Skip to content
OCR-AI
Open a document
All articles

Recognising document types: invoice, receipt or delivery note?

Before extracting data you need to know what the document is. How the type is recognised, why it shapes the fields you get, and what to do when it is wrong.

Published February 5, 2026Updated October 25, 20263 min read

An invoice, a receipt and a delivery note can look almost the same: a logo at the top, a table in the middle, an amount at the bottom. But the data you need from each is different, and so is what you do with it next. So the first step in reading a document is working out what it is.

Why the document type matters

  • The fields differ. On an invoice you look for an allocation number and a due date. On a receipt, for the payment method. On a delivery note, quantities matter more than prices, which are sometimes not shown at all.
  • The checks differ. On an invoice you check that the VAT adds up. On a delivery note you check what arrived against the order.
  • The destination differs. An invoice goes to the books, a delivery note to the warehouse, and a letter to the archive.

Which types are recognised

OCR-AI recognises the document type of each page and shows it above the result:

TypeExamples
InvoiceTax invoice, proforma, an invoice from a supplier abroad
ReceiptTill receipt, receipt for a payment
Purchase orderAn order sent to a supplier
Delivery noteWhat comes with the goods
QuoteAn offer before an order
Bank statementA bank or card statement
PayslipA monthly payslip
ID documentID card, licence, passport
ContractAn agreement, an appendix
LetterA letter, notice or request
FormA filled-in form
TableA page that is mostly a table
DocumentAnything that fits none of the types

How the type is recognised

Recognition is based on the content, not the file name. The engine reads the whole page and relies on what a person would rely on: the title ("tax invoice", "receipt", "delivery note"), the fields present, and the layout of the page. A document that combines two types, such as a "tax invoice/receipt", gets the main type, and the fields of both come out when they are printed.

The type is set for each page separately. A PDF holding an invoice followed by a delivery note gets a different type on each page.

When recognition is wrong

Mistakes happen mostly with documents that have no clear title, or pages that look like another type. Two ways to deal with it:

  • Upload from the matching page. Uploading from invoice OCR, receipt scanner or PDF to Excel tells the engine what to look for, and it concentrates on that type's fields and tables.
  • Check the fields, not only the label. Even when the label is off, the fields that came out are what matters. If a field is missing, you can find the value in the page's full text.

Sorting an archive by type

The document type is kept in the result. In a JSON export it appears for each page as document_type, so you can sort a whole archive by type in another tool. In the workspace, the history in your browser lists recent documents, so you can return to any of them without reading it again.

More from the blog