Skip to main content
A batch run takes a list of inputs and runs the workflow once for each item — in parallel where possible. Use batch runs when you have a volume of items to process.

When to use a batch run

  • You have multiple items to process with the same workflow
  • Processing them one at a time would be slow or impractical
  • The items share the same structure
Examples:
  • Extract key clauses from 200 contracts uploaded to @GoogleDrive
  • Screen 500 insurance applications against underwriting criteria
  • Enrich a list of 1,000 company names with data from @companies_house (custom connector)

How to run a batch

  1. Open the workflow
  2. Select Batch run from the Run dropdown menu
  3. Provide your input list — upload a CSV, connect to a data source, or paste a list of values
  4. Felix processes each item and collects the outputs into a single result

Structuring your input

CSV upload — one row per item, with column headers Felix uses as variable names.
company_number, company_name, jurisdiction
12345678, Acme Legal Ltd, England & Wales
87654321, Smith & Partners, Scotland
Connected data source — point the batch at a table or sheet in a connected service.
Run this workflow for each row in the Applications tab of
@GoogleSheets. Use the columns: applicant_name, loan_amount,
application_date.

Human Checkpoints in batch runs

If a workflow includes a Human Checkpoint, execution pauses for each item when it reaches that step. For high-volume batches, consider an alternative:
After processing all applications, generate a summary of items
flagged as high risk and send to @Slack for review.

Tips

  • Test on a small sample first. Run with 3–5 items before processing the full dataset.
  • Keep input data clean. Missing values and inconsistent formatting are the most common cause of batch failures.
  • Write outputs to a connected service. For large batches, writing directly to @GoogleSheets or @Notion is more practical than downloading a single large file.
  • Use exception-only checkpoints for volume. Pausing on every item defeats the purpose. Flag exceptions and review them as a group.