Frequently asked questions about document processing with AI
(01)Which document formats can the AI process?
Virtually all common formats are supported: PDF (including image-based), scanned paper receipts, Office documents, emails, photos and faxes. The system also handles multi-page documents, rotated pages and weak scan quality. Before the project we test with your real examples to find out where the limits lie.
(02)How accurate is the data extraction?
Accuracy depends on the document type, scan quality and layout variety. Standard fields such as invoice number, date and amount can usually be read very reliably, while free-text fields and complex layouts are more demanding. We measure accuracy on your own documents, set target values per field and improve the models continuously through feedback.
(03)Is automated document processing GoBD-compliant?
Yes, we design the solution to meet the GoBD requirements for process documentation, traceability and immutability. Every processing step is logged and the original documents are archived in an audit-proof way. We create the process documentation required by the GoBD as part of the project, together with you and your tax advisers.
(04)How long does training for a new document type take?
For a standardised document type, such as invoices from a particular supplier, 50–100 annotated examples and one to two weeks of training are often enough. For complex, variable documents we typically need 200–500 examples and two to four weeks. Transfer learning speeds up training for similar document types.
(05)What happens if the AI cannot recognise a document reliably?
Documents with a low confidence score are automatically routed into a manual review workflow. The clerk corrects or confirms the extraction, and this feedback flows back into the model as training data. This keeps quality high, and the share of manual reviews decreases over time.
(06)Can the AI also extract tables from documents?
Yes. Layout-aware models such as LayoutLM recognise table structures, columns and rows even in complex layouts. The extracted tables are output as structured data, for example JSON or CSV, and can be imported directly into your systems – such as the line items of an invoice into your ERP.
(07)How are confidential documents protected?
Processing takes place in encrypted environments with access control. Operation in your own infrastructure is possible, so documents never leave your network. Role-based access, audit logging and automatic deletion after defined periods provide additional protection for sensitive content. We conclude data processing agreements with all service providers.
(08)What does AI document processing cost?
Costs depend on document types, volume and integration effort. The more documents are processed manually today, the faster automation can pay off; we calculate the concrete business case with your figures during scoping. Fixed price after scoping, proposal within 48 hours.