AirmailerairmailerDocs

OverviewApproval QueueManaging ThreadsEmail Statuses

Sign In
Airmailerairmailer
  1. Docs
  2. Conversations
  3. Approval Queue

Approval Queue

Review and approve AI-generated email responses before they're sent to customers.

What Is the Approval Queue?

The Approval Queue provides human oversight for AI-generated responses. Instead of automatically sending every response, you can review drafts, make edits, and approve them before delivery.

This Human-in-the-Loop approach gives you:

  • Quality control before emails reach customers
  • Confidence-based routing to focus review on uncertain responses
  • Full editorial control with inline editing capabilities

Accessing the Approval Queue

Navigate to Conversations in the sidebar, then select Approval Queue. A badge displays the number of pending drafts awaiting review.

Approval Queue Interface

Queue Table

The approval queue displays all AI-generated drafts:

| Column | Description | |--------|-------------| | Created | When the draft was generated | | Customer | Customer name and email address | | Subject | Email subject line | | Preview | First 100 characters of the draft | | Confidence | AI confidence score for the response | | Status | Current state: pending, approved, rejected, or edited | | Actions | Quick approve, edit, or reject buttons |

Filtering

Filter the queue by status:

  • Pending Review (default) - Drafts awaiting your decision
  • All Drafts - Complete history
  • Approved - Sent responses
  • Rejected - Discarded drafts
  • Edited - Modified before sending

Reviewing a Draft

Click any row to open the Draft Review Sheet.

Draft Review Sheet

The review interface shows:

  • Original customer message - What the customer wrote
  • AI-generated response - The draft ready for review
  • Confidence score - Visual indicator of AI certainty
  • Edit capability - Modify the response before sending

Actions

| Action | Result | |--------|--------| | Approve & Send | Sends the email immediately as-is | | Approve with Edits | Sends your modified version | | Reject | Discards the draft; marks conversation as skipped |

AI Confidence Scores

The AI assigns a confidence score to each response based on how well the answer matches your knowledge base.

| Score | Indicator | Meaning | |-------|-----------|---------| | 85%+ | Green | High confidence - information clearly available | | 70-84% | Yellow | Moderate confidence - review recommended | | Below 70% | Red | Low confidence - careful review needed |

Low-confidence drafts often indicate:

  • Questions not covered in your documents
  • Ambiguous customer requests
  • Complex topics requiring nuance

Approval Modes

Configure how your brand handles AI responses in Agent Settings under "Response approval":

Auto Send

All AI responses are sent immediately without review. Best for well-documented, straightforward use cases.

Draft All

Every AI response requires approval before sending. Maximum control for sensitive communications.

Confidence Threshold

Only responses below your configured threshold require approval. Balances efficiency with oversight.

Example: Set threshold to 80% - responses with 80%+ confidence send automatically, while lower-confidence drafts queue for review.

Configuring the Threshold

The confidence threshold slider ranges from 50% to 100%:

| Range | Mode | Behavior | |-------|------|----------| | 90-100% | Strict | Most AI responses queued for review | | 70-89% | Balanced | Medium-confidence responses reviewed | | 50-69% | Permissive | Only low-confidence responses reviewed |

Lower thresholds mean more automation; higher thresholds mean more human oversight.

Per-Conversation Overrides

You can override the brand default for specific conversations:

  • Flag sensitive customers for mandatory review
  • Allow auto-send for trusted, routine threads
  • Adjust based on topic complexity

Set overrides from the conversation detail view.

Best Practices

Review Regularly

  • Check the approval queue at least daily
  • Set up notifications for pending drafts
  • Don't let drafts age - customers expect timely responses

Prioritize Low Confidence

  • Review red (below 70%) drafts first
  • These are most likely to need edits
  • Yellow drafts often just need a quick check

Edit for Accuracy

  • Fix any factual errors before approving
  • Adjust tone if needed
  • Add details the AI may have missed

Use Patterns to Improve

  • Notice what the AI gets wrong repeatedly
  • Update your documents to fill gaps
  • Add rules to prevent recurring issues

Know When to Reject

  • If the response is fundamentally wrong, reject it
  • Consider responding manually for complex cases
  • Use rejections to identify training opportunities

Escalation Notifications

Get notified when drafts need review so you can respond promptly.

Notification Email

Enter an email address to receive notifications when drafts are pending approval. This is useful for:

  • Alerting on-call support staff
  • Sending to a shared team inbox
  • Triggering email-based workflows

Configure in Agent Settings > Escalation settings > Notification email.

Slack Notifications

Connect a Slack webhook to receive notifications in your team's channel.

Setting Up Slack Webhooks

  1. Go to api.slack.com/apps and click Create New App
  2. Choose From scratch, name your app, and select your workspace
  3. In the app settings, navigate to Features > Incoming Webhooks
  4. Toggle Activate Incoming Webhooks to On
  5. Click Add New Webhook to Workspace and select your channel
  6. Copy the webhook URL (starts with https://hooks.slack.com/)
  7. Paste in Agent Settings > Escalation settings > Slack webhook URL

Notifications include:

  • Customer name and email
  • Subject line of the pending draft
  • Link to the approval queue

When Notifications Trigger

Notifications are sent when:

  • A new draft enters the approval queue
  • The draft's confidence score falls below your threshold (if using Confidence Threshold mode)
  • All drafts (if using "Draft All" mode)

No notifications are sent in "Auto Send" mode since responses don't queue for review.

Workflow Example

Customer asks about warranty extension
        ↓
AI generates response (72% confidence)
        ↓
Draft appears in Approval Queue
        ↓
You review: answer is mostly correct
        ↓
You edit: add specific warranty terms
        ↓
Approve with Edits → Email sent
        ↓
Customer receives accurate response

Next Steps

  • Managing Threads - Work with conversation history
  • Email Statuses - Understand conversation states
  • AI Agent Overview - Learn how responses are generated
PreviousOverviewNextManaging Threads

Command Palette

Search for a command to run...