Document processing with AIIncoming documents become checked data.
Invoices, orders and forms arrive as PDFs, scans or e-invoices, and someone often retypes them by hand. We build processing that classifies each document, extracts the agreed fields and checks them. Your software receives structured data, and your team resolves the uncertain cases.
Documents we process
Capture incoming invoices
Supplier, invoice number, date, line items, amounts and currency arrive in your software as one record. The original stays stored, and every value points to where it appears in the document.
Import orders
We read orders sent as email attachments and match item numbers and quantities against your master data. Your team sees any discrepancy before the order is created.
Classify forms and applications
We recognise the document type, assign it to the right case and extract the fields the next step needs. Missing required details are flagged straight away.
From inbox to checked record
- 01
Define documents and fields
We gather representative examples with you, including poor scans and rare layouts. Together we decide which document types and fields the processing covers and which software receives them.
- 02
Extract and check
We read structured e-invoices such as XRechnung or ZUGFeRD directly from their XML. We use OCR when text first has to be recovered from an image. Rules then check required fields, amounts, currency and duplicates.
-
03
Route uncertain cases for review
A confidence score alone does not prove a value is correct, so we combine it with the checks. When a value is missing or a total does not add up, your team sees the document with the spot marked and corrects it directly.
- 04
Test and hand over
We measure errors per field on examples the processing has not seen before. We also test that a retry after a failure writes each record to your software only once.
-
Structured data with its origin
Your software receives the agreed fields as a record, linked to the original document and the location of each value. Your team reviews every record or just the flagged cases, following your approval rules.
You see for each field how often it is correct on your test documents. We add new layouts to the same test set and see whether a change makes existing results worse.
How much should run automatically?
The processing can prepare data for review or pass checked data on directly. Decide for each document type when your team confirms.
How one invoice moves through the processing
An invoice arrives by email as a PDF. The processing recognises the type, extracts supplier, number and amounts, and finds that the supplier, invoice number and date match an existing invoice. It holds the record and shows your team both documents side by side.
Automate the next stepsCommon questions
Does the processing post invoices automatically?
It delivers structured data to your accounting or ERP software. Your approval rules and your team decide whether and when anything is posted there. We set up the handover as agreed for each document type.
How do you handle poor scans?
We include poor scans in the test examples from the start. When OCR reads a value with low certainty or not at all, the document goes to your team with the spot marked.
Which software can receive the data?
We connect the software you use today through its interface or an agreed import format. At the start we review the target system and state the effort for the integration.
Which documents does your team retype today?
Tell us which documents you process and which software needs the data. We’ll look at anonymised examples together.