Recognition of Expense Reports

Incoming expense reports are now automatically recognized and processed in the background as (received) invoices with a reduced set of fields—without creating a new, visible document type.

How does the detection work?

Recognition is performed entirely by the LLM. No full-text search or pattern recognition is performed—the model evaluates only the document title on the first page of the document.

If this title contains one of the known keywords for expense reports, the document is classified accordingly. Other pages in the document are not taken into account for this assessment.

The following keywords in the document title currently trigger recognition:

• Spesenabrechnung
• Auszahlungsbeleg
• Expensesummary

The list is based on the documents reported so far and will be updated as needed.

What is detected?

An approved expense report is always tracked internally as document type “invoice-i” and is read as follows:

• Document Type: Invoice (Received)
• Supplier: corresponds to the person who submitted the expense report
• Total
• Currency
• IBAN
• Date

All other information is not recognized, particularly the VAT and individual line items. There are currently too few examples of these fields in the submitted documents to ensure reliable recognition.

Requirements and Setup

• This applies equally to DeepO and AbaScanPro—no distinction is made.
• Recognition is based solely on the document title on the first page.
• The document title must contain one of the stored keywords.

No setup by customers or partners is required. Recognition occurs automatically in the background and cannot be configured. No new, visible document type is displayed in the program—the processing remains invisible to users.

Further information

How will the document be processed next?

The document remains an invoice and is transferred to Accounts Payable (Kredi) as requested.

What happens with VAT and line items?

This information is not currently recognized and remains blank because there are not yet enough sample documents for reliable recognition.

Are there any differences between DeepO and AbaScan?

No, the recognition works the same way in both products.

What should I do if a document title doesn’t trigger recognition?

Please report the relevant example so that the list of keywords can be updated as needed.