A scanned or photographed invoice is an image, not data — so before it can go into Excel, you need OCR to recognise the text. On-device OCR extracts the words and numbers so you can move the line items into a spreadsheet, and because it runs in your browser, the invoice stays private.
Why a scanned invoice won't convert directly
A scan or photo of an invoice is just pixels — there's no underlying text to pull into columns. OCR (optical character recognition) reads the characters out of the image first; only then can you get the amounts and line items into Excel.
Step 1: extract the text with OCR
Image to Text recognises the invoice text on your own device — nothing uploaded, which matters for supplier and payment details. You get the numbers and labels back as editable text you can work with.
Step 2: move it into a spreadsheet
Paste the extracted text into Excel and split it into columns (Data → Text to Columns), or line up the items by hand for a short invoice. If your invoice is a digital (non-scanned) PDF, skip OCR and convert PDF to Excel directly instead.
Getting accurate OCR from an invoice
- Use the highest-resolution scan or photo you can.
- Keep it straight and evenly lit; flatten creases and avoid shadows.
- Check the totals and any hand-stamped figures by eye — OCR can misread a smudged digit.
- Plain printed invoices read far better than faint thermal-printer receipts.
Doing this at scale
For the occasional invoice, on-device OCR plus a quick paste is fast and private. For hundreds of invoices a month, dedicated accounts-payable software with automated capture will be faster — but for a handful, you don't need to upload anything or pay for a service.
The private-vs-server trade-off
The OCR step here is fully on-device. The related PDF to Excel conversion (for digital PDFs) runs on a secure server instead — see extract data from a PDF into Excel for when to use which.
What decides whether OCR gets it right
- Resolution: scan at 300 dpi. Below about 200, character shapes break down and digits start substituting for each other
- Contrast and flatness: a photographed invoice at an angle, or with a shadow across it, loses whole columns
- Font: printed invoices read well; handwritten totals and stylised logos do not
- Ruled tables help. A table with no lines and inconsistent spacing is hard for any tool to segment into columns
Always check the numbers before you use them
The failure mode that matters is not OCR refusing — it is OCR being confidently wrong about one digit. 8 and 3, 1 and 7, 5 and 6, and a misplaced decimal are the recurring substitutions. Check the total against the source before the figures go anywhere near an accounting system, because an error that reconciles is far more expensive than one that fails loudly.
The recognition runs on your own device, which is the point for invoices: they carry supplier terms, bank details and pricing you may be contractually obliged not to share.