Escalations allow your AI agent to seamlessly hand off conversations to human support when needed. This ensures customers always have a path to human assistance while maximizing the efficiency of your AI agent.
How Escalations Work When a conversation is escalated:
Ticket created — An escalation ticket is logged in your dashboardTeam notified — Configured notification channels are triggered (email, Slack, webhook)Customer informed — The user sees a confirmation messageAgent paused — AI responses are held until a human responds or the escalation is resolved
Escalation Triggers Conversations can be escalated in several ways:
User-Initiated Escalation Users can request human support by:
Clicking the escalation button (if enabled in widget settings)
Typing phrases like “talk to a human” or “speak to support”
Submitting a form that triggers escalation
AI-Initiated Escalation The AI can automatically escalate when:
Low confidence — Response confidence falls below your thresholdTool failure — A critical tool fails repeatedlySpecific topics — Keywords or intents trigger escalation (configured in system prompt)
Programmatic Escalation Escalate via API or webhook when external conditions are met:
curl -X POST "https://api.ansa.so/conversations/{conversationId}/escalate" \ -H "Authorization: Bearer $ANSA_API_KEY " \ -H "Content-Type: application/json" \ -d '{ "reason": "VIP customer flagged for priority support" }'
Configuring Escalation Settings Navigate to Settings → Escalations in your agent dashboard.
Auto-Escalation on Low Confidence Setting Description Enable Auto-Escalate Automatically create tickets when confidence is low Confidence Threshold Minimum confidence score (0-100%) before escalation Delay Wait time before escalating (allows user to clarify)
Set the confidence threshold to 40-50% to catch genuinely uncertain responses without over-escalating.
SLA Configuration Define response time expectations:
Setting Description Target Response Time Expected time for first human response Escalation Priority Default priority level (low, medium, high, urgent) Business Hours When human support is available
Notification Channels Configure how your team is notified:
Email Notifications
Recipients — Email addresses to notifyInclude Transcript — Attach conversation historyFrequency — Immediate or batched (hourly digest)
Slack Notifications
Connect your Slack workspace in Integrations → Slack
Select the channel for escalation alerts
Configure message format and mentions
Webhook Notifications Send escalation data to your own systems:
{ "type" : "escalation.created" , "data" : { "escalationId" : "esc_abc123" , "conversationId" : "conv_xyz789" , "reason" : "low_confidence" , "confidence" : 35 , "priority" : "medium" , "customerEmail" : "jane@example.com" , "transcript" : [ { "role" : "user" , "content" : "..." }, { "role" : "assistant" , "content" : "..." } ] } }
Control the “Talk to Human” button visibility:
Option Description Always Show Button visible on every message After Threshold Show after N messages or low confidence Never Hide button (escalation still possible via phrases)
Customize the button text:
Default: “Talk to a human”
Examples: “Contact Support”, “Get Help”, “Speak to Agent”
Confirmation Message The message shown after escalation:
Your request has been escalated to our support team. We typically respond within 2 hours during business hours.
Managing Escalations Escalation Dashboard View all escalations at Conversations → Escalations :
Column Description Status Open, In Progress, Resolved Priority Urgent, High, Medium, Low Assigned To Team member handling the ticket Created When the escalation was triggered SLA Status On track, At risk, Breached
Responding to Escalations
Click an escalation to view the full conversation
Review the AI’s responses and confidence scores
Type your response in the reply box
Click Send — the customer receives your message in the widget
Resolving Escalations When the issue is resolved:
Click Resolve on the escalation
Optionally add a resolution note
The AI resumes handling the conversation
Re-enabling AI After resolution, the AI automatically resumes. To keep AI paused:
Toggle Keep AI Paused before resolving
Manually re-enable via the conversation menu
Best Practices Tune Confidence Thresholds
Start with a 50% threshold
Review escalated conversations weekly
Adjust based on false positive rate
Set Realistic SLAs
Consider your team’s capacity
Account for timezone coverage
Build in buffer for complex issues
Train Your AI Reduce escalations by:
Adding common questions to your knowledge base
Improving system prompt for edge cases
Creating tools for frequent actions
Monitor Escalation Metrics Track in Analytics → Escalations :
Escalation Rate — % of conversations escalatedResolution Time — Average time to resolveSLA Compliance — % meeting response targetsTop Reasons — Why escalations happen
API Reference Create Escalation POST /conversations/{conversationId}/escalate
Request Body: Field Type Description reasonstring Reason for escalation prioritystring low, medium, high, urgentassignTostring Team member email (optional)
Update Escalation PATCH /escalations/{escalationId}
Request Body: Field Type Description statusstring open, in_progress, resolvedprioritystring Updated priority assignTostring Reassign to team member notestring Internal note
List Escalations GET /escalations?agentId={agentId} & status = open
Next Steps
Confidence Scores Understand how confidence affects escalations
Slack Integration Set up Slack notifications
