down what actually happens, and finding that the documented process and the real one rarely match.
That order matters. A model that reads invoices flawlessly will not help an organisation that cannot say which approval step everything waits on. Conversely, once a process is genuinely understood, the technical part is usually smaller than anyone expected — and occasionally unnecessary.
Talk it through
Thirty minutes on your actual process.
Describe where the time goes and I will tell you what I would tackle first — including the cases where the honest answer is that software is not your problem. No charge, no obligation.