Approvals

The approval system gives you control over AI actions. At medium autonomy levels, AI drafts actions for your review instead of executing immediately.

How Approvals Work

1. AI generates action - Response, refund, ticket update, etc.
2. Action queued - Appears in Approvals tab
3. You review - Read the draft, check reasoning
4. Approve or reject - Action executes or is discarded

When Approvals Are Required

Based on autonomy level and action type:

Autonomy Level	Responses	Simple Actions	Sensitive Actions
1 (Draft Only)	Always	Always	Always
2 (Suggest)	Always	Always	Always
3 (Semi-Auto)	Sometimes*	Always	Always
4 (High Auto)	No	No	Always
5 (Full Auto)	No	No	Configurable

*Routine responses may auto-send; complex ones need approval

The Approval Queue

Accessing Approvals

1. Click Approvals in the sidebar
2. Or use keyboard shortcut g a

Queue Overview

See all pending approvals:

• Action type
• Related ticket
• AI confidence
• Time waiting
• Preview

Filtering

Filter by:

• Action type (response, refund, etc.)
• Ticket priority
• Customer segment
• Age of request

Reviewing an Approval

1. Read the Context

• Original ticket/conversation
• Customer history
• Previous attempts

2. Check AI Reasoning

Click "Why?" to see:

• What triggered this action
• Information AI used
• Confidence level
• Alternatives considered

3. Review the Draft

Read the proposed:

• Response text
• Action details
• Affected data

4. Make a Decision

• Approve - Execute as-is
• Edit & Approve - Modify then execute
• Reject - Discard, optionally provide feedback

Approval Actions

Approve

Click Approve to execute the action immediately.

Edit & Approve

1. Click Edit
2. Modify the draft
3. Click Approve
4. Modified version executes

Reject

1. Click Reject
2. Optionally add feedback
3. Action is discarded
4. Ticket returns to AI for new approach

Bulk Approval

For multiple similar approvals:

1. Select checkboxes
2. Click Approve Selected
3. Review count
4. Confirm

The Why Panel

Every AI action includes reasoning:

What's Shown

• Intent - What the AI understood
• Sources - Information used (KB, history, etc.)
• Confidence - How sure the AI is
• Alternatives - Other approaches considered

Using Why Panel

• Understand AI decision-making
• Identify training opportunities
• Build trust in AI capabilities
• Catch potential errors

Notification Settings

Configure approval notifications:

1. Go to Settings → Notifications
2. Find Approval Alerts
3. Enable:
   • In-app badges
   • Email digest
   • Slack alerts (if connected)

Urgency Levels

• Immediate - High-priority tickets
• Digest - Batched notifications
• Dashboard only - No active alerts

Best Practices

Review Regularly

Check approvals frequently to avoid delays:

• Morning review
• After lunch review
• End of day review

Provide Feedback

When rejecting, explain why:

• Helps AI improve
• Creates training data
• Documents decisions

Trust but Verify

Even at higher autonomy:

• Spot-check approved actions
• Review the activity log
• Monitor customer responses

Adjust Thresholds

If too many approvals:

• Increase autonomy level
• Adjust confidence threshold
• Add knowledge base content

If too few (missing things):

• Decrease autonomy level
• Add sensitive action rules
• Review trust controls

Metrics

Track approval metrics in Analytics:

• Approval rate
• Edit rate
• Rejection rate
• Time to approval
• Approvals per agent