When the Invoice Does Not Match the Order

Ninety invoices arrive and eighty two match the order and the delivery. The other eight consume most of the week.

The quantity is different, the price moved, part of it was delivered separately, the reference number is wrong, or somebody ordered it over the phone and there is no order to match against at all.

The exception costs twenty times the routine

A matching invoice takes seconds. An exception takes an email to the supplier, a conversation with whoever ordered it, a wait, a follow up, and a decision.

So a 10% exception rate is not 10% of the workload. It is most of it. And because each one requires somebody to chase a person who is busy, the elapsed time is days rather than minutes, which is why the pile only ever grows.

The consequences run outward from there. Late payments damage supplier relationships and forfeit early settlement discounts. Month end takes a week because nobody knows what is genuinely owed. And the person handling it is doing detective work rather than finance.

The scale of the wider problem

This is one instance of a burden that has been measured. The Canadian Federation of Independent Business, in the January 2025 edition of its Red Tape Report, found that Canadian businesses spent 768 million hours on regulatory compliance, equivalent to nearly 394,000 full-time jobs, and that the average small business owner lost the equivalent of 32 business days to red tape in 2024. The smallest firms pay over five times more per employee than the largest.

That report is about government regulation rather than supplier invoices, and it says nothing about automation. What it establishes is the shape of the problem: administrative work falls hardest on the businesses least able to absorb it, because a large company hires a department and a company of forty people gives it to somebody who already has a job.

Why the software you bought did not fix it

Accounting systems do exact matching well. Invoice references order, quantities agree, prices agree, post it.

They fail the moment anything is approximately right, which is the entire population of cases that needed help. The supplier invoices in units of ten where you ordered singles. The description is worded differently. The delivery came in two parts and the invoice is for one. A human recognises all of these instantly. A rule cannot express any of them.

So the software handles the invoices that were never a problem, and the exceptions land in a person’s inbox exactly as before.

What AI checks on arrival

This is one of the cleanest cases for AI in a business, because the work is entirely reading and judgement and there is a right answer.

It reads whatever arrives. PDF, scan, photograph, an email body with the figures typed in. Supplier layouts differ and none of them need to be configured in advance.

It matches on meaning rather than string. That “150mm galv bracket, box of 10” and “Bracket, galvanised, 150mm, x10” are the same line is obvious to a person and impossible for exact matching. It is now routine.

It reconciles the arithmetic. Part deliveries against a single invoice, price changes within tolerance, freight and tax lines that do not appear on the order.

It says what is wrong, specifically. Not a flag. “This invoice is for 240 units, the order was 200, and the delivery note records 200. The 40 unit difference is not accounted for.” That sentence is the difference between a five minute resolution and a twenty minute investigation.

It chases. The email to the supplier goes out the same morning, with the discrepancy stated, and the follow up happens without anybody remembering.

What reaches a person is the small set requiring an actual decision, with the work already done.

The control that has to stay

One rule, and it is not negotiable: nothing gets paid on an automated judgement alone above whatever threshold you set.

Payment approval stays with a person. What changes is that the person approves in seconds with the reasoning in front of them, rather than spending twenty minutes assembling it first. Anything the system is not confident about does not get a guess, it gets escalated with the reason attached.

A system permitted to say it is unsure can be trusted with the rest. One that always produces an answer cannot be trusted with any of it.

Fix the ordering, not only the matching

A share of your exceptions are created upstream and no amount of clever matching removes them.

The most common cause is orders placed without a record: somebody rings a supplier, agrees a price on the phone, and the first written trace of the transaction is the invoice. There is nothing to match against, so every one of those becomes an exception by construction.

The second is a purchase order raised loosely, with a description that will not correspond to how the supplier bills. That produces a mismatch on every line and somebody spends twenty minutes proving the two documents describe the same goods.

Both are worth an hour of attention before any technology, because they set the floor on how good the automated result can be. Match what you can, and remove the exceptions that should never have existed.

Where the money actually is

The hours saved are the obvious return and usually not the largest one.

Duplicate payments are more common than most businesses believe, because the same invoice arriving twice in slightly different form is exactly the case exact matching misses. Early settlement discounts get taken when approval takes two days instead of three weeks. Price increases that were never agreed get caught on arrival rather than discovered a year later during a review.

And month end stops being an event, because the position is current rather than reconstructed.

The number to find this week

Ask whoever handles payables what proportion of invoices need a conversation, and how long the average one takes from arrival to resolution.

They will know both figures immediately, because they live it. Multiply them out and you will have the annual cost of the exceptions, which is almost always several times what anybody in the business assumed.

Sources

• Canadian Federation of Independent Business, Canada’s Red Tape Report, 7th edition, January 2025.

AI Optimize reads every invoice on arrival, matches it on meaning rather than exact text, and puts only the genuine decisions in front of a person. That work sits under Document Intake & Validation.