Running flows in batch

Execute Flows across multiple inputs simultaneously using batch processing. Useful for bulk operations, testing, and data processing tasks.

When to use batch processing

• Bulk operations: Process hundreds or thousands of items
• Testing: Run Flows against a dataset to verify behavior
• Data migration: Transform or enrich existing data
• Evaluation: Compare Flow performance across many inputs

Running a batch

1. Open the Flow
2. Click Run
3. Select Batch tab
4. Choose input source:
   • Upload CSV file

• Select Record collection
• Paste JSON array

5. Map columns to Flow input variables
6. Click Start Batch

CSV input format

CSV files should have a header row with column names matching your Flow variables:

customerId,message,priority
12345,"Need help with order",high
67890,"Question about pricing",low

Each row becomes one Flow execution.

JSON input format

Provide an array of objects:

1
[
2
  {"customerId": "12345", "message": "Need help with order", "priority": "high"},
3
  {"customerId": "67890", "message": "Question about pricing", "priority": "low"}
4
]

Using Records as input

Run the Flow against a Record collection:

1. Select Record collection as input source
2. Choose the Record type
3. Optionally filter Records
4. Map Record fields to Flow variables
5. Click Start Batch

The Flow executes once per Record.

Monitoring batch progress

After starting a batch:

1. Go to Logs → Batch Jobs
2. Find your batch execution
3. See:
   • Total items

• Completed count
• Failed count
• Progress percentage
• Estimated completion time

Batches run in the background—you can close the browser.

Batch results

After completion, download results:

1. Go to the batch job page
2. Click Download Results
3. Choose format:
   • CSV (inputs + outputs)

• JSON (full execution details)

CSV includes all input columns plus an output column with the Flow result.

Handling failures

If some items fail:

1. Download the failed items CSV
2. Review error messages
3. Fix issues (update Flow or data)
4. Re-run the batch with only failed items

See Handling batch failures for detailed troubleshooting.

Batch limits

Batch size and concurrency limits depend on your plan tier. For large batches, split into smaller batches to make monitoring and error recovery easier.

Cost considerations

Batch processing can consume significant AI credits:

• Each item counts as one Flow execution
• 1,000 items = 1,000 executions
• Monitor usage to avoid surprise bills

Use smaller models (e.g., GPT-4o-mini) for batch operations when possible.

Scheduling batches

Automate batch processing with schedules:

1. Go to Schedules
2. Create a schedule
3. Select your Flow
4. Configure to run against a Record collection
5. Set frequency (daily, weekly, etc.)

See Automating batch processing for details.

Best practices

• Test with small batches first: Run 10-100 items before scaling to thousands
• Monitor costs: Check usage estimates before starting large batches
• Handle failures gracefully: Design Flows to fail gracefully on bad input
• Use appropriate concurrency: Higher concurrency is faster but can hit rate limits
• Save results: Download and archive batch results for auditing

Next steps

• Handling batch failures to troubleshoot errors
• Automating batch processing with schedules
• Running an evaluation for model/prompt testing
• What are records? to understand Record-based batches