Human-in-the-Loop: Approval Steps

Not every automation should run from start to finish without any human review. Chase Agents provides a built-in mechanism to pause an automation at a critical decision point, send an approval email to a designated person, and only continue once that person has explicitly approved. This is called human-in-the-loop governance.

The seek_human_approval_for_next_step Step Type

When your automation reaches a seek_human_approval_for_next_step step, execution pauses. The platform sends an email to the address you configure, displaying a summary of what the automation has done so far and what action it is about to take. The email contains two buttons: Approve and Reject.

Once the recipient clicks one of those buttons, the step completes and returns a result object with three fields: success (a boolean indicating the step ran correctly), choice (a string containing either Approve or Reject), and message (any optional note the approver provided).

Configuring the Approval Step

When adding a seek_human_approval_for_next_step step, you specify the email address of the person who should receive the approval request, a subject line for the email, a message body that explains the context including what the automation has done and what it is about to do, and an optional timeout duration. If the approver does not respond within the timeout window, the step resolves with a default outcome you configure.

Reading the Result in Subsequent Steps

After the approval step completes, the next step in the automation can access the result via the dataPrev variable, which holds the output of the immediately preceding step. Specifically, dataPrev.choice will equal either Approve or Reject. Use an if step immediately after to branch based on that value: if dataPrev.choice equals Approve, proceed with the action; otherwise, take the reject path such as notifying the requester, logging the decision, or stopping the workflow.

Common Use Cases

Human-in-the-loop steps are most valuable when automations operate on high-stakes or irreversible actions. Examples include sending a bulk email campaign where a manager should review the draft before it goes out, issuing a refund or credit that requires finance approval before the transaction executes, deploying a configuration change that needs a senior engineer to sign off before the automation modifies infrastructure, and publishing generated content that requires a content lead to approve AI-written copy before it goes live.

Keeping Approvers Informed

Write approval request messages that give the reviewer everything they need to make a confident decision. Include the specific data the automation processed such as the number of records, the customer name, or the dollar amount, the exact action about to be taken, and a clear explanation of what Approve and Reject mean in this context. Vague approval emails slow things down and often result in rejections out of caution.

Viewing Approval History

Every automation run that included an approval step is logged in Workflow History. You can see when the approval request was sent, when the approver responded, what their decision was, and any message they provided. This audit trail is available to workspace admins and is useful for compliance reviews.

Combining with AI Auto-Fix

Approval steps and AI auto-fix are complementary governance mechanisms. Approval steps pause execution and wait for a human decision. Auto-fix runs after execution and attempts to repair a failed step automatically. For maximum reliability on sensitive workflows, consider using approval steps before high-stakes actions and enabling post-execution auto-fix to recover from unexpected runtime errors after those actions complete.