Conversation transcripts
Review past conversation history between users and your AI applications
Reviewing conversation history
Conversation history provides historical information about sessions that have taken place with a deployed agentic CX designer application.
Use these historical transcripts to review what happened during real user interactions, troubleshoot unexpected behavior, inspect transcripts, and identify opportunities to improve flows, routing, prompts, integrations, or escalation paths.
What is conversation history?
Conversation history gives you a session-level view of user interactions with your deployed application.
Each conversation record can help you understand:
What the user asked
How long the session lasted and latency of the system
Whether the user engaged or abandoned the experience
Whether analytics tags were reached
What the full transcript looked like
Use conversation history when you need to move from high-level performance data into the details of a specific user session.
How to access conversation history
To access conversation history for an application:
Open Applications.
Select a deployed application.
Open the Observe tab.
Select History.
Review the Transcripts table.
The Observe tab appears after an application has been deployed for the first time.
Conversation filters
Use filters to narrow conversation history to the sessions you want to inspect.
Start / End time
Filter conversations within a specific date and time range.
User engagement
Filter for conversations where the user did or did not respond.
Tags
Review conversations where selected analytics tags were reached.
Search
Find conversations containing specific user utterances or keywords.
Filters are useful when investigating a known issue, reviewing behavior from a specific release window, or finding examples of repeated user friction.
Viewing conversations
Each row in the conversation history table represents a unique conversation session.
Select a conversation row to view more details.
First message
The first user utterance in the conversation. If a user started a session and did not engage, this appears as N/A.
Time
When the conversation session started.
Duration
How long the conversation lasted.
Scores
Confidence or AI-related scoring details available for the session.
Average response time
How long the application took to respond on average.
Flows
Flows invoked during the conversation.
Tags
Analytics tags reached during the session, when available.
Transcript
The full exchange between the user and the application.
Conversation details
When you open a conversation, you can review the full transcript and supporting session details.
Conversation details include:
Conversation ID (select the Information icon)
User ID (select the Information icon)
Full message transcript
User messages
Application responses
Debugger/event log (select a message in the transcript to review all events in the selected turn)
Copy options are available for conversation and user IDs. These IDs can be useful when investigating a specific session or reviewing that same path in in-Canvas analytics.
Using conversation history with In-Canvas analytics
Conversation history helps explain what happened in a transcript. In-Canvas analytics helps you see where that conversation traveled inside a flow.
A common troubleshooting workflow is:
Find the relevant conversation in Observe > History.
Open the conversation and review the transcript.
Select the information icon > Copy the Conversation ID.
Open the flow involved in the conversation.
Use In-Canvas analytics to filter by that Conversation ID.
Review the path the user took through the flow.
This is useful when a conversation shows unexpected fallback behavior, missed routing, repeated questions, drop-off, or escalation.
Common review scenarios
A user says the application did not help
Review the transcript to see what the user asked and how the application responded.
Escalations increased
Filter by escalation-related flows or tags and inspect the conversations leading up to escalation.
Users are dropping off
Review transcripts near drop-off points and compare with In-Canvas analytics.
A flow is being invoked unexpectedly
Search for user utterances and review whether routing matched the intended flow.
A Data request appears slow or unsuccessful
Review affected conversations and compare timing or failure behavior.
A new user need is emerging
Search repeated phrases and identify whether a new flow, prompt, or knowledge base content is needed.
Last updated