Governance Policies

Governance policies control what AI can and cannot do. Set rules to block sensitive actions, require approvals, or allow routine operations automatically.

Accessing Governance Policies

1. Go to Automation in the sidebar
2. Find the Governance Policies card
3. Click Manage Policies

Overview

The Policies page shows:

Summary Bar

Creating a Policy

1. Click Create Policy button
2. Fill in the policy dialog:

Policy Name

Give your policy a descriptive name (e.g., "Block Refund Discussions", "Require Approval for Discounts").

Description

Explain what the policy does and why it exists.

Conditions

Define when this policy applies:

1. Click Add Condition
2. Select a field:
   • signals.confidence - AI confidence score
   • context.action_type - Type of action being taken
   • context.customer_tier - Customer segment
   • context.topic - Conversation topic
3. Select an operator:
   • For numbers: <, >, <=, >=, =, !=
   • For strings: equals, contains, starts with, ends with
   • For existence: exists, does not exist
4. Enter the value to compare

Add multiple conditions for complex rules. All conditions must match for the policy to trigger.

Outcome

Choose what happens when conditions are met:

Priority

Set priority using the slider (1-100):

• Higher priority: Evaluated first
• Lower priority: Evaluated after higher ones
• When policies conflict, higher priority wins

1. Click Create Policy or Save Changes

Policy Cards

Each policy shows:

• Outcome Icon: Color-coded (green/red/yellow)
• Policy Name: Title you assigned
• Status Toggle: Active or Disabled
• Priority Badge: Priority number
• Conditions Preview: Summary of conditions

Managing Policies

Enable/Disable

Toggle the switch on any policy card to activate or deactivate it.

Edit Policy

1. Click Edit on the policy card
2. Modify settings in the dialog
3. Click Save Changes

Delete Policy

1. Click the trash icon on the policy card
2. Confirm deletion

Example Policies

Block Sensitive Topics

• Name: Block Legal Discussions
• Condition: context.topic contains legal
• Outcome: Block
• Priority: 90

Require Approval for Refunds

• Name: Refund Approval Required
• Condition: context.action_type equals refund
• Outcome: Require Approval
• Priority: 85

Allow Routine Queries

• Name: Allow FAQ Responses
• Condition: signals.confidence > 0.9
• Outcome: Allow
• Priority: 50

Best Practices

1. Block first: Set blocking policies at high priority
2. Be specific: Use multiple conditions to avoid false matches
3. Document policies: Use descriptions to explain reasoning
4. Review regularly: Check blocked actions for false positives
5. Start conservative: Block more initially, then loosen as you learn