Overview
Ansa is ideal for customer support teams who want to:
- Deflect repetitive questions with instant, accurate answers
- Escalate complex issues to human agents seamlessly
- Build a growing knowledge base from real support interactions
- Provide 24/7 support without scaling headcount
Recommended Configuration
1. Knowledge Base Setup
Index your existing help documentation to give your agent comprehensive knowledge:
Recommended sources:
- Help center articles (web crawl or upload)
- FAQ documents
- Product documentation
- Troubleshooting guides
Use the auto-detect feature during onboarding to automatically discover and index your help pages. Ansa will probe common paths like /help, /support, /faq, and /docs.
2. Agent Settings
| Setting | Recommended Value | Why |
|---|
| Model | Claude 3.5 Sonnet or GPT-4o | Best balance of accuracy and speed |
| Temperature | 0.3 - 0.5 | Lower temperature for more consistent, factual responses |
| System Prompt | Include tone, escalation rules, and boundaries | See example below |
Example System Prompt:
You are a helpful customer support agent for [Company Name].
Your role:
- Answer questions accurately based on the knowledge base
- Be friendly, professional, and concise
- If you don't know the answer, say so honestly
- For billing issues, account access, or complex technical problems, offer to connect the user with a human agent
Never:
- Make up information not in the knowledge base
- Share internal processes or confidential information
- Make promises about refunds or credits without escalating
3. Escalation with Confidence Thresholds
Use confidence scores to automatically escalate low-confidence responses:
| Confidence Level | Action |
|---|
| High (80%+) | Respond automatically |
| Medium (50–79%) | Respond but offer human escalation |
| Low (below 50%) | Show contact form or escalate immediately |
Create a form tool for human escalation:
{
"name": "contact_support",
"description": "Use this when the user needs human assistance, has a billing issue, or when you cannot answer their question confidently.",
"executionType": "form",
"formSchema": {
"fields": [
{ "name": "email", "label": "Email", "type": "email", "validation": { "required": { "value": true, "message": "Email is required" } } },
{ "name": "issue", "label": "Describe your issue", "type": "textarea" }
],
"submitLabel": "Contact Support",
"successMessage": "We've received your request. A support agent will reach out within 24 hours."
},
"formPostActions": [
{ "type": "email", "recipientIds": ["support-team-user-id"], "subject": "Support Escalation" },
{ "type": "slack", "webhookUrl": "https://hooks.slack.com/...", "messageTemplate": "🚨 New escalation from {{email}}: {{issue}}" }
]
}
Set up proactive engagement to help users before they get frustrated:
| Trigger | When | Action |
|---|
| Time on support page | 30 seconds on /help/* | Open chat with “Need help finding something?” |
| Scroll depth | 75% scroll on troubleshooting article | Show bubble “Still having issues?” |
| Exit intent | About to leave support page | Show form for feedback |
Email Channel for Async Support
Enable the email channel to handle support emails with AI:
- Forward support emails to your agent’s inbound address
- Set confidence threshold (recommended: 70%)
- Configure low-confidence action to “forward” to human reviewers
- Enable Auto Learn to automatically create Q&A entries from human responses
Monitoring & Improvement
Review Conversation History
Regularly review conversations to identify:
- Common questions that need better documentation
- Low-confidence responses that need knowledge base updates
- Patterns in escalated issues
Track Key Metrics
Monitor these analytics:
- Average confidence score — Target 75%+
- Escalation rate — Track form submissions vs. resolved chats
- Response quality — Review feedback ratings
See Also