Invoice OCR and KSeF: what should you scan or import?

Import the data when an invoice is available in KSeF; use document extraction for a photo or scan received outside that system. Invoice OCR helps transfer information from an image into software fields. KSeF supplies invoices in XML format. Both routes can coexist, but the same purchase should be recognised as one transaction.

This guide covers incoming purchase documents for businesses operating in Poland. It helps you establish a practical rule for supplier files, scanned documents and KSeF imports. It does not determine VAT rates or provide a list of documents that always qualify as deductible expenses. Reading the document correctly is the beginning of the accounting workflow.

What is the difference between OCR, AI extraction and XML import?

OCR stands for optical character recognition. It reads text from a photograph or scan. A further step assigns the values to fields such as invoice number, supplier or amount. Tools can combine OCR with AI models that interpret document layouts. The label “AI” alone does not tell you which fields have been extracted correctly.

For example, Microsoft’s invoice model documentation describes extracting header information and line items from documents. This illustrates the technology; it does not identify Norman’s processing provider or guarantee the accuracy of any particular application.

XML imports start with values already stored in defined fields. Poland’s Ministry of Finance specifies XML as the format for KSeF structured invoices. Printing, scanning and reading those values again usually adds an unnecessary step. That is a practical conclusion about the data format, not a prohibition on using a printed copy.

Which route should you use for each document?

Establish the source first. A file named “invoice.pdf” does not tell you whether the invoice originated in KSeF. It could be a visual representation of an invoice, a separate supplier document or a scanned sheet of paper. Check the number, supplier and existing imports before adding another copy.

If a PDF contains text, avoid replacing it with a screenshot. The original file gives the software more opportunities to read its contents than a reduced image. When a supplier sends several attachments, identify the invoice and relevant supporting material instead of treating every attachment as a separate purchase.

When does scanning invoices still make sense?

Scanning remains useful when you actually have paper or an image and are not importing that document from another source. For a file from an overseas supplier, first establish how that particular invoice reaches you rather than assuming it will appear in Poland’s KSeF. A foreign language or currency does not, by itself, determine its accounting treatment.

Make sure a photograph clearly shows the number, dates, parties and all amounts. Lay the document flat, remove shadows and include its edges. Check focus before uploading. For a document with several pages, preserve their order and ensure that the final summary belongs to the same invoice.

Do not alter the image in a way that changes the document’s contents. Ask the supplier for a better copy when a section is unreadable. Entering a guessed amount may remove a warning in the application, but it does not resolve the missing evidence. Keep a clear note of what needs checking until the issue is settled.

How can you organise both routes in Norman?

For KSeF, follow the token connection instructions. For other files, open the Documents page and choose Upload. The current expenses and documents guide covers PDF, JPG, PNG, WEBP and XML uploads, forwarding documents by email and adding photos through the mobile application.

There is no need to retype an invoice simply because it arrived through a different channel from the previous one. Set one consistent rule: retrieve available invoices through KSeF and upload the remaining documents as files. When an invoice arrives again, search for its existing record by supplier and invoice number first.

The extracted-data section in the document details lets you compare the result with the source. Before adding your company’s real files, you can try Norman’s sample receipt. This introduces the review screen without re-entering documents from a period you have already settled.

What should you check after extraction?

Work from document identity to amounts, then to the meaning of the purchase. Reading the large number at the bottom of a page is insufficient if the software has confused currencies, dates, the seller or the buyer.

1. Parties and number: compare the supplier, tax identification number and invoice number. Confirm that the document belongs to your business.
2. Dates: distinguish issue date, sale date and payment deadline. Do not substitute one for another just because a field is empty.
3. Amounts and currency: check net, VAT, gross and any earlier payments. The outstanding balance may differ from the invoice total.
4. Line items: when there are several rates or different purchases, compare the details rather than only the final sum.
5. Document type: establish whether it is an invoice, correcting invoice, pro forma or confirmation. Similar layouts do not give documents the same purpose.

Then review the category and linked transaction. A Polish sole trader, or JDG, keeping a KPiR uses that register’s categories; a sp. z o.o. works with accounts in its accounting books. Uploading a file does not establish VAT deductibility or whether an expense qualifies for income-tax purposes. The purchase documents and input VAT guide explains the relevant Norman controls.

Example: three files, two purchases and one payment

Consider this fictional document-handling example. A business has invoice A for PLN 1,230 imported from KSeF, a PDF of that same invoice A and a scan of invoice B for PLN 246. Its bank records show a PLN 1,230 payment for A. Before settling the period, check that there are two purchases totalling PLN 1,476 gross, rather than three entries totalling PLN 2,706.

The PDF of A does not represent another purchase. The PLN 1,230 payment does not establish that B has been paid either. Identify the documents first, then connect the transfer to the correct liability. If B’s number is unreadable, resolve it before trying to match documents using the amount alone.

This example checks document handling and outstanding balances, not tax deductions. It does not assume that the whole gross amount reduces taxable income or that all VAT is recoverable. The correct treatment depends on the business and the purchase involved.

How should you finish the review?

If several people review documents, agree who will ask suppliers for missing information and who will check the accounting treatment. Record a specific reason for each exception, such as a missing second page or an inconsistent currency. The next reviewer should understand what remains to be done without reconstructing the entire email exchange.

In Norman, marking a transaction as completed records your review. According to the product documentation, a transaction with the required information can already affect KPiR and JPK_V7 before that mark is applied. Check the category and amounts in the transaction and register, rather than relying solely on the task list.

Keep three questions separate: has the document been read, has it been accounted for correctly, and has it been paid? If you later prepare a report, check submission and the receipt confirmation separately. A successful document-extraction message does not confirm payment of either an invoice or a tax liability.

If your documents are disorganised, begin with one supplier and one period. Record where invoices arrive, who reviews exceptions and where you check for missing items. Then extend that same process to the remaining documents. For a broader evaluation, use our five tests for AI bookkeeping and explore the available capabilities on Norman’s AI bookkeeping page.