A scanned invoice vs an emailed PDF invoice differs in one thing that matters: the text layer. An emailed PDF from a vendor's billing system usually contains its characters, so they can be copied exactly. A scanned invoice is a picture of a page, so every character must be recognised first, and recognition can misread digits.
What separates a scanned invoice from an emailed PDF invoice?
A scanned invoice is an image of a paper page, made by a scanner, a phone or a fax. An emailed PDF invoice produced by a vendor's billing system contains the characters themselves, stored as text in the file. That stored text is called the text layer, and it is the difference that changes everything downstream.
Both can look identical on screen. Both are often PDF files. The difference is inside: one holds pixels, the other holds letters and digits with fonts attached.
When the text layer is there, software reads the characters directly, with no guessing about whether a smudge is a 3 or an 8. When it is not, optical character recognition (OCR) has to turn the picture into text first. The OCRmyPDF project, an open-source tool for this job, describes what OCR adds: text that can be selected, searched and copied, in the words of the OCRmyPDF introduction.
Scanned invoice versus emailed PDF invoice, side by side
| Scanned invoice | Emailed PDF invoice | |
|---|---|---|
| Text layer | None until OCR adds one | Usually present from the billing system |
| Typical errors | Misread digits, lost decimal points, columns merged, stamp or handwriting read as text | Correct characters mapped to the wrong field, such as the customer number taken as the invoice number |
| File size | Larger per page, because each page is an image | Smaller per page, because text takes little space |
| Select-text test | Selecting text highlights the whole page or nothing | Individual words and numbers highlight |
| What to ask the vendor for | The original PDF from their billing system, or a portal download | Nothing more; keep the file as sent |
How to tell whether a PDF invoice has a text layer?
To tell whether a PDF invoice has a text layer, open it in any PDF viewer and try to select the invoice total with the cursor. If individual characters highlight, the text layer is there. If the whole page highlights as one block, or nothing highlights, the page is an image and will need OCR.
Then copy and paste. Select the total and the invoice number, copy them, and paste them into a blank spreadsheet cell or text file. Three results are possible:
- The values paste exactly. The PDF has a real text layer from the billing system.
- Nothing pastes. The page is an image with no text layer.
- Something pastes, but with odd characters, spaces in the wrong places or wrong digits. The page is an image that someone has already run through OCR. The text layer exists, but it is a guess and needs the same checks as any OCR output.
For a firm that handles many files, two free command-line tools from the Poppler project give the same answer faster. The pdffonts manual page says the tool lists the fonts used in a PDF, so an empty font list points to a file that is images only. The pdftotext manual page says it converts a PDF to plain text, and its -layout option keeps the original physical layout as far as possible, which makes an invoice's columns easy to inspect.
Which invoice format gives cleaner extracted data?
An emailed PDF invoice with a real text layer gives cleaner extracted data than a scan, because the characters are exact and no digit can be misread. It still needs field mapping, which means deciding which correct number is the total, which is the invoice number and which date is the due date.
Exact characters remove the most common OCR errors: 0 read as O, 1 read as l, a lost decimal point, a stamp read over the total. Those errors are the hardest to spot, because the wrong value still looks like a value.
Mapping errors remain on native PDFs. A vendor prints an invoice number, an order number and an account number in the same corner. Correct text does not stop software from picking the wrong one. The arithmetic check, subtotal plus tax and freight equals total, is still worth running on every invoice whatever its format.
If a client can choose, ask for the vendor's emailed PDF or a download from the vendor's portal rather than a scan of a printout. A firm logging invoices as they arrive can add a column for the format, so a client who sends only scans can be asked for the originals; the invoice tracking spreadsheet post lays out such a log. Bank statements follow the same pattern, covered in scanned vs digital bank statement PDF.
Emailed PDF invoices that are really scans
Some emailed PDF invoices are really scans. A vendor or client prints the invoice, runs it through an office scanner or a phone scanning app, and emails the result as a PDF. The attachment looks like a native PDF, but it is an image inside a PDF wrapper, with no text layer or only an OCR guess.
Clues that a PDF attachment is a scan:
- The page is slightly crooked, or the margins are grey or uneven.
- A stamp, signature or handwritten note sits on the page.
- The file is large for a one-page invoice.
- The select-text test highlights the whole page as one block.
- The email came from the client, not from the vendor's billing address.
When a scan already carries an OCR text layer, that layer can be poor. The OCRmyPDF cookbook notes the tool normally exits with an error when asked to process a file that already has OCR, offers --redo-ocr to replace it, and --force-ocr as a fallback that rasterizes every page. The lesson for a firm is simpler: an earlier OCR layer is not the same as a native text layer, so treat the file as a scan and check its totals.
Stamps, handwriting and faxed pages each cause their own misreads. Which marks cause which errors, and what to check for each, is covered in invoice OCR edge cases: stamps, handwritten notes and faxed pages.
Schema CPA accepts both kinds in the same client file: a vendor's PDF forwarded by email and a phone photo of a paper invoice go through the same upload link, and an invoice forwarded a second time is flagged rather than counted twice. The client tracker demo shows how a client's documents appear once they are in. When the invoices are in, the post on invoice to Excel covers turning a stack of them into one sheet.