Three-way matching: every invoice, every line, every time
Every supplier invoice should agree with two things: what you ordered, and what was actually delivered. That is the three-way match, and it is the foundation of purchasing control.
In theory. In practice most finance teams glance at the total, nod, and pay. Not out of carelessness — there simply isn't time. Checking a twelve-line invoice against two other documents, line by line, takes around six minutes. At two hundred invoices a month that is twenty hours.
So teams spot-check, or don't check at all. And that is where overcharges quietly get paid.
What the system checks
Automated matching doesn't look at the total. It looks at every line, comparing three sources: the purchase order, the delivery note, and the invoice.
For each line it asks three questions. Were we invoiced for what was delivered? At the price we agreed? Is this item on the purchase order at all?
The typical discrepancies fall out of those three questions:
- Quantity mismatch — 100 boxes delivered, 120 invoiced
- Price mismatch — €4.20 on the order, €4.50 on the invoice
- Item not ordered — a line that appears on no purchase order
In one real example, nine of twelve lines matched cleanly and three were flagged. Those three came to €368.50 — on a single, entirely ordinary office supplies invoice.
The team works the exceptions, not the invoices
This is the shift. Your finance person isn't reviewing twelve lines, they're reviewing three — the three that actually warrant attention. The other nine are ready to approve.
The system recommends: hold the three flagged lines, release the remaining nine for payment. A query goes to the supplier for the flagged items automatically, with the reason attached.
A person still approves every payment. The system checks and recommends; the decision stays with the team. That isn't caution for its own sake — a supplier relationship is a commercial matter, and the consequences of withholding a line need human weighing.
The numbers
- Checking one invoice: from around six minutes to seconds
- Lines matched automatically: roughly three quarters, with no human involvement
- Supplier turnaround: clean invoices move straight through rather than waiting for a batch review
That last point is about supplier relationships. When clean invoices don't get stuck behind problematic ones, suppliers are paid within days. That is a negotiating position as much as an efficiency gain.
What happens to the flagged lines
Finding a discrepancy isn't worth much on its own. The question is what happens next.
For every flagged line the system assembles the basis of the query: what each document says, the size of the difference, and the recommended action. The message to the supplier goes out from there in a couple of clicks — not as a vague "please check", but with a specific reference to the purchase order number and the line in question.
This matters because composing the query often costs more time than finding the error. If disputing a forty-euro difference means twenty minutes of correspondence, most teams simply pay it. At which point finding it achieved nothing.
Part deliveries, tolerances, reality
In practice things rarely agree exactly, and they don't need to.
Part delivery is normal: the order was for a hundred, sixty arrived, sixty were invoiced. That isn't a discrepancy, it's an open order. The system has to treat it as one, or every part delivery produces a false alarm — and after a few weeks the team starts ignoring the flags altogether.
The same goes for tolerances. A few cents of rounding, a small variance on material priced by weight, apportioned delivery charges — these can carry a tolerance, set per company and per supplier. The goal isn't for every difference to stop the process. It's for the ones that matter to stop it.
Where it fits
The system connects to your existing accounting or ERP platform and runs in your own EU cloud environment. Nothing has to be replaced to adopt it.
Purchasing data — who you buy from, at what price, on what terms — is sensitive commercial information. It is not used to train a third party's general models.
What it requires
One thing is worth stating up front, because most demonstrations leave it out: three-way matching only works where purchase orders exist.
At many mid-sized companies, buying happens by phone or email with no formal order. In that case the system has nothing to compare against — matching the invoice to the delivery note is a two-way match, which is still useful, but it will never catch a price discrepancy, because nothing recorded what was agreed.
That isn't a technical obstacle, it's a process question. If the goal is purchasing control, purchase-order discipline is the first step and automation is the second. It doesn't work in the other order, and it would be dishonest to claim otherwise.
What it's worth
€368.50 on one invoice. The real question is how many of those slip through unnoticed across two hundred invoices a month, when nobody has twenty hours to check them all.
The usual answer is that nobody knows. That is precisely the problem.