Fully loaded cost to process one invoice
Top fifth: $2.78
Everyone else: $12.88
- Cost to process one invoice, top fifth of organisations
- $2.78Cost to process one invoice, top fifth of organisations
- Cost to process the same invoice, everyone else
- $12.88Cost to process the same invoice, everyone else
- Of invoices flagged as exceptions outside the top fifth, against 9% inside it
- 22%Of invoices flagged as exceptions outside the top fifth, against 9% inside it
The spread is four and a half times the cost, for the same invoice.
The average is not the interesting number. The distance between the top fifth and everyone else is.
Invoices processed without human intervention
Top fifth: 49.2%
Everyone else: 23.4%
Invoices flagged as exceptions
Top fifth: 9.0%
Everyone else: 22.0%
The three move together, which is why buying one of them fails
Exceptions are the input. An invoice that cannot be matched has to be touched, an invoice that is touched cannot be straight-through, and cost per invoice is mostly the cost of touching. Automation bought without fixing the exception rate produces a faster version of the same rework, which is the shape of most disappointing accounts payable projects.
The sum most finance leaders have never done
Three numbers you already have, multiplied. We have left the answer blank because it is yours, not ours.
Annual invoice volume
Minutes a person spends on each one
Loaded cost per hour
Your number
It usually comes out as a count of full-time roles rather than as a line item, which is exactly why it has never been challenged: it sits in the budget as salary, so it never appears as the cost of not automating. Then add the late-payment side, where discounts are missed because approval took three weeks and suppliers price you accordingly the next time.
Count PO and non-PO separately, or you have an average rather than a case
These two invoices cost different amounts to process, and most organisations have very different volumes of each. Adding them together loses the only distinction that matters.
With a purchase order behind it
An order, a goods receipt and an invoice that agree confirm each other. It posts, nobody sees it, and the marginal cost of the next one is close to nothing. In most organisations this is the majority of the volume, which is why the headline cost per invoice always looks manageable.
With no purchase order behind it
There is nothing to confirm against. It has to be coded and routed by rules, and if the rules do not cover it, by a person who knows which cost centre this supplier usually belongs to. Almost all of the remaining manual work is here, in what is usually the smaller pile.
Which is why a proposal quoting one blended cost per invoice is describing neither case. Ask for the two numbers separately, from us or from anyone else. If a vendor cannot give you the split, they have not looked at your ledger.
The timetable is not yours
Receiving structured electronic invoices is already mandatory for domestic B2B in Germany. Issuing follows in 2027 for larger businesses and in 2028 for everyone else, and the format is EN 16931 in its 2026 edition rather than the 2017 one. Any implementation scoped now should name the edition it targets, because the two are not the same document and a project that does not say which one it means will find out late.
What you get
Four consequences, and each one is the exception rate falling rather than a separate feature. Open any of them for what it means in practice.
The ordinary invoice stops involving a person
An invoice that matches its order and its receipt has nothing in it for a human to decide. It agrees, it posts, and nobody sees it. That is what the top fifth’s 49.2% touchless rate is: not a different technology, a smaller pile of invoices that need looking at.
The work that remains is the work that always needed judgement. A coding question, a price that moved, a receipt that never arrived. Those are the cases an accounts payable team is for, and they are the ones it currently has least time for because the ordinary invoices are in the way.
This is also why the saving is not measured in minutes per invoice. The minutes barely move. What moves is how many invoices anyone has to spend minutes on.
Approval time collapses, which is what unlocks early payment
Early-payment discounts are usually available and usually missed, and they are missed on the calendar rather than on the decision. Two per cent for paying inside ten days is worth having; it is not worth having if the invoice sits for eight days before anyone routes it.
When approval is triggered by arrival and the route is known in advance, the ten days stop being tight. The discount becomes a policy choice about working capital instead of a race the process keeps losing.
It is worth checking what your terms actually offer before treating this as a benefit. Some supplier agreements have no discount in them at all, in which case the gain here is only the cycle time, which is still worth having but is a different argument.
Nothing is lost between systems
The usual estate has an invoice arriving in one place, being posted in another and archived in a third, with a person carrying it between them. Each handover is a place where a document can be dropped, duplicated or filed under a rule nobody chose, and the gaps are only visible when something is asked for.
One path removes the handovers rather than documenting them. The retention rule applies because the system applies it, not because someone remembered, and the document that comes back at audit is the one that was posted rather than a copy of it.
ISO 15489 is the standard this follows. It is worth naming because it is what separates a retention policy from a folder structure that everyone agrees to respect.
A payback you can calculate before you start
Most business cases in this field are built on someone else’s average, which is why they are usually wrong in one direction or the other. The input that matters here is a number you already have: how many of your invoices need a person, and how long each one takes.
Count that for a month and the arithmetic is yours rather than ours. It also tells you whether this is worth doing at all, and on a small enough volume the honest answer is sometimes no.
Ardent’s figures on this page are context for what the spread looks like across a market, not a forecast for your company. We would rather you checked the claim against your own ledger than took the benchmark on trust.
Where this sits
Invoice-to-pay is a cost problem on the surface and an information problem underneath it. That underneath is Information Value Management: what the information behind it is worth, what it costs to hold, and what a change to either returns.
We count what proportion of your invoices could post untouched today, and what stands in the way of the rest. That count is the business case, and it takes days rather than a quarter.
Invoice-to-pay, asked plainly
What is invoice-to-pay automation?
Invoice-to-pay automation captures a supplier invoice, matches it against the order and receipt, routes what needs approval, posts it and archives it under a retention rule, without anyone typing it in. It is the tail end of procure-to-pay, and it is the part with the clearest payback because the manual effort is countable before you start.
What is the difference between PO and non-PO invoices?
A PO invoice has a purchase order and a goods receipt to match against, so a system can confirm it automatically and post it. A non-PO invoice has nothing to match, so it has to be coded and routed by rules, and that is where nearly all of the remaining manual work sits. Any credible business case counts the two separately.
What is the ROI of invoice automation, and when does it pay back?
The calculation is unglamorous and reliable: invoices per year, multiplied by minutes of human handling each, at loaded cost, minus what remains after automation. Payback is usually inside a year for a volume above roughly twenty thousand invoices, and longer below it. Be wary of any figure that does not start from your own exception rate, because that is the only variable that moves it much.
How is invoice-to-pay different from AP automation?
Mostly branding. Accounts payable automation usually describes the same scope. Invoice-to-pay is the more precise label because it names both ends of the process, which matters when a supplier claims to automate it but stops at capture and leaves posting and archiving to you.
Do the e-invoicing mandates change this?
They set the timetable. Receiving structured electronic invoices is already mandatory for domestic B2B in Germany, issuing follows in 2027 for larger businesses and 2028 for the rest, and invoices must conform to EN 16931. That converts invoice automation from a project competing for budget into one with a date attached.