What one supplier invoice really costs, and why the range is so wide

Ardent Partners’ 2025 survey of 204 accounts payable and finance leaders puts the all-inclusive cost of processing one invoice at an average of $9.84, and the time to process it at 8.2 days.¹ Split the same respondents into the 20% with the lowest average cost and the shortest cycle time, which Ardent calls Best-in-Class, and the figures are $2.65 and 2.9 days. The remaining 80% average $12.42 and 13.5 days.¹

That is a gap of about 4.7 times on cost (12.42 divided by 2.65), inside one survey, for the same basic task. For a finance leader who signs off a budget for an AP department, the useful question is what sits in that gap. Some of it is software. A good part is the behaviour of suppliers, the quality of purchase orders and the way approvals are routed, none of which a new tool changes on its own.

What the benchmarks actually say

Two sources are worth reading side by side, with a warning about combining them. Ardent’s State of ePayables 2025 is a web survey of 204 AP and finance leaders run from March to May 2025: 54% work at companies with more than $1 billion in revenue, 26% at mid-market firms and 20% at small ones; 58% are in North America, 31% in EMEA and 11% in APAC.¹ The cost figure is self-reported and defined as all-inclusive. The PDF is a Bottomline-branded edition, which suggests a vendor-sponsored report, so treat the framing with that in mind.

The second source is older. APQC’s database held cost-per-invoice data for 1,485 organisations when CFO.com reported it in February 2018. The top quartile spent $2.07 or less per invoice, the median $5.83 and the bottom quartile $10 or more.² Industry medians in the same data included $1.14 in distribution and transportation and $9.43 in the public sector.² APQC’s own commentary adds that top performers’ cost per invoice is about one-fifth of the bottom quartile’s, and that they handle five times as many invoices per full-time employee.³

The two series cannot be merged. APQC’s 2018 numbers include labour, systems, outsourcing and overhead; Ardent’s 2025 numbers are a self-reported all-inclusive cost from a different respondent pool.²¹ What they share is the shape: a wide spread, with the best performers at roughly a fifth to a quarter of the typical cost.

What drives the gap: exceptions, purchase orders and approvals

Ardent’s second table sets the two groups against each other on the operating metrics behind the cost. Best-in-Class departments report an invoice exception rate of 11.1%, against 20.9% for all others, and process 51.0% of invoices straight-through, against 29.0%.¹ They also have 84.0% of invoices linked to a purchase order, against 47.3%, and spend 12.8% of staff time on supplier inquiries, against 24.0%.¹

Ardent is direct about which of these matters most. Across all respondents the exception rate is 18.4%, and the report states:

Exceptions are typically the biggest single reason why the benchmarks in Table 1 are not lower.

An exception is any invoice that cannot go through on its own: a price that differs from the order, a quantity that differs from the receipt, a missing purchase order, an unreadable field. Each one needs a person to find the cause, contact someone and decide. Ardent does not publish a cost per exception, so the arithmetic of the gap cannot be reconstructed from the report. The direction is clear, though. Respondents ranked approvals that take too long (49%) and a high percentage of exceptions (48%) as their top two AP challenges in 2025.¹

Purchase order linkage explains part of the exception rate, because an invoice with an order and a receipt can be checked by rule, while one without needs a human to decide whether the spend is legitimate. Across all respondents 65.4% of invoices are linked to a purchase order and 35.4% go through straight-through; 57.4% of suppliers invoice electronically.¹ Each of those numbers is a policy and supplier-management outcome as much as a technology one.

Where automation moves the number, and where it stops

There are three places where software acts directly on cost per invoice. The first is capture: reading the invoice, extracting the fields and putting them in the system without retyping. Ardent notes that automated data capture is traditionally one of the first tools an AP department deploys, and calls it a starting point rather than the destination.¹ The second is matching: comparing the invoice with the order and the receipt and passing it when they agree. The third is exception triage: classifying why an invoice failed, routing it to the right person and proposing the fix.

Adoption data shows how much of this is still open. In Ardent’s survey, 44% of AP teams use AI in some capacity, and more than 75% are expected to within 12 months.¹ Invoice validation and reconciliation is fully automated at 22% of respondents and partly automated at 62%. Exception management is fully automated at 4%, partly at 53%, and not at all at 44%.¹ Of the two steps, the one Ardent blames for holding the benchmarks back is the less automated.

The limits are equally concrete. A model that reads an invoice cannot correct a supplier master record that holds two bank accounts for one vendor. It cannot make a supplier send an electronic invoice when 42.6% of them still do not (the 57.4% average leaves that remainder).¹ And it cannot create a purchase order after the fact for a purchase nobody requested. Those are master data, procurement policy and supplier relationship problems. Automation can surface them faster, for example by flagging that the same exception keeps coming from the same few suppliers, but fixing them is operational work.

What the gap is worth: a worked example

The arithmetic is simple, and so is the main trap. A department that has not measured its own cost per invoice cannot say whether it is at $12.42 or $5.83, and the saving from a project is the difference between two numbers it has to produce itself. Ardent’s definition is all-inclusive; APQC’s includes personnel, systems, overhead and third-party vendors.³ Choose one definition, apply it to last quarter and keep applying it.

Caveats and counter-evidence

Three cautions apply to every number above. First, Ardent’s figures are self-reported: 204 leaders answered a survey, and the top 20% were defined by their own reported cost and speed. Second, the Best-in-Class group is defined by outcome, so the comparison shows what low-cost departments look like, not what caused them to be low-cost. A department could have a low exception rate because it handles simpler invoices from fewer suppliers.

Third, the older APQC data comes with its own warning. The CFO.com author told readers to take their findings with a grain of salt.² The numbers are from 2018, and APQC’s cost definition is not identical to Ardent’s, so a 2018 median should not be read as a 2025 target.

There is also a risk in the other direction. Taking cost out of AP by raising the share of invoices that pass without a human touch also raises the share that pass without a human noticing an error. Ardent lists fraud risk as a top challenge for 31% of respondents.¹ Any automation that approves more invoices on its own needs a stated tolerance, a sample check and an audit trail, otherwise cost per invoice falls while the cost of mistakes rises somewhere it is not measured.

What to do next

1. Measure your own cost per invoice for one quarter, using one written definition of what counts as cost, and record the volume it is divided by.
2. Count exceptions by cause for the same quarter: price, quantity, missing purchase order, coding, duplicate, approval delay. The top two causes are your first targets.
3. Find the share of invoices linked to a purchase order and the share of suppliers sending electronic invoices, and compare them with the 65.4% and 57.4% survey averages.¹
4. List the suppliers that generate the most exceptions and the most inquiries, and decide whether the fix is a conversation, a contract term or a system change.
5. Run any capture, matching or triage tool against a baseline and a sample of manually checked invoices, and report the exception rate and error rate next to the cost.
6. Newmind sells a paid accounts payable benchmark report for comparing your figures, and the free pre-audit returns a first estimate of where AI could cut cost in your workflows.