Automated Document Verification with AI

1. Intake — Documents arrive via upload, email, scan folder or API. feld.ai processes PDF, image and electronic formats (ZUGFeRD, XRechnung).

2. Classification — Each document is automatically assigned a type (50+ document types). Mixed bundles are separated.

3. Extraction — Header and line-item fields are read — even on unseen layouts, without prior templates.

4. Verification & validation — Completeness, correctness and plausibility are checked: master data matching, rule and format checks, duplicate detection.

5. Human-in-the-loop on low confidence — Uncertain or deviating fields are presented for review. Corrections feed directly back into model training.

6. Handover to the target system — Verified, structured data is handed over to ERP, accounting or DMS — as a posting suggestion or via the API.

Invoice & Receipt Processing — automatically verify invoices, match against master data and generate posting suggestions.

Freight Invoice Processing — verify freight papers, customs documents and CMR consignment notes with a 3-way match.

Compliance Agent — automatically monitor conditions, deadlines and permits.

Glossary: Visual Grounding and Confidence Score — the terms behind AI verification explained.

What is automated document verification?

Automated document verification means checking documents for completeness, correctness and plausibility with AI — without manual review. The AI reads invoices, freight papers or contracts, extracts the relevant fields, matches them against master data and rules, and flags discrepancies for approval.

How accurate is the AI check?

In production environments, feld.ai achieves a header field recognition rate of over 96%. At a major German logistics company, 96% of freight invoices are processed correctly on the first pass (first-pass STP). Every extracted value is traced back to its source in the original document via visual grounding, and confidence scores make uncertain fields visible.

Is the data processed in the EU?

Yes. All data is processed exclusively on our own hardware in Austria. No US cloud provider is involved, no CLOUD Act exposure. ISO 27001 certification is in preparation.

Can a human intervene?

Yes. When confidence is low, the AI hands the document over to human-in-the-loop review. A team member checks, corrects and approves it — and every correction feeds directly back into model training. This way, verification quality improves continuously.

Which documents can the AI check?

feld.ai verifies invoices and receipts, freight papers and customs documents, contracts, notices and case files — across industries and without rigid templates. New document types can be trained with a few examples.