Skip to main content

Common Workflows

Under Review

This article is currently under review. Some content may be incomplete or inaccurate.

A few pipeline shapes cover most projects. Use these as starting points, then add or remove activities to fit your documents.

Standard extraction​

The default end-to-end pipeline: convert, read, split, classify, extract, then hand off for review.

Import → OCR → Split → Classify → Extract → Review

Best when uploads may contain multiple documents of different types and you want a person to confirm the results.

Pre-classified, single document​

When each upload is exactly one document of a known type, skip Split and Classify.

Import → OCR → Extract → Review

Set the document class on the project so Extract knows which fields to pull.

Straight-through processing (no review)​

For high-confidence, low-risk documents, drop the Review step and export automatically.

Import → OCR → Split → Classify → Extract → Validate → Export

Use Validate business rules to catch problems, and consider a Notification step to signal completion.

Human-in-the-loop with branching​

Route only the documents that need attention to a reviewer, and auto-approve the rest.

Import → OCR → Split → Classify → Extract → Validate
   → If (has errors?) → true:  Review → Export
                      → false: Export

See Control Flow for how to add the branch.

Restructure before extraction​

When splitting is imperfect (for example, a cover page that belongs with the next document), add a Data Transform step to move pages and reassign classes, then let a downstream Classify or Extract reprocess only the documents that changed.

Import → OCR → Split → Data Transform → Classify → Extract → Review

Export and notify​

Produce output files and tell an external system where to download them.

... → Export → Notification

The Notification step posts a webhook with download links for the Export results.