AI Automation
Autonomy Levels
Configure how independently Keva operates
The autonomy level is the primary control for how Keva handles tickets. Choose from five levels, from fully manual to fully autonomous.
Autonomy Spectrum
┌─────────────────────────────────────────────────────────────────────────┐
│ │
│ HUMAN CONTROL ◄────────────────────────────────────────► AI AUTONOMY │
│ │
│ ┌───┐ ┌───┐ ┌───┐ ┌───┐ ┌───┐ │
│ │ 1 │ │ 2 │ │ 3 │ │ 4 │ │ 5 │ │
│ └───┘ └───┘ └───┘ └───┘ └───┘ │
│ Draft Suggest Semi-Auto High Full │
│ Only Autonomy Autonomy │
│ ▲ │
│ │ │
│ Recommended Start │
└─────────────────────────────────────────────────────────────────────────┘The Five Levels
Comparison Table
| Feature | L1 | L2 | L3 | L4 | L5 |
|---|---|---|---|---|---|
| AI drafts responses | ✓ | ✓ | ✓ | ✓ | ✓ |
| Auto-send routine | ✗ | ✗ | ✓ | ✓ | ✓ |
| Auto-send complex | ✗ | ✗ | ✗ | ✓ | ✓ |
| Auto platform actions | ✗ | ✗ | ✗ | ✗ | ✓ |
| Sensitive still approved | - | - | ✓ | ✓ | ✓* |
*Configurable via Trust Controls
Setting Your Autonomy Level
Choosing the Right Level
Start with Level 2-3 if:
- You're new to AI-powered support
- You have strict quality requirements
- Your tickets are highly varied
- You're in a regulated industry
Consider Level 4-5 if:
- You've validated AI quality at lower levels
- Most tickets are routine
- You have high ticket volume
- Speed is critical
What Changes at Each Level
Level 1 → 2
- Same behavior, but now clearly framed as "suggestions"
- Approval workflow becomes prominent
Level 2 → 3
- Simple, high-confidence responses auto-send
- Complex tickets still need approval
- You'll see "Auto-sent" in activity logs
Level 3 → 4
- Most responses auto-send
- Only escalated tickets need review
- Approval queue becomes much smaller
Level 4 → 5
- Platform actions execute automatically
- Only Trust Control triggers require approval
- Essentially "lights out" operation
Level 3 Detail: Semi-Autonomous
Recommended Starting Point
Level 3 is the recommended starting point for most teams. It provides meaningful automation while maintaining oversight for complex situations.
What Auto-Sends
Responses that are:
- High confidence (90%+)
- Clear question/answer format
- Match knowledge base content
- From non-VIP customers
- Not flagged by any rule
What Needs Approval
Responses that are:
- Medium/low confidence
- Complex multi-part questions
- From VIP customers
- Contain sensitive topics
- First contact from customer
What Escalates
Situations that:
- AI can't determine response
- Customer requests human
- Sensitive keywords detected
- Multiple failed attempts
Monitoring by Level
| Level | Review Approach |
|---|---|
| Level 1-2 | Review every draft and approval |
| Level 3 | Spot-check auto-sent, review all approvals |
| Level 4-5 | Monitor escalations and sensitive action approvals |
Adjusting Over Time
Signs to Increase Level
- High approval rate (>95%)
- Few edits to AI responses
- Team spending too much time approving
Signs to Decrease Level
- Customer complaints about responses
- Frequent incorrect actions
- Important nuances being missed
FAQ
Can I change levels mid-ticket?
Yes, but the change only affects new actions. In-progress approvals remain.
Does level affect all ticket types?
Yes, but Trust Controls can override for specific action types.
What happens to pending approvals when I increase level?
They remain in the queue. They won't auto-execute.
Can different team members have different levels?
Currently, autonomy level is organization-wide. Use roles to control who can approve.