Most tools can read a clean invoice. The difference shows on the ones that are not clean, and in what happens when a reading is unsure. The costly failure is not a misread figure. It is a misread figure that looks exactly like a correct one and is found weeks later, if at all.
Every invoice is checked by its own arithmetic
An invoice states its own totals. The lines have to add up to the subtotal, and the subtotal plus tax has to equal the total. We check that on every invoice we extract, and any invoice that fails is listed with both figures, so you see the problem before it reaches a ledger. Scans and photos are also read twice by independent methods, and a figure where the readings disagree is listed rather than guessed. See two readings that disagree.
Labels can mislead
A figure labelled “total” may be the total before a discount. A registration number printed beside the tax may belong to the customer, not the supplier. We read each figure by its place on the page and confirm it by the arithmetic, rather than trusting the label next to it.
Not everything in the folder is an invoice
A cost folder collects quotes, payment confirmations and statements as well as invoices. Each carries figures that would enter a total if nothing separated them first. We classify each document before extracting anything, and list the ones that are not invoices. See not everything is an invoice.
Duplicates that share nothing as files
A photographed receipt and the emailed PDF of the same expense have different names, formats and sizes, so a check on the files calls the folder clean. We match on supplier, amount, date and lines, and flag the pair for you to decide. Our note on duplicate invoices from photographed receipts shows the pattern.
Dates that read two ways
A date like 04/09/2025 is April or September depending on who wrote it, and a wrong guess moves an invoice into the wrong period. We settle it from other evidence on the document, or list it as unresolved. See a date with two readings.