Your AP automation ceiling is the share of invoices that can clear without a person, and it is set by four inputs you already own. Those are vendor master quality, the ratio of purchase-order-backed to non-PO-backed invoices, supplier electronic capability, and the structural portion of your exceptions. Software raises the ceiling only after those inputs improve.
Key takeaways
- Straight-through processing requires every condition to resolve at once, so the ceiling is the intersection of four inputs rather than their sum.
- Ardent Partners reported an average invoice cost of $9.84, a cycle time of 8.2 days, and an 18.4 exception rate for 2025, measured across operations that already run capture and workflow software.
- Ardent publishes its top tier only in relatives, because absolute performance is not comparable across different invoice mixes. A vendor may quote one touchless percentage to every prospect. That is a claim the research firm declined to make.
- Run the six-step diagnostic before the demonstration, and take your measured ceiling into the vendor conversation as a constraint.
Every accounts payable (AP) automation business case states a target touchless rate, and that number usually comes from a vendor deck. Most programs end up well underneath it and stay there. The cause is in your own data rather than in the product you bought.
The number in the business case, and the number you get
Ardent Partners published its State of ePayables 2025 benchmarks in January 2026, reporting 2025 data. Average invoice processing cost was $9.84. Average cycle time was 8.2 days. Average exception rate was 18.4 percent. Those are averages across operations that already own capture software and approval workflow, which is what makes them useful. They describe the state after automation rather than before it.
Gartner's November 2025 survey covered 183 CFOs and senior finance leaders. It found 59% using AI in finance. Among adopters, accounts payable automation is the second most common use case at 37 percent. In the same survey, 91% of adopters reported low or moderate impact initially. That is the ceiling expressed as a survey result. Plenty of teams have bought the software, and far fewer have the outcome it was bought for.
The pattern inside a program repeats. The first ninety days go well, because the cleanest population automates first, meaning repeat suppliers, purchase order (PO) backed volume, structured formats, and familiar coding. The touchless rate climbs fast, the steering committee is happy, and then it flattens. That flattening is the point where the system runs out of invoices meeting all its conditions at once. Our guide to accounts payable automation shows where each condition gets tested.
Your AP automation ceiling is a property of your inputs
Straight-through processing (STP) requires the system to answer every question an invoice raises without asking a person. Who is this supplier, and against which record? What was bought, and against what commitment? Which account and cost center apply? Who approves, and did anything already pay it?
An invoice clears untouched only when all of those resolve. One failure and the system routes it to a queue. Your ceiling is therefore the intersection of the conditions rather than the sum, which is why individually respectable numbers multiply into a disappointing one.
That also explains the demonstration. A vendor demonstrates on the population where every condition holds, because that population exists in every company. Your production average gets dragged down by everything else, and nobody demonstrates the everything else.
Four inputs determine how big that intersection is. You own all four before signing a contract, and three of them take quarters rather than weeks to improve.
Vendor master quality
Every match, duplicate check, and payment instruction resolves against the vendor master. An automation project leaves several problems untouched, namely duplicate records, legal name against trading name, missing tax identification numbers, stale remit-to addresses, and unverified bank details. Extraction reads a supplier name perfectly and still attaches the invoice to the wrong record.
The downstream cost is measurable. In 2022 research, the Washington State Auditor put duplicate payments at 0.8% to 2% of total payments. A fragmented vendor file is how most get created. Where more than a small share of your active records are near-duplicates, no configuration change fixes your duplicate rate.
The PO-backed to non-PO-backed ratio
Three-way matching works because three independent records exist. Take away the goods receipt and the PO, as services spend does, and the deterministic control disappears with them. Non-PO-backed invoices need a general ledger (GL) coding decision and a budget owner who is willing to approve them.
Where half your volume has no PO behind it, half your volume cannot use your strongest control, whatever software you install. Measure that ratio first, because it caps the ceiling harder than any other input.
Supplier electronic capability
Ardent reports that 57% of suppliers can send invoices electronically on average, and its top-performer tier has 1.4 times more suppliers enabled. Notice what that is. Supplier enablement is a program run by people who call suppliers, negotiate formats, and chase onboarding. It takes quarters, and no purchase shortens it.
Exception rate and its composition
The 18.4% average matters less than what is inside it. Sort your exceptions into two piles. Structural exceptions exist because a record cannot exist, as with services that generate no receipt or contract-priced work billed after the fact. Better extraction never removes them. Correctable exceptions exist because something upstream is wrong, such as a missing PO number, a changed remit-to, an unwritten coding rule, or a supplier who never got enabled. Those do go away.
Take a team whose exceptions are 70% structural. Its ceiling is far lower than that of a team at the same 18.4% whose exceptions are mostly correctable. The headline rate does not show that.
