There is a folder on a shared drive at a 140-room hotel called AP – TO CODE.
It has four hundred things in it. Some are PDFs. Some are photos of paper invoices taken at an angle. Some are emails that were dragged in whole, with the invoice buried three replies down. A few are duplicates of ones already in the system, which nobody will discover until the vendor calls about a double payment.
Every Thursday, the controller opens that folder. Every Thursday, she closes it around 3 p.m. having made real progress, and every Thursday it refills.
Nobody in the building wants to touch that folder. That's the tell. In our experience, the workflow that everyone avoids is almost always the most expensive one in the company, precisely because avoidance is what let it get that bad.
What it actually costs
Let's build the number, because "it's a mess" is not a number.
A property that size in a full-service configuration is processing somewhere in the range of 350 to 500 vendor invoices a month — linens, produce, beverage, engineering parts, landscaping, elevator service, the pest company, the guy who fixes the ice machines. Say 400.
The manual path for each one: open it, identify the vendor, match it to a PO or a receiving document if one exists, code it to a GL account and a department, key it into the accounting system, route it for approval, and file it. Even at a brisk five minutes each — and it is not five minutes for the ugly ones — that's 33 hours a month. Call the loaded rate $45. That's about $18,000 a year in pure keying.
Then add the things that aren't keying. The approvals that sit in a GM's inbox for eleven days, so the early-pay discount expires. The duplicate that got paid. The invoice that was coded to the wrong department, so the F&B manager's P&L is wrong, so he makes a staffing decision on bad information. The late fee. The vendor relationship that got a little colder.
Nobody tracks those individually, which is exactly why they add up. On a property this size, missed early-pay terms alone can run into five figures annually — a 2% discount on even a modest share of $2 million in annual vendor spend is real money left in the road.
Now the number is not $18,000. It's somewhere north of $35,000, and it's climbing, because that folder is growing.
Why this is a good automation candidate and the board packet isn't
Because everything in that list is deterministic.
Reading a PDF and pulling the vendor name, invoice number, date, line items, and total is a solved problem. Matching to a PO is a lookup. Coding to a GL account is pattern recognition — this vendor has been coded to 6420 the last ninety-one times, and the machine can say so with 97% confidence and flag the one that looks different. Routing for approval is a rule. Detecting a duplicate is a hash.
What is not deterministic — what should never be automated — is the judgment call. Should we still be using this landscaping vendor. Is a 14% increase on linen acceptable or is it time to bid it out. Why is engineering spend up forty percent this quarter.
That is the controller's actual job. It is the job she was hired for, it's the job she is very good at, and she is currently doing about six hours of it a week because the rest of her time is spent being a very expensive data entry clerk.
The build, honestly
Roughly: invoices arrive by email, portal, or photo. A system reads them, extracts the fields, checks for duplicates, proposes a GL code and department with a confidence score, matches against receiving where it can, and drops it into an approval queue with everything already filled in. The human's job becomes review and approve, not read and retype. Anything below a confidence threshold, or above a dollar threshold, gets kicked to a person by design.
Build cost for a property or small group: meaningful, but well inside the range where a $35,000 annual problem clears a 2x return in year one and keeps returning after. That's the bar. If it didn't clear the bar, we'd say so.
The second-order effect is the one owners actually end up caring about. When invoices are coded within a day instead of within a month, the P&L closes faster, and a GM who can see accurate departmental spend on the 3rd instead of the 20th makes better decisions for seventeen more days a month. That's not a savings line. That's an operating advantage, and it's very hard to put on a slide, which is why almost nobody sells it.
The folder is a symptom
Here's the honest close. The AP – TO CODE folder is not a filing problem. It's a signal that your finance function is spending its scarcest hours on the least valuable work in the building, and has been for years, because there was never a good alternative.
There is now. And the point of using it isn't to have a smaller finance team — it's to have a finance function that actually functions: one that tells you what's happening in your business while you can still do something about it, staffed by people who have the time to think.
Go look at your version of that folder. Count what's in it. The number will tell you what to do next.