How does automatic document analysis work?

The Two Steps

OCR (Optical Character Recognition). Pixels become characters. For clean PDF files with embedded text, this step is skipped – the text is already present there.

Extraction. The desired fields are read from the text. Here, language models are significantly better than older rule-based methods because they can handle different layouts.

What Determines Quality

The Most Important Setup

Never book without a verification step. The system delivers values with a security indication. Below a threshold, the document is sent for manual review.

For amounts, an additional plausibility check is recommended: Sum of positions against total amount. A calculation error is detected this way before it is booked.

What Really Reduces Effort

Not recognition, but avoiding paper. An e-invoice requires no recognition – the data is already structured. Document AI is the solution for what cannot be avoided.

Realistic Expectations

With clean documents, high recognition rates are achievable. With mixed batches including smartphone photos, significantly less. Measure with your actual documents, not with the provider's examples.