If there's one invoice (always the same one, always long, always from the same supplier) that never quite matches what you ordered, the cause is almost never that the supplier billed you wrong. In most cases, the real problem is how that document gets read: a cover page with a summary that gets added on top of the line items, or pages after the first one that don't get read the same way it does. And that kind of mistake never raises a flag: the total still looks "reasonable" even when it's wrong.
The symptom isn't an error, it's a list that never empties out
Any business that buys from several suppliers ends up, at some point, with a short list of invoices waiting to be checked because their total doesn't match the sum of their line items. Normally that list clears out every week. The problem shows up when the same invoices stay on it week after week, and they're almost always the longest ones: documents running dozens of pages with thousands of products, the kind nobody checks line by line by hand because it would eat a whole afternoon.
That's exactly what was happening to one of our clients, who receives documents like this from several suppliers. The weekly list of invoices that didn't add up was sitting around 105 every week. Looking closely at the longest ones turned up the reason: it wasn't a supplier problem, it was a reading problem.
Why does this happen more with long invoices?
Two different causes, both with the same underlying pattern: the system assumes whatever is true on the first page holds for the whole document.
The first page sets the rules, and every other page has to fit them. A PDF invoice reader usually figures out where each column starts and ends (product, quantity, price) by looking at the first page. If a later page shifts those columns even slightly, the reader throws out that whole row instead of flagging it as unreadable. No warning appears: products just go missing, and the total comes out lower than it should with nothing on screen explaining why.
The cover page repeats what's already added up inside. Some suppliers open the document with a summary page (taxes, totals by category) before the line-by-line detail. If the system adds up that summary and then also adds up the detail on the following pages, the same amount gets counted twice. The result is the opposite of the first cause: the total comes out higher than it should, and it also looks perfectly reasonable at a glance.
Both issues were fixed by reading every page with its own header (instead of inheriting the first page's) and teaching the system to recognize when a table is just a summary of the others, so it never gets added twice. With that, our client's weekly list of unmatched invoices dropped from 105 to 22. Of the 12 long invoices checked by hand in full detail, 7 ended up matching exactly.
What can you check yourself, without touching any software?
You don't need any technical knowledge to run this check. Next time a supplier you deal with regularly sends you an invoice longer than ten or fifteen pages:
- Add up only the product detail by hand, leaving out any summary page at the start or end, and compare it to the total the invoice declares. If it doesn't match, look for a summary page that's getting added in on its own.
- Count how many product lines the invoice actually has, and compare that to how many lines made it into your accounting or management software. If fewer came through than are on paper, some are getting dropped along the way, almost always on pages other than the first.
- Ask, don't assume, when someone last checked one of those long invoices line by line by hand. If the answer is "never" or "months ago," that's exactly the invoice to look at first.
The decision rule is simple: if your invoice process, manual or automated, decides how to read a page by looking only at the first one, and one of your suppliers sends you long documents, you have a blind spot there, even if everything else works fine. You don't need AI to find it: the same check, done by hand, uncovers the same problem.
What changes once you know
Before the fix, someone on the team had to open every one of those long invoices, count the lines, and match them by hand, week after week. Now the system reads them correctly the first time and only sends up for review the ones that genuinely have something off, not every invoice that happens to be long. It's the same work we do for other clients who receive PDF invoices from multiple suppliers: read them, match them line by line against the order, and only post what actually matches. You can see how it works in a real case at invoices that book themselves.
If you want to look at how invoices come into your own business today, with or without AI, our services page walks through how we approach it. And if you'd rather talk it through directly, get in touch and we'll look together at where your reconciliation is leaking.
