AP automation: what to automate in accounts payable, and what to leave alone

AP automation is the practice of removing manual handling from the path a supplier invoice travels between arriving and being paid. Vendors sell it as a single product, which is why so many evaluations go badly: the phrase covers four distinct jobs — capture, matching, approval, payment — and they do not repay automation equally. A team that automates the two cheap ones first usually finds that most of the pain was never in the software.

This article is about the automation decision. The underlying process — what a payables ledger is, why invoices get paid twice, how to close the month — is accounts payable, and it is worth having that straight before you buy anything, because automating a process nobody has agreed on simply produces disagreements faster.

The four jobs hiding inside AP automation

• Capture: getting the invoice off a PDF or an email and into structured fields — supplier, number, date, net, tax, total. This is the job OCR was built for and the one people mean when they say automation.
• Matching: checking the invoice against what was ordered and what was received. Two-way matching compares invoice to purchase order; three-way adds the goods receipt.
• Approval: routing the invoice to whoever is allowed to say yes, with a limit attached to the amount, and keeping the record of who said it.
• Payment: producing the payment run, exporting it to the bank, and marking the ledger once the money leaves.

How much each one repays

1. Approval first. It costs nothing but a decision: write down who approves what up to which amount, and what happens when they are on holiday. Most late-payment problems and nearly all maverick spend trace back to this being informal rather than to a lack of software.
2. Payment second. Batch the runs — weekly is enough for almost everyone — and pay from a list rather than one invoice at a time. Paying continuously is what creates duplicate payments, because the same invoice arrives twice by two routes and nobody is looking at a list.
3. Capture third. This does need software, and it is worth it above roughly a few hundred invoices a month. Below that, the OCR correction time is comparable to typing, and the failure mode is worse: a wrong number that looks confident.
4. Matching last, and only if you actually raise purchase orders. Three-way matching against orders that half the company does not create is a control on paper and a queue of exceptions in practice.

The number that decides everything is invoices per month per person handling them. Under about two hundred, the constraint is almost never keying speed — it is waiting for approvals. Automating capture in that situation buys you a faster route to the same bottleneck, which is the most common way an AP automation project disappoints everyone who signed off on it.

What to measure before and after

• Cycle time from invoice received to approved, split from approved to paid. If the first is long and the second short, your problem is human, not technical.
• Percentage of invoices touched more than twice. Rework is the real cost, not the initial handling.
• Duplicate payments caught and missed, counted per quarter rather than argued about.
• Early-payment discounts taken versus available. This is the one benefit that shows up in cash rather than in a slide.
• Exceptions as a share of volume. If more than about one in ten invoices needs a human to intervene, the rules are wrong, not the invoices.

Where Ettex Invoices fits, and where it does not

Ettex Invoices covers the sending side properly — issuing invoices, tracking what is unpaid, chasing it — and Ettex Books holds the ledger those payments land in. On the incoming side you can record supplier bills, route them for approval with an amount limit, and keep the approval history attached to the document.

It is not an AP automation platform, and it is worth being blunt about the gaps. There is no OCR capture: supplier invoices are entered or imported, not read off a scan. There is no three-way matching engine, because there are no goods receipts to match against. There are no ERP connectors and no direct bank payment execution — a payment run is exported, not fired. If your evaluation turns on those four things, you are shopping for a dedicated AP product, and you should.

The honest version of the business case

The savings claimed for AP automation are usually calculated as invoices multiplied by minutes saved multiplied by a loaded hourly rate. That number is real only if the time released is actually removed from the payroll or redirected to something you would otherwise have hired for. In a five-person finance team it typically is not, and the genuine return is elsewhere: fewer duplicate payments, fewer late fees, discounts captured, and an audit trail that does not need reconstructing from a mailbox at year end. Those are smaller numbers and much more likely to appear.