What is invoice OCR, and why is it not enough?
By Tim White · Glossary · Last updated
Invoice OCR (optical character recognition) turns a scanned or photographed invoice into machine-readable text. You need it for paper and photos. On its own it only gives you raw text. It doesn't know which number is the total, and it can't tell you when it misread one.
What OCR does
OCR reads characters off an image. A PDF exported from accounting software already has a text layer, so it barely needs OCR. A scan or a phone photo is only pixels until OCR reads it.
What OCR can't do
OCR gives you a block of text. It won't tell you which line is the vendor or which figure is tax and which is the total. It has no sense of confidence either. If glare turns an 8 into a 3 on the tax line, the wrong digit looks just as certain as the right one.
So OCR has to be paired with extraction that structures the text and checks it. InvoiceJet uses OCR as a fallback for scans and photos, then has two models read the result, checks that the totals add up, and shows the text behind each field.
Better captures, fewer misreads
Most OCR errors start at capture. Lay the page flat, fill the frame, avoid hard shadows and keep the phone square to the paper. A photo taken at an angle in a dim truck cab will produce broken characters no matter what reads it. When a capture is still poor, the per-field confidence tells you which numbers to check.
Common questions
Does InvoiceJet work on phone photos of invoices?
Yes, including HEIC photos from an iPhone. Each field still gets a confidence level, so you can see where the photo caused doubt.
Is OCR enough to automate invoice processing?
No. OCR produces text with no idea which number is the total or whether it was read correctly. You need extraction and checks on top.
Do digital PDFs need OCR?
Usually not. A PDF generated by billing software already contains selectable text. OCR matters for scans and photos.
Sources
Keep reading
Try it on your own invoices
10 invoices a month free, no card. Every field comes back with a confidence level and the text it was read from.