Approval Queue
The Approval Queue is where you review AI-drafted messages before they are sent. Any message that exceeds the agent’s risk threshold lands here for human review, giving you full control over what goes out under your name.
Accessing the Approval Queue
Navigate to Inbox > Approval in the left sidebar. The approval queue shows a badge count of pending items so you always know when drafts need attention.
Queue Layout
The approval queue uses a split-panel layout:
- Left panel — List of all pending drafts, showing contact name, company, job type, risk level, and how long ago the draft was created
- Right panel — Full detail view of the selected draft with editing and action controls
Reviewing a Draft
Click any item in the left panel to see the full detail view. Each draft shows:
Contact Information
- Contact name and email
- Company name
- Job type (e.g., “initial outreach”, “follow up email”, “re engagement”)
Risk Score
A visual risk gauge shows the job’s risk score as a percentage:
- Low risk (green) — Below the agent’s threshold; the job landed in the queue due to autonomy level settings
- High risk (red) — Above the threshold; extra caution recommended
The gauge also displays the current policy settings: risk threshold, daily email cap, and working hours.
Draft Content
The full AI-generated email appears in an editable text area. You can:
- Read the draft as-is
- Edit any part of the content directly
- Click Restore original to undo your edits
When you edit a draft, an “Edited” badge appears, and the approve button changes to Approve (edited) so you know the sent version will differ from the AI original.
Qualifying Questions
If the AI included a BANT (Budget, Authority, Need, Timeline) qualifying question in the draft, a chip appears above the draft editor highlighting the question type. You can edit or remove the question from the draft before approving.
Rating Drafts
Below the draft editor, you can rate the AI’s work on a 1—5 star scale:
| Rating | Label |
|---|---|
| 1 | Poor |
| 2 | Below average |
| 3 | Acceptable |
| 4 | Good |
| 5 | Excellent |
Ratings are optional but valuable — they feed into the agent’s confidence score and help the AI improve over time. Higher confidence scores unlock higher autonomy levels.
Approving a Draft
Click Approve & Send to approve the draft. The job moves to the Queued state and is sent by the background worker, typically within seconds.
If you edited the draft, the button reads Approve (edited) — the edited version is what gets sent.
Rejecting a Draft
- Click Reject
- Enter a rejection reason (required) — for example, “Tone is off”, “Wrong offer”, “Contact is DNC”
- Click Confirm Reject
Rejected jobs are moved to the Cancelled state. The rejection reason is logged and visible in the History tab.
Tip: Detailed rejection reasons help the AI learn what to avoid in future drafts.
Bulk Actions
Use checkboxes in the left panel to select multiple drafts at once. This is useful when you have a large queue and want to quickly approve or process several items.
History Tab
The approval queue includes a History tab that shows all previously processed jobs:
- Sent (green) — Approved and successfully delivered
- Queued (blue) — Approved and waiting to send
- Executing (purple) — Currently being sent
- Failed (red) — Sending failed after approval
- Cancelled (gray) — Rejected or halted
Click any history item to expand it and see the draft content, decision details, and timestamps.
History refreshes automatically every 15 seconds.
Who Can Approve
The approval queue requires the conversation:approve_ai permission. The following roles have this permission:
| Role | Can Approve |
|---|---|
| Owner (RevOps Admin) | Yes |
| Manager | Yes |
| SDR (Approver) | Yes |
| AE (Closer) | No |
| Viewer | No |
See Team and Roles for the full permissions breakdown.
Best Practices
- Review daily — A backlog of unapproved drafts means contacts are waiting. Aim to clear the queue at least once per day.
- Rate every draft — Even a quick star rating gives the AI signal to improve. The more you rate, the faster the confidence score climbs.
- Be specific when rejecting — “Bad” tells the AI nothing. “Too aggressive CTA for a first touch” helps it adjust.
- Edit rather than reject — If a draft is 80% right, edit the remaining 20% and approve. This gives the AI a positive signal with a correction, which is more useful than a rejection.
- Watch for patterns — If you find yourself making the same edit repeatedly, update your agent’s custom instructions to address the issue at the source.