Advanced users8 min read

Multi-Step Automations

This page focuses on sequencing, validation, operational safety, and handoff points in workflow design.

Design principles

  • Keep each workflow purpose-specific.
  • Validate source quality before expensive downstream steps.
  • Add monitoring and retry logic where files may vary in quality.
  • Separate human-review checkpoints from fully automated paths.

Example chain

  1. 1

    Collect files through a guest upload link into a workspace folder.

  2. 2

    Run a workflow on those files: repair or deskew the source if needed, then run OCR.

  3. 3

    Save the output to a workspace folder, then index it for the AI Document Assistant or review it.

Where human handoff still matters

  • Legal or policy changes that require interpretation, not just extraction.
  • Hiring decisions that need evidence review and recruiter judgment.
  • Low-quality source files that fail validation repeatedly.

Design mistakes to avoid

  • Automating a process before the team agrees on its true output.
  • Skipping validation to make the workflow look simpler than it is.
  • Combining unrelated business purposes into one oversized automation.

Ordering steps

The order of steps changes the result. These orderings follow from how each tool works.

  • Repair first when a PDF fails to open or converts inconsistently; every later step depends on a readable file.
  • Deskew before OCR. Deskew PDF re-renders the pages it straightens as high-resolution images, so any text layer on those pages is lost and OCR should run afterwards.
  • Flatten before compressing when form fields or annotations should be fixed into the page.
  • Compress last, once the content is final, so later steps don't undo the size savings.