Three of the four things your buyers do all day have no benchmark, no report line, and no owner. That is where the hours are.
Every finance team can tell you what an invoice costs them to process. Ardent Partners put the 2025 average at 9.40 dollars, against 2.78 for best-in-class teams, drawn from a survey of 212 AP and finance professionals. APQC runs its own version. There are calculators. There are conference talks.
Now ask the same company what it costs them to handle an order confirmation that comes back with a changed delivery date. Nobody has that number. Nobody has ever been asked for it.
That asymmetry is the actual problem in procurement operations, and it is not obvious from inside the department.
Four loops, one of them measured
- Requisition intake
- Order confirmations
- Expediting
- Invoice exceptions
Only the last one sits inside accounts payable, which is the only one of the four with an industry benchmark attached. So it is the one that gets a project, a business case, and a vendor. The three upstream loops get absorbed.
Absorbed work has a specific signature. It never appears in a report, it has no budget line, and it grows with volume without anyone noticing the slope. A buyer who spends ninety minutes a day comparing confirmations to purchase orders is not doing something the company can see. If she leaves, the ninety minutes leave with her, and the replacement takes six months to build the same instincts.
Ask a procurement lead how many order confirmations arrived last month with a changed date, price or quantity. In most companies the honest answer is that the information does not exist anywhere. The comparisons happened. The results were acted on. Nothing recorded them.
The exception rate is the whole cost
The published AP numbers are worth reading closely, because the headline figure hides the mechanism.
Ardent's 2025 research puts the industry average invoice exception rate at 22%, against 9% for best-in-class teams. Cycle time follows the same split: 3.1 days for best-in-class, 17.4 days for the least automated organisations, with an overall average of 9.2 days.
Read those two pairs together and the cost per invoice makes sense. The clean invoices are nearly free either way. The gap between 2.78 dollars and 9.40 dollars is not spread evenly across every document. It is concentrated in the fifth of them that needed a person, and the difference between a good AP operation and a poor one is almost entirely how many documents fall into that fifth.
There is a second thing hiding in the same number. Analysts who work with these benchmarks estimate that most organisations undercount their true cost per invoice by 30 to 50%, because the calculation usually covers labour and software while leaving out rework, duplicate payments, late-payment penalties and missed early-payment discounts. The number your CFO believes is probably the optimistic one.
Touch time is not cycle time
Here is the part that surprises people who have never timed the work.
A manual invoice takes something like 10 to 15 minutes of actual human attention to key in and route, dropping to under two minutes once automated. But the average invoice takes 9.2 days to get from arrival to approved.
Nine days of elapsed time. Fifteen minutes of work.
The other 8.99 days are queueing. The document sits in an inbox until someone opens it. The question goes to a requester who answers tomorrow. The approver is travelling. Nothing is being done to the invoice for almost all of the time the invoice exists as an open item.
This matters because it changes what automation is for. If you think the problem is keystrokes, you buy something that types faster and you save fifteen minutes. If you understand the problem is waiting, you buy something that starts the clock on the day the document arrives rather than on the day a person gets to it, and you save eight days.
The same logic runs through the three unmeasured loops. The order confirmation that arrives Monday and gets compared on Thursday was always going to need the same ten minutes of comparison. What it did not need was the three days.
Count it for one week
The way out of this is not a benchmark, because there isn't one. It is a tally sheet.
Pick one buyer and one week. Every time an order confirmation arrives, log four things.
- When it arrived
- When someone compared it against the order
- Whether anything differed
- Which supplier sent it
Do the same for requisitions that went back for a missing field.
At the end of the week you will have three numbers nobody in your company has ever had. What share of confirmations disagree with the order. How many days pass between arrival and comparison. And which suppliers account for most of both.
The third number is usually the surprise. In most operations a small group of senders produces the majority of the rework, and everyone can name them in the corridor but nobody can prove it in a meeting.
Those numbers also give you a real business case rather than an imported one. Not what an analyst says an invoice costs an American company, but what this specific work costs you, measured in your own hours, on your own supplier base.
One week. One spreadsheet. It is the cheapest diagnostic in the department.
Where we fit
An order confirmation lands in a shared mailbox at 07:40. Orchestra reads it, checks it against the purchase order, and flags the two lines where the date moved. Sonata posts the rest into your ERP. The buyer opens her queue at nine and sees two questions instead of forty documents.
Nothing in that sequence is faster typing. It is the removal of the waiting.
Your suppliers keep sending what they send. Your buyers stop reconstructing it by hand, and for the first time you have a record of which suppliers make them do it most.