AI document analyzer — supplier bills into a spreadsheet

Suppliers send paper and PDFs. A return wants rows. The gap between those two facts is filled, in most businesses, by somebody typing.

Drop the bills in above — photographed, scanned or emailed — and each one comes back as a row: date, supplier, invoice number, supplier VAT number, net, VAT and total. Suppliers become pages. The register downloads as an .xlsx with a tab per supplier, or opens in the spreadsheet with the formulas still live.

What it actually does

• Reads a photograph. The page is straightened, cropped and levelled in your browser first — the same processing as the document scanner — and that step is most of the difference between reading 1,250 and reading 7,250.
• Extracts the fields a bookkeeper needs, not everything on the page: supplier, invoice number, date, VAT number, and the three totals, plus the line items behind them.
• Checks itself. It adds up the line items and shows that beside the total the document states. When they disagree the row says so rather than choosing one.
• Groups by supplier, folding two spellings of one name into one page.
• Builds the purchase side of a VAT return, boxes 7 to 11, using the same arithmetic as the VAT return summary.

Nothing is final. Every cell in the register is editable, and the treatment box on each row is a dropdown you can change — because a bill with no VAT on it might be zero-rated, exempt, or from a supplier who is not registered, and those are three different boxes. Guessing between them would be invisible and wrong about a third of the time.

This one sends your document somewhere

Every other tool on this site runs entirely in your browser and says so. This one cannot: reading a photograph of an invoice needs a model, a model runs on a server, and a key to that server cannot be handed to a web page. So, plainly:

• The document leaves your device. Each page is sent, as an image, to an AI provider.
• It is not kept. Not by them after the read, and not by us at any point — what we store is a counter, so the free allowance can be counted.
• The offline alternatives are real. A text-layer PDF goes through PDF to Excel in your own browser, free, for nothing. A photo you only need straightened goes through the document scanner, also without leaving the device.

If the documents are ones you are contractually barred from sending to a third party, use the app: it reads them on your own machine.

Free to try, then metered

This one calls an AI model, so it costs real money per document. There is a monthly allowance you can use without paying anything, and the tool prints what is left of it every time it reads a bill. Past that you either wait for the next month or move to a paid plan. Every tool here that does not call a model is simply free — this page says otherwise because it is otherwise.

Accuracy, honestly

It is good at consistent layouts and good at tables. It is weaker on poor scans, on handwriting, and on documents that put critical figures inside a paragraph. Check the totals against the document before posting anything to a ledger — the row shows both numbers precisely so that check takes two seconds.

It does not read a bank statement into a register yet, and a purchase order is read as though it were a bill, which gets the header right and the treatment wrong. Tell us what you are trying to extract and from what. It moves the order this gets built in.

Doing this from a phone

This page takes a photograph, and so does the app — the document inbox there opens the camera, takes a stack of bills one after another without leaving the screen, and files each one against the supplier it came from. That is the difference worth knowing about: here the register is a download, and there it is the purchase ledger the return is built from.