Why the top tier gets published only as a relative
Ardent describes its highest performers entirely in relatives. Invoice cost runs 79% lower than the average. Cycle time runs 79% faster. The exception rate runs 47% lower. The tier enables 1.4 times as many suppliers, and it processes more than 1.8 times as many invoices straight through. There is no absolute cost, no absolute cycle time, and no straight-through percentage anywhere in it.
A research firm holding the underlying data made that choice deliberately. Absolute performance is not comparable across companies with different invoice mixes, supplier bases, and enterprise resource planning (ERP) environments, so an absolute would mislead more than it informed. Hold that next to a vendor quoting the same touchless percentage to every prospect in every industry. That vendor is making a claim the research firm with the data declined to make.
An AP automation diagnostic to run before you buy anything
Give this to one analyst for a week.
- Pull twelve months of posted invoices. Calculate the share with a PO reference. That is your PO ratio.
- Count active vendor records, then count those sharing a normalized name, tax identification number, or bank account. That is your duplicate exposure.
- Split intake by channel. Count structured arrivals, meaning electronic data interchange, e-invoicing networks, and portals, against unstructured ones, meaning email attachments and paper. Compare the structured share to the 57% average.
- Export one quarter of exceptions and classify each by cause, then tag each cause structural or correctable. Rank by volume, then rank again by hours consumed.
- Time supplier inquiry work for two weeks. Ardent puts the average at 21.9% of AP staff time, so anything near that is a second workstream rather than a rounding error.
- Multiply the PO-backed share by the electronically capable share. The product is a crude floor for the population where matching is deterministic and the document arrives structured.
That last figure is deliberately rough, and still more honest than a vendor's number. Where it lands near 30%, a 70% touchless target in year one is a forecast without a mechanism. Either the target comes down or the inputs improve. Our guide to the accounts payable process maps each stage to its owner. That mapping tells you who fixes which input.
Where current approaches fall short
Pilots run on a curated invoice set, the same population that automates first anyway, so the result predicts the first ninety days and nothing after. Business cases state one touchless number with no invoice mix, no supplier enablement rate, and no year attached to the benchmark behind it.
The bigger gap is what happens to the invoices that fail. Most tooling routes an exception to a person and calls that handling. An approver receives a price variance and has no authority to settle it, so the invoice makes a second lap, and the supplier emails to ask why. That is how 21.9% of staff time ends up on inquiries while the workflow itself looks faster.
How we approach the AP automation ceiling
We publish automation rates of 80% or more, and we qualify them, because the honest version is a range that depends on the four inputs above. Qualcomm moved from 14 percent to 61% autonomous invoice processing on SAP S/4HANA. TruGreen reached 60 percent autonomous processing and identified $870,000 in duplicates. The numbers differ, because the operations differ.
Our AI is built for the population that sets the ceiling. It predicts GL coding for non-PO-backed invoices from vendor and cost center history, and resolves suppliers against a messy vendor master rather than assuming a clean one. It also works the supplier mailbox through AP Inbox Service Center, so inquiries stop consuming analyst hours. Every decision leaves an audit trail, which is what lets you tune a tolerance instead of guessing. The maturity stages appear in our overview of AI maturity in accounts payable, and the capabilities are listed on our AP automation page.
The bottom line
Run the diagnostic before the demonstration, and take your ceiling into the vendor conversation as a constraint rather than a question. A target can exceed your measured inputs. That is a forecast, and forecasts are what make year two disappointing. To read your own numbers against the 2025 benchmarks, talk to us.
Frequently asked questions
What is a realistic touchless invoice rate?
It depends on inputs you can measure, chiefly your PO-backed share, supplier electronic capability, vendor master quality, and the structural portion of your exceptions. Ardent Partners publishes its top-performer tier only in relatives, processing more than 1.8 times as many invoices straight through, never as an absolute percentage. Treat any vendor quoting one universal number with caution.
Why did our AP automation return stall after the first year?
The cleanest population automates first, so the early curve is steep and unrepresentative. What remains is non-PO-backed volume, suppliers who cannot send structured data, and structural exceptions that better extraction does not remove. Gartner found in November 2025 that 91% of finance AI adopters report low or moderate impact initially, which is the same pattern at survey scale.
Does better invoice capture raise the ceiling?
It raises the ceiling only for the first stage. Extraction accuracy governs how well a document is read. Everything after that depends on records you maintain yourself, including supplier resolution, matching, coding, approval, and duplicate detection. Two things raise the ceiling further than any change to the reading engine, cleaning the vendor master and raising supplier enablement.
Which of the four inputs improves fastest?
Vendor master quality improves fastest, because it is a data cleanup you control end to end. The PO ratio depends on procurement policy, supplier enablement depends on suppliers, and the structural share of exceptions depends on what your business actually buys.