DashboardSign inStart your trial

Batch & Task Approvals

The four approval types, fanning one step out into many approval cards, and human work steps that collect data

Not every approval is a yes/no on a single item. A step can ask for a choice, collect free text, fan out into dozens of individual cards, or hand a reviewer a checklist of work to do. This page covers the approval types and the two advanced shapes: batch (fanout) and task.

#Approval types

Each Human Review node has one approval type, set on the node:

TypeWhat the reviewer does
Approve / RejectApprove or reject the proposed action, with an optional comment. The default.
Multiple ChoicePick from predefined choices you configure. Each choice has a label (what the reviewer sees), a value (passed to the next step), and an optional color.
Freeform ResponseType a free-text response. Useful for feedback, notes, or open-ended input the workflow reads downstream.
TaskComplete a unit of work, not just a decision: instructions, a checklist, and fields to fill in. See Task approvals .

The choice or response flows to the next step like any other node output, so a Multiple Choice approval can branch the workflow and a Freeform Response can feed a later AI or data step.

#Batch (fanout) approvals

When a step produces a list, one Human Review node can fan it out into one approval card per item instead of a single wall-of-text card. Thirty drafted follow-ups become thirty cards you can work through, or approve in bulk.

You configure fanout with two fields on the node:

  • Items expression: points at the array to fan out (for example, the drafts produced by an earlier AI step).
  • Item template: maps each item onto its card using simple dot-notation paths:
    • Title path (required): the card headline.
    • Description path: the card body.
    • Image URL path: a preview image.
    • Meta path: extra metadata to show.
    • Confidence path: a 0–1 score on each item. When it lands at or above the node’s confidence bar, that individual item auto-approves the moment it’s created and never reaches your queue.

All the cards from one run share a single batch, so they stay linked as a group.

Item cap. A single fanout is capped at 50 items by default, up to 100. If the list is longer than the cap, the step fails rather than silently dropping items. Split the work or filter the list upstream.

In the queue. Each item shows up as its own card. Select several at once and the batch action bar lets you Approve or Reject all of them in one move, so you can clear routine items in bulk and slow down only on the ones that need a closer look. Per-item confidence auto-approvals never appear. They’re already decided.

What the workflow gets back. After the batch resolves, the node outputs a decisions array: one entry per item, each with the item’s index and its decision (approved, rejected, or edited), plus the batchId. Downstream steps read this to act only on the approved items (send the approved drafts, skip the rest).

#Task approvals

A task approval is a human work step, not a yes/no gate. Use it when you need a person to do something mid-workflow: run a QA checklist before a send, paste in a discount code, or enter data the workflow can’t derive on its own.

A task node is built from three optional blocks:

  • Instructions: free-text guidance for whoever picks up the task.
  • Checklist: a list of steps to tick off. Any item marked required must be checked before the reviewer can finish.
  • Collect fields: inputs the reviewer fills in. Each field has a label and a type (URL, text, or number; text by default), plus an optional placeholder and a required flag.

Downstream output. When the reviewer finishes, the collected fields become the node’s output, keyed by each field’s key, so later steps can read them ({{ node.discount_code }}). If the reviewer skips the task, the node outputs a skip marker instead, which a Data node can branch on to take a different path.

#Related