Under 200 invoices a month? Maybe don't automate them yet

The most useful thing I've read about invoice automation this year is a firm telling you not to buy it. Osher Digital is a consultancy that builds these systems for a living, and in its write-up on invoice processing automation it says it isn't worth doing under about 200 invoices a month. The reason it gives is pretty blunt: "The fixed cost of the build is too high relative to the labour saved."

I like advice like that, because anyone selling something has every reason to tell you it fits everyone... so when a firm that builds these things draws a floor under its own market, I assume it's telling you something true, since it would make more money by keeping quiet.

What the numbers say

Osher's figures come from its last six AP rollouts, so they're one firm's experience and not an industry survey -- and honestly that's part of why I trust them more than I'd trust a glossy market report. Model cost came to roughly $0.03 to $0.06 AUD per invoice and the post-audit error rate was 0.3 to 0.6 percent. Payback for mid-market AP teams took 6 to 14 months, faster when the team was over 4 FTE before automation and slower for small finance teams where the cost reduction is partial.

What I notice is how little the model itself costs, and a few cents an invoice is clearly not where the money goes. My reading is that the money goes into the build, so connecting the inbox, mapping fields, handling exceptions and wiring approvals into your accounting system, and most of that doesn't get much smaller just because you have fewer invoices... spread it over 150 a month and it takes a long time to earn back.

The other end of the scale is a good contrast. Bloomberg reported, in a piece republished by Insurance Journal, that five years ago almost no accident and health claims at Ping An ran without humans, and today nearly 60% are automated, with some settled in just 51 seconds. At that kind of volume every second saved adds up to something real, but most small finance teams are nowhere near it -- and that's fine.

You don't need a vendor quote to get a first answer, and I'd actually want you to work this out before you talk to anyone selling a build. Take a normal month, not your quietest one, and write down four numbers: how many invoices came in, the minutes per invoice from arrival to approved and entered (including any chasing, and time a handful honestly instead of guessing), the exception rate, meaning the share that needed something unusual like a missing PO, a wrong amount or a new supplier, and the extra minutes those unusual ones took on top of the normal handling.

Multiply volume by minutes per invoice, add the extra exception time, and that's the hours this work really takes each month. Automation won't remove all of it, because someone still has to review the output and exceptions still need a person -- often more of their attention than before, since the easy ones have been taken away. So I'd be conservative, count the routine minutes as savings and leave the exception minutes out.

Then set those saved hours against whatever a build quote comes back at and see how many months it takes to cover it. If the answer lands well past Osher's 6 to 14 month range, you probably have your answer for now, but if you're comfortably above 200 invoices and several people share the AP work, you're likely in the range where it pays. I've left prices out on purpose, because labour rates and build quotes vary too much for any number from me to mean anything for your team.

If you're under the line

Deciding "not yet" isn't the same as deciding to do nothing, and Osher's own view for small volumes is plain: a part-time bookkeeper and Xero's native bills feature is fine. I'd add three habits that cost very little and make any later automation smaller and cheaper when you do get there.

First, tidy the inbox so there's one address for invoices and nothing else goes to it, because hunting for an invoice is time nobody ever counts. Second, standardise what you capture... pick the fields you always record (supplier, invoice number, date, amount, tax, PO) and record them the same way every time, and that list quietly becomes your spec if you ever automate. Third, keep the source next to the value, so when someone types in an amount the PDF it came from is one click away. That keeps checking quick now, and if you build something later your team is already in the habit of verifying against the page.

There's also a middle step between doing it all by hand and automating it. A simple review app where people check each value against its page is a reasonable place to start -- it's the kind of thing we build at Workbricks -- and the automated steps can come later, once your volume makes the sum work.

I should be honest that this post will date, since model costs keep falling and building these systems keeps getting cheaper. The 200 figure is Osher's, taken from its own rollouts as of this summer, and a year from now the floor may well be lower. So I don't want anyone treating it as a law. Do the sum before you sign anything, then do it again every few months, because the right answer for a team of two in September might look quite different by spring.

Do the sum first.