Accounts payable invoice workflow automation is sold as removing the payables workload and delivers something narrower and still worth having: it removes touches from the invoices that behave. Capture reads the header so nobody types it. Matching compares three documents without a person. Routing sends the invoice to the right approver and chases them. All of that is real, and none of it touches the exceptions, which are judgements about what to do when reality and the paperwork disagree. This page is about being clear-eyed on the split, so that a business case is built on the part that automates rather than on a demo of the happy path.
What genuinely automates
Reading the invoice header, checking for duplicates, comparing against the order and receipt within a tolerance, routing to an approver and reminding them, and posting the result. These are deterministic and they are most of the volume.
What stays human, and should
Deciding whether to accept a price increase, whether a partial delivery should be paid in full, whether a no-order invoice is legitimate, and whether an unusual supplier is real. Each is a judgement, and an automated rule making it silently is the failure mode worth fearing.
The number to judge any quote against
On the worked example on this site, 400 invoices at three touches of six minutes with a 12% exception rate is 134.4 hours a month, $4,300.80 at a $32 loaded rate and $10.75 an invoice. Automation should be measured against the routine portion of that, not the whole of it, because the 14.4 exception hours will still be there afterwards. A vendor whose case relies on removing all of it is describing somebody else's invoices.
Questions people ask about accounts payable invoice workflow automation
How much of the work can realistically be automated?
The routine portion, which depends entirely on your exception rate. That is why counting exceptions before buying matters more than any published percentage.
Will automation reduce headcount?
Usually it changes what the time is spent on rather than removing a person, at small scale. If the business case depends on a headcount reduction, be sure the remaining exception work is genuinely smaller than one person.
What is the risk of automating too early?
Encoding a route nobody has written down, then maintaining that encoding. The cheapest version of this project is always documentation first.