Conversation history

The conversation history tool provides an interface for browsing and analyzing actual production conversations between your agent and end-users. This can be useful to evaluate the functional performance of your agent or to debug issues.

Flow and page analysis helps you understand traffic and escalations through the flow graph of an agent. This enables agent builders to analyze traffic flow (high and low traffic paths), discover issue hotspots (for example, to pinpoint high escalations), and drill down to the specific pages and intents causing issues. Flow and page analysis can be used along with conversation history filters, to allow for maximum flexibility in studying a relevant subset of traffic.

Limitations

Loading the Flow Analysis Table and Flow Analysis Graph views can be slow for large agents. To improve load times, use an intent filter and a one week date range.

At most 500 turns will be logged for each conversation.

Permissions required

Using this tool requires the dialogflow.conversations permission. This permission can be added to a custom role. It is also available in the Admin role, as well as other Dialogflow roles.

Enable and disable conversation history

The conversation history tool uses data from interaction logs. You can enable and disable conversation history with the Enable interaction logging agent setting.

You can also control how long data is retained by using the retention_window_days security setting.

Browsing the conversation history

To use the conversation history tool:

  1. Open the Dialogflow CX Console.
  2. Choose your project.
  3. Select your agent.
  4. Select the Manage tab.
  5. Click Conversation History.
  6. Conversations are listed in a table.
  7. Optionally apply filters using the filter control above the table.
  8. Select one of the tabs, each of which are described below.

Conversations

Conversations are shown in two views: conversation summary view and conversation details view.

Conversation summary view

The conversation summary view lists all conversations. The following metadata is provided for each conversation:

X Item
Conversation ID An identifier for the conversation
Duration Duration of the conversation
Turns The number of conversational turns
Channel Either chat (text), voice, or undetermined (no end-user input has been provided to the conversation yet)
Language The language for the conversation
Environment The agent's environment
Start time Start time for the conversation
Flags May be set to Live Agent Handoff (if conversation escalated to a human agent) or Abandoned (if end-user left the incomplete conversation) or TU,TD (if answer in the conversation got end-user rating)

You can filter the results by metadata (except environment and duration) and the following additional filter options:

X Item
Intent The provided intent was matched at some point in the conversation
No Match Some subset of conversational turns resulted in an intent no-match
Flow The provided flow is the final active flow at the end of some conversation turn
Page The provided page is the final active page at the end of some conversation turn
Is Live Agent Handoff The Live Agent Handoff flag is set
Is Abandoned The Abandoned flag is set
Conversation ID A specific conversation is chosen
Start Time A date range is provided
Agent Utterance The provided text is a substring of an utterance that was used by the agent at some point in the conversation
User Utterance The provided text is a substring of an utterance that was used by the user at some point in the conversation
Has Thumbs Up Feedback Thumbs Up feedback provided for a response in the conversation, the TU flag is set
Has Thumbs Down Feedback Thumbs Down feedback provided for a response in the conversation, the TD flag is set

Screenshot of conversation summary view

Conversation details view

You can browse the turn-by-turn view of a specific conversation in the conversation details view. Each conversational turn provides the end-user message, the agent message, and the following metadata:

X Item
Intent An intent display name that was matched or No Match
Page The final active page name
Flow The final active flow name
Parameters Parameters collected during the conversational turn

Screenshot of conversation details view

Example use case

The tool can be used to identify agent issues that result in escalations to a human agent. From the conversation summary view, you can apply the Is Live Agent Handoff filter for Yes. Examine some of these conversations and try to find common patterns. For example, you may find that in most of the conversations, there is a no-match flag for turns on a specific page. This would indicate that an intent route needs to be created, or that an existing intent route is not in scope when it needs to be.

Flow analysis table

The Flow Analysis - Table tab is used to analyze traffic and exits in flows and pages in tabular form.

The following metrics are shown:

X Item
Flow name Flow name for each flow in your agent
Relative traffic The number of conversations that went through this flow
Escalation rate The percentage of conversations that result in a request for human escalation out of all conversations that went through this flow
Exit rate The percentage of conversations that ended after this flow or transitioned to SESSION_END from this flow, out of all conversations that went through this flow

Screenshot of previously described interface

Clicking the list button opens a sample of relevant conversations for the row. You can read through the samples to better understand specific and nuanced issues. For example, you can examine why escalations are happening. To view the full details of a sample, click the Conversation details link.

Screenshot of previously described interface

Clicking Detailed Stats for a flow opens the traffic stats for flows. This shows flow transition stats from and to the selected flow.

The following metrics are shown:

X Item
Next Flow Stats for transitions from selected flow to another flow
Previous Flow Stats for transitions from another flow to the selected flow

Screenshot of previously described interface

Clicking any flow row drills down into the pages that compromise that flow.

The following metrics are shown:

X Item
Relative Traffic The number of conversations that visited this page.
Escalation rate The percentage of conversations that result in a request for human escalation out of all conversations that visited this page
Exit rate The percentage of conversations that exit the flow or exit the session (END_SESSION) directly after visiting this page, out of all conversations that visited this page
No Match rate The percentage of conversations that resulted in no-match out of all conversations that visited this page

Screenshot of previously described interface

Clicking Detailed Stats for a page opens the traffic stats for pages. This shows page transition stats from and to the selected page.

The following metrics are shown:

X Item
Intent Stats for intent matches for the page
Next Page Stats for transitions from selected page to another page
Previous Page Stats for transitions from another page to the selected page

Screenshot of previously described interface

Flow analysis graph

The Flow Analysis - Graph tab is used to analyze traffic and exits in flows and pages in graph form.

Clicking on any flow drills down into the traffic for the pages within the flow.

Screenshot of previously described interface

You can use the zoom slider to vary the level of detail that you see on the graph. Low volume paths are dropped from the visualization at lower settings. Even at higher settings, very low volume paths are always hidden.

Screenshot of previously described interface

Metrics are color-coded to easily find high severity issues.

Screenshot of previously described interface

Clicking the list button shows relevant conversations.

Screenshot of previously described interface

Clicking on any flow takes you to the pages within that flow. Here you get more overlay metrics: relative traffic, no match rate, escalation rate, and exit rate.

Screenshot of previously described interface

Access conversation history with the API

You can access conversation history with the API. See the V3beta1 reference documentation.