Conversation History

​

Accessing Conversations

1. Go to Conversations in the sidebar
2. Select an agent from the dropdown
3. Browse conversations in the list

• Title — Auto-generated from the first message
• Channel — Widget, email, or API
• Timestamp — When the conversation occurred
• Confidence — How confident the agent was

​

Conversation Details

​

Messages

• User messages (questions)
• Agent responses
• Confidence scores per response
• Feedback (thumbs up/down)
• Human responses (if escalated)

​

Form Submissions

• Which form was submitted
• The data they provided
• Submission status
• Timestamp

​

Metadata

• Conversation ID
• User ID (if identified)
• Channel source
• Start and end times

​

Filtering Conversations

​

By Feedback

​

By Channel

• Widget — Website chat widget
• Email — Email channel responses
• API — Direct API integrations

​

By Confidence

• High — 80-100% confidence
• Medium — 60-79% confidence
• Low — Below 60% confidence

​

Common Workflows

​

Weekly Review Routine

1. Filter by Thumbs Down
2. Review each negative-feedback conversation
3. Identify what went wrong:
   • Missing information → Add to knowledge base
   • Incorrect response → Update documentation
   • Poor formatting → Adjust system prompt
4. Make improvements and track changes

​

Finding Training Data

1. Filter by Low Confidence
2. Look for common question patterns
3. Add missing content to knowledge base
4. Monitor if confidence improves

​

Monitoring Channels

1. Filter by Email channel
2. Review auto-responses sent
3. Check confidence on each response
4. Adjust email threshold if needed

​

Message Feedback

​

Where Feedback Appears

• Conversations — Filter by feedback
• Analytics — Feedback trends over time
• Message level — Each message shows its feedback

​

Using Feedback Data

1. Prioritize improvements — Fix thumbs-down patterns first
2. Validate changes — Did thumbs-up increase after updates?
3. Train the team — Show examples of good/bad responses

​

Human Responses

• Messages are marked with “Human Response” badge
• Human responses are excluded from confidence calculations
• You can review how humans handled edge cases

​

Deleting Conversations

1. Open the conversation
2. Click the Delete button
3. Confirm deletion

​

When to Delete

• Test conversations
• Spam or irrelevant messages
• Privacy/compliance requests
• Duplicate conversations

​

Exporting Conversations

​

Via API

# Get all conversations for an agent
curl "https://api.ansa.so/conversations?agentId=xxx" \
  -H "Authorization: Bearer $ANSA_API_KEY"

# Get a specific conversation with messages
curl "https://api.ansa.so/conversations/{id}" \
  -H "Authorization: Bearer $ANSA_API_KEY"

​

Bulk Export

import { AnsaClient } from "@ansa/sdk";

const client = new AnsaClient({ apiKey: "xxx" });

async function exportAllConversations(agentId) {
  const conversations = [];
  let offset = 0;
  const limit = 100;

  while (true) {
    const response = await client.listConversations({
      agentId,
      limit,
      offset,
    });

    conversations.push(...response.conversations);

    if (!response.hasMore) break;
    offset += limit;
  }

  return conversations;
}

​

Understanding User Journeys

​

Common Patterns

1. Quick answers — Single question, immediate resolution
2. Multi-turn — Follow-up questions to drill down
3. Form completion — Questions leading to form submission
4. Escalation — Complex issues handed to humans

​

Identifying Issues

• Repeated questions — User didn’t get answer first time
• Abandoned conversations — User left without resolution
• Low confidence chains — Multiple uncertain responses
• Form drop-offs — Started but didn’t complete forms

​

Best Practices

​

Next Steps