Accounts payable is one of the clearest places to look for savings because every invoice follows the same path. You can count the steps, time them and put a price on them. Most owners have never done it, so the true cost of paying a bill stays invisible.
What the benchmarks say
A 2026 benchmark roundup from Lido, which cites APQC and Ardent Partners research, reports a manual cost of roughly $10 to $22 per invoice, with APQC's median at $21.40 and its top quartile at $10.18, and Ardent Partners citing a wider range of $15 to $40.[1] I could not check the underlying APQC and Ardent reports directly, so treat these as published ranges relayed by a secondary source.
The same page gives much lower figures for automated processing. Those numbers are not tied to a named independent study. They come from the vendor's own assumptions and customer data, so treat them as marketing estimates until you test them on your own invoices.[1]
The takeaway is not the exact dollar figure. It is that the wide range tells you the cost depends heavily on how your process runs. Two companies can both be "manual" and have very different costs.
Where the cost actually hides
Entering the data is the visible part. The expensive parts are usually elsewhere:
- Receiving: invoices arrive by email, mail, text photo and the vendor portal. Someone gathers them.
- Coding: deciding which account, job or cost code each line belongs to.
- Matching: checking the invoice against the purchase order, contract or delivery ticket.
- Chasing approvals: waiting for the project manager or owner to say yes.
- Exceptions: wrong amounts, missing documents, duplicates and vendor questions. These often take ten times longer than a clean invoice.
- Paying: scheduling, issuing and recording payment.
- Looking things up later: an auditor, vendor or job-cost review asks for the backup.
Which steps AI helps with
| Step | Does AI help? | Comment |
|---|---|---|
| Reading the invoice | Yes, often well | Extraction from PDFs and photos has improved a lot. Still check amounts. |
| Coding to account or job | Partly | Works when your coding patterns are consistent. It learns from history, so bad history produces bad suggestions. |
| Matching to PO or contract | Yes, with rules | Rules do most of the work. AI helps with messy line descriptions. |
| Approval routing | Automation more than AI | Send it to the right person and remind them. Simple and effective. |
| Exceptions | Triage only | AI can flag and explain a mismatch. A person resolves it. |
| Payment | No | Keep a person on release of funds. |
Controls matter more than speed
Faster invoice handling is only good if the controls hold. Before you automate, decide:
- Duplicates: how will you catch the same invoice arriving twice?
- Vendor changes: any change to bank details should trigger a call-back to a known number, never an email reply. Payment-redirect fraud depends on exactly that gap.
- Approval limits: who can approve what amount, and what happens when they are out?
- Segregation of duties: the person who sets up vendors should not also release payments.
- Audit trail: every automated action should be logged with who approved it.
The construction wrinkle
If you run job-cost accounting, invoices carry extra weight. Each one needs the right job and cost code, and you may need lien waivers, retainage tracking and pay-when-paid timing tied to it. Rules and forms vary by state and contract, so work out your requirements with your attorney and accountant. For the automation, the principle is simple: capture the job, cost code and required documents at intake, and flag anything missing before it reaches approval. A PlanGrid and FMI survey found project teams lose hours every week to searching for data and resolving conflicts, which is the same problem showing up in the accounting office.[2] (That survey is from 2018 and covers construction broadly, not invoicing alone.)
Find your own cost per invoice
- Count invoices processed last month.
- List everyone who touched them and estimate their minutes per invoice, including exception handling.
- Multiply minutes by the loaded hourly cost of each person.
- Add software, postage and bank fees for paying.
- Divide the total by the invoice count.
Do this for a typical invoice and for an exception, since exceptions are where the time goes. If your number is under the published manual range, your process is already efficient, and the savings case is weaker. If it is well above, you have found money.
Ask any vendor to process 50 of your real invoices, including your ugly ones, and show you the exception rate, not just the clean-invoice speed.
Sources
- Lido, "Invoice processing cost benchmarks" (secondary source citing APQC and Ardent Partners)
- PlanGrid and FMI, construction survey (Autodesk summary)
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.