AI Automation
Configure how Keva's AI handles your support tickets
Keva's AI is highly configurable. You control how autonomous it is, what actions require approval, and how it learns from your preferences.
Core Concepts
Autonomy Level
How independently Keva operates, from fully manual to fully autonomous. The autonomy slider is the primary control for AI behavior.
Trust Controls
Additional safeguards that require confirmation for sensitive actions, regardless of autonomy level.
Action Limits
Daily caps on actions Keva can take, preventing runaway automation.
Key Features
Autonomy Levels
Control AI independence from draft-only to fully autonomous
Trust Controls
Require confirmation for sensitive actions
Action Limits
Set daily caps on AI actions
Client Brain
How AI learns from your tickets
How AI Processes Tickets
When a ticket arrives, Keva:
- Analyzes - Understands the request, intent, and sentiment
- Researches - Checks knowledge base, customer history, connected platforms
- Decides - Determines the best response or action
- Acts - Drafts, queues for approval, or executes based on autonomy level
Quick Configuration
Recommended Starting Point
| Setting | Value | Why |
|---|---|---|
| AI Autonomy Level | Medium Auto (Level 3) | Handles routine tickets, escalates complex |
| Ask customer before executing actions | On | Makes customer confirmation available for actions |
| Spending & Action Limits | Defaults | Prevents runaway automation |
As You Build Confidence
| Week | Autonomy | Notes |
|---|---|---|
| 1-2 | Low Auto → Medium Auto (Levels 2-3) | Review most actions |
| 3-4 | Medium Auto → High Auto (Levels 3-4) | Trust routine handling |
| 5+ | High Auto → Full Auto (Levels 4-5) | Minimal human intervention |
AI Settings Location
Open Settings → Copilot in your dashboard. The Copilot tab shows:
- AI Autonomy Level - Main slider for reply and action behavior
- Ask customer before executing actions - Makes customer confirmation available for eligible actions
- Spending & Action Limits - Daily and refund caps
- Brand Voice - Tone and style guidance for AI-generated replies
Monitoring AI Performance
Track AI effectiveness in Analytics:
- Resolution rate - % of tickets resolved by AI
- First response time - Speed of AI replies
- Escalation rate - How often humans needed
- Customer satisfaction - CSAT for AI-handled tickets
- Accuracy - How often AI responses are edited
Improving AI Over Time
Add Knowledge Base Content
The more information Keva has, the better responses:
- FAQs
- Product documentation
- Policy documents
- Common scenarios
Review AI Reasoning
Check the Why Panel to understand decisions:
- Correct good behavior
- Catch mistakes early
- Identify training gaps
Provide Feedback
When editing AI responses:
- Explain what was wrong
- Show the correct approach
- AI learns from edits
Adjust Gradually
Increase autonomy incrementally:
- Start conservative
- Verify quality at each level
- Step back if issues arise