AI document processing: reading invoices, receipts and contracts
Supplier invoices, delivery notes, contracts and forms arrive as PDFs, scans or phone photos, and are often still typed in by hand. An AI reads the fields you need, validation rules check the result, and your team only sees the cases that are not clear. Sortia Labs sets this up for companies in Munich and throughout Germany, on site or remote.
- AI document processing
- Invoice data extraction
- Receipt capture with AI
- AI OCR
- Intelligent document processing
Which documents the AI reads for you
Supplier invoices and receipts
Supplier, invoice number, date, VAT ID, IBAN and amounts per tax rate, even when one receipt carries several rates. The values go to your accounting or ERP as a posting proposal.
Delivery notes and purchase orders
Line items, quantities and article numbers are extracted and compared with the order or invoice. Differences in quantity or price are flagged before the payment is released.
Contracts and correspondence
From contracts come terms, notice periods, parties and amounts; from letters the sender, the request and any deadlines. The AI summarises and assigns; the judgement stays with you.
Forms and applications
Completed forms, applications and questionnaires become structured data that your system takes over directly. Empty or unclear fields are flagged for your team instead of being guessed.
Recognise and route
Before extracting anything, the AI recognises what kind of document it is and routes it to the right workflow: invoices to accounting, contracts to the archive, complaints to customer service.
How the AI gets to work on your documents
- 1
Sample
You give us a selection of typical documents, including the difficult ones: poor scans, unusual layouts, multi-page invoices. With them we define which fields should be extracted.
- 2
Prototype on real documents
Usually after two to three weeks the recognition runs on your examples. For every field you see how often the value is right and which cases go to review.
- 3
Validation and approval
Totals, tax rates, mandatory fields and the match with master data are checked by fixed rules. Whatever fails appears in a review list with the original document alongside.
- 4
Integration
The extracted data goes to your ERP, your accounting, your document management or as an export to your tax advisor, through an API, a file or a mailbox.
Software we build and run ourselves.
Questions about AI document processing
How accurately does the AI read invoices?
That depends on document quality and layout, which is why we measure it in the prototype on your own receipts, field by field. What matters is the combination with validation rules: anything that does not add up goes to review, and only clear cases pass through without rework.
How is this different from classic OCR?
Classic OCR turns an image into text. The AI also understands what that text means: which amount is the gross total, which number is the invoice number and which date is the delivery date. Separate templates per supplier are usually not needed.
Can the data go straight into DATEV or our ERP?
Usually yes, through the interface of your system or as an import file. Our own tool Fibu, for example, creates a DATEV export file for the tax advisor. Which route fits depends on your ERP and your accounting firm.
Do our documents leave the company?
That is your decision. Processing can run through an AI provider with a data centre in the EU, on your own server or with a local model when invoices and contracts should stay in the building. A data processing agreement is part of every delivery.
Read on.
Which documents does your team still type in by hand?
Send us a handful of typical examples. We show you how well the AI recognises the fields you need and where rules are required.