Skip to content

PDF to e-invoice

What for. To turn an ordinary PDF invoice into a real e-invoice: read it out, review it, release it.

PDF only, up to 15 MB: “PDF invoice (scans too) — max. 15 MB” and “Scanned PDFs are processed by text recognition (OCR) — this can take a few seconds per page.”

Possible rejections: “Please choose a PDF file.”, “The file is not a valid PDF file.”, “This PDF file could not be read.”

The PDF is read, with a hard time limit of 60 seconds. Per page a decision is made: text layer or text recognition. Pages with fewer than 100 legible characters, or with more than 30 % illegible characters, go through OCR, whose language follows the portal language, then German, then English.

Fields, parties, line items and totals are derived from labels, zones and columns, and then cross-checked. If the PDF carries a Swiss QR code, its content overrides the inferred payment details.

The currency is determined from the code in the text, the currency symbol or the seller’s country — and every currency the recognition can name is also carried as such. This expressly includes the world beyond the euro and the dollar (TWD, CAD, INR, TRY, ZAR, ILS, SAR and others). RMB — the spelling customary on Chinese invoices — is normalised to the ISO code CNY.

Two columns: on the left the original with the sources highlighted, on the right the recognised values. “Clicking a highlight jumps to the field — and the other way round.”

Highlights and field borders are colour-coded by confidence: certain from 0.85, check from 0.60, uncertain below that. Because colour alone must not carry information, every field also carries the same value as a hidden label (“Recognition: check”) for screen readers.

Put up for review are: invoice number, invoice date, due date, delivery date, service period, invoice type, currency, buyer reference, per party the name, VAT ID, street, postcode, town, country, email and telephone, the line items (description, quantity, unit price, tax rate, unit) as well as IBAN, BIC and payment terms.

If no line item could be read, an empty row appears there — so that the gap is visible next to the original. Without at least one line item you cannot continue: “Please enter at least one invoice line item — without one, no e-invoice can be produced.”

What the individual messages mean is listed under PDF import: messages.

At the bottom you choose the target format (XRechnung by default) and click “Apply & create e-invoice”. You then land in Create, and with everything the recognition found — including the fields the compact review does not even show. Mandatory checks, the format gate and the quota only take effect there.

  • Text recognition needs a licence and installed language data. Without both, the import still runs — only without OCR, and with the corresponding warning. At most 10 pages per import go through OCR, and OCR is serialised across the process: parallel imports wait for one another.
  • The label vocabulary covers 14 languages: the nine portal languages plus Czech, Swedish, Danish, Norwegian and Turkish.
  • There is no learning. Your corrections do not improve the recognition; only which fields you changed is logged, never what you changed them to.
  • The result and the uploaded PDF live only in transient storage; once the review has expired, only a fresh upload helps (“Upload a new PDF”).