Document Data Extractor

Why template-based OCR is a thing of the past

Conventional OCR software works like a rigid stencil: you define coordinates, and as soon as a layout shifts, the process fails. The Data Extractor uses large language models to understand documents semantically – just like a human. It does not look for pixel positions, it looks for meaning.

Example: Instead of maintaining 50 templates for 50 suppliers that break with every layout change, the AI recognises the “amount due” in the top right for supplier A and the “grand total” in the bottom left for supplier B – without predefined rules.

How extraction works

1. 1

   A document comes in

   Via REST API from your software or workflow – as a PDF, scan, photo or Word file.

2. 2

   The AI reads the content

   The model understands the document and fills exactly the fields you defined in your JSON schema.

3. 3

   Automatic validation

   Every result is checked against your schema – missing required fields or wrong formats stand out immediately.

4. 4

   Human in the loop

   Your team reviews uncertain cases in the validation portal before the data moves on.

5. 5

   Handover to your systems

   The validated data reaches ERP, CRM or legacy systems as JSON – automated via Frends if you wish.

One model for every channel

Crumpled phone photos, low-contrast scans, digital PDFs or handwritten notes: the engine recognises text even under difficult conditions and understands several languages without retraining.

Travel expenses example: a French fuel receipt (JPG) yields the “Montant total”, a German hotel invoice (PDF) the “Gesamtbetrag” – both end up in the same field of your database.

Where the Data Extractor takes work off your hands

• Incoming invoices – invoice data from any number of suppliers, without a template per layout.
• Travel expenses and receipts – fuel receipts, hotel invoices and receipts from different countries, cleanly in the same data field.
• Orders and delivery notes – data that used to be typed in by hand goes straight into your system as a record.
• Forms and notes – handwritten documents are processed, too.

Maximum flexibility for your compliance

Data protection is not one-size-fits-all. Use the cloud API for instant scalability – ideal for e-commerce with peak loads such as Black Friday – or run the solution on premises as a Docker container, for example in banking or healthcare. The AI then runs isolated in your own infrastructure, and no sensitive data leaves your network.

Licensing

Model	For whom	Price
API only	High-volume use cases with direct REST API and pay-per-page billing	from €0.10 per page
Hybrid quality	Critical data – the AI extracts, your team reviews in the validation portal	from €100 per month
Enterprise end-to-end	Integration into ERP, CRM or legacy systems, custom workflows with Frends, optionally on premises	on request

All models include the human-in-the-loop workflow. Hybrid quality and Enterprise also come with the validation portal; the Enterprise model adds full system integration, custom workflows with Frends and, if required, on-premises operation.