Best practices in accounts payable automation are easier to describe as traps, because the failures in these projects rarely come from doing something obviously wrong. They come from doing something that looked like good practice: modelling the process thoroughly, automating the biggest pain first, setting tolerances tightly for control, and measuring the result by the number of automated invoices. Each is defensible and each has a characteristic failure. This page is the four traps, what each one looks like while you are walking into it, and the version of the same instinct that works instead.
Trap one: modelling the process too thoroughly
Thorough modelling produces a configuration that describes every case including the ones that happen twice a year, and which then has to be maintained. The version that works is modelling the routine and leaving exceptions to people, deliberately.
Trap two: automating the loudest pain first
The loudest pain is usually approval delay, which depends on people. Automating it first delivers faster delivery of unfinished work. The version that works is the sequence: document, capture, match, route.
Traps three and four: tight tolerances, and measuring the wrong thing
Tolerances set tight for control produce exception queues that get bypassed, which is less control than before. And measuring automation by percentage of invoices automated rewards excluding difficult ones. Measure the hours and the cost instead. 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, $10.75 an invoice and $51,609.60 a year.
Questions people ask about best practices accounts payable automation
How do we know if we fell into one?
The signals are a configuration nobody wants to touch, an exception queue that is never empty, and a metric that improves while the team is no less busy.
Is it recoverable?
Usually, by simplifying: widen tolerances to what you actually approve, remove rules covering rare cases, and re-measure in hours.
What is the single best practice?
Write the route down before automating it. Everything else follows more easily from that one step.