Scanned PDFs into Excel: the honest version
A scanned PDF is a photograph with extra steps. Nothing about it is machine-readable until OCR runs, and OCR on a bad scan produces confident nonsense. Here is how to tell whether your scan is worth processing.
Check the resolution first
Below about 200 dpi, characters start merging. A 150 dpi scan of a 9pt invoice will lose the difference between 5 and 6. If you can zoom to 200% and the text is crisp, OCR will do well; if it is fuzzy at 100%, no software will save it.
Deskew matters more than resolution
A page scanned at 3 degrees rotation reads much worse than the same page straight. Most scanners and phone apps have a straighten option — use it. This single step improves accuracy more than swapping OCR engines.
Grayscale beats colour, and black-and-white beats both
Colour scans carry compression artefacts around text. Converting to grayscale before OCR usually helps; hard black-and-white can help more, but only if the scan was clean to begin with. If the paper has a coloured background, grayscale is the safer choice.
Verify the totals, not the words
OCR mistakes cluster in names and addresses, which are long and rarely checked. The dangerous ones are in amounts, because a wrong amount propagates into your accounts. Sort by amount, check the outliers, and compare the sum against the document's own total.
Common questions
Can OCR read a photo of a document?
Yes, if it is flat, evenly lit and in focus. Shadows and perspective are the two biggest accuracy killers.
Why does OCR insert strange characters?
Usually a stamp, a signature or a fold crossing the text. Cropping that region out before OCR often fixes it.
Is a text-layer PDF better than a scan?
Always. If you can get the original digital file from the supplier, do that instead of scanning a printout.