Human Escalation
Not every question should be answered by AI. Some questions are too nuanced, too specific, or too important to handle automatically. Keep'em detects these situations and routes them to your team.
When Escalation Happens
The system triggers human escalation in several scenarios:
Low AI confidence — When the AI searches your transcript and documents but doesn't find sufficiently relevant content to give a confident answer.
Viewer request — When a viewer explicitly asks to speak with a person (e.g., "Can I talk to someone?" or "I need a human").
Repeated similar questions — When a viewer asks variations of the same question multiple times, suggesting the AI's responses aren't satisfying their need.
What Your Team Receives
When a question is escalated, your team gets a notification with full context:
- The viewer's name and email
- Which video they were watching
- Their exact timestamp in the video
- Their question
- What the AI attempted (if anything)
- The full chat history from that session
This context means your team can respond effectively without asking the viewer to repeat themselves.
Notification Channels
Configure where escalation notifications go from your project settings:
| Plan | Available Channels |
|---|---|
| Starter | |
| Growth | Email, Slack, Webhook |
| Scale | Email, Slack, Webhook, Custom Routing |
Slack Integration
On Growth and above, escalated questions post to a designated Slack channel. Your team responds directly in the Slack thread. The viewer sees the response in their chat — seamless, no context switching for the viewer.
The Viewer's Experience
From the viewer's perspective, the transition is smooth. The chat shows a message like "I've flagged this for the team — they'll respond shortly." When a human responds, the message appears in the same conversation. There's no separate support ticket, no email chain, no second interface.