Conversation Management

Review, evaluate, and manage individual conversations with admin tools

Open a conversation to read its messages and review its quality, assignment, tags and feedback.

Accessing Conversation Management

Staff members and viewers can open the administration panel; available actions depend on their permissions.

1. Navigate to Chat History in the sidebar.
2. Click a conversation to open it.
3. Click the Open administration icon at the right of the conversation header.

The panel sits beside the conversation when there is enough space. On smaller screens, it replaces the messages within the page. The same header control becomes Close administration and returns you to the conversation. The layout adapts when you resize the window or open the AGO Assistant.

Customer information, creation date, the linked ticket and translation controls appear at the top of the panel. The three-dot Actions menu in the conversation header groups evaluation, topic analysis, debugging and conversation access controls, according to your permissions and enabled features.

Below the conversation details, the panel shows admin rating and comments, co-pilot evaluation when enabled, topics, admin tags, user feedback, and reviewer and status assignment. When available, Sharing & copies appears above these controls. Your unsaved admin comment and expanded evaluation stay in place when you close and reopen the panel or resize the window. Click Save to store a comment before moving to another conversation.

Topics and Admin Tags show their current labels as compact chips. Use the small + beside a section to add a label. On desktop, the remove button appears over the end of a chip on hover or keyboard focus, without widening it. Add, remove and comment-edit controls remain visible on touch screens. Comments show a short preview until you open the editor. Empty User Feedback stays collapsed until you open it.

When the Thread Evaluator is enabled, the Co-pilot evaluation row shows the available CX Score as a colored badge and the review status. Expand it to see the score breakdown and improvement suggestions. Dates follow your interface language; hover over a date for the full timestamp and time zone.

When you open a conversation from a list, use the previous and next arrows in the conversation header to move through that list. The back arrow returns to the list.

Pasted code

When a user pastes a large block of code into a message (2,000 characters or more), the block is moved into an attachment and the message keeps the user’s own text, with a reference such as [pasted-html-1.html] where the code was. This covers the HTML of an email or a form, an email’s original source with its headers, an API response in JSON and other code or logs.

The agent reads the attachment with the message: the whole block up to 20,000 characters, otherwise its first 20,000 characters plus, for HTML, up to 10,000 characters of its visible text. Your knowledge is searched with the user’s words only, so a pasted newsletter or log no longer pulls unrelated articles into the answer. Open the attachment to see everything the user pasted.

View a linked ticket

If a ticket was created from the conversation, click View ticket at the top of the administration panel to view its details. The ticket subject appears below the link when available. If an external ticket link is available, you can open it from the ticket details page.

Create a task

When a conversation shows work your team must do afterwards, such as an article to fix or a bug to report, click Create task at the top of the administration panel to add it to your team’s Tasks board. To point at one answer in particular, use Create task in that message’s actions instead: the task then links to the message as well. The task opens with a generic title and the conversation linked as evidence; no customer text is copied into it. The form warns you when tasks already exist for this conversation. Available to staff users when Tasks is enabled for your workspace.

Translate a conversation

Use the language selector at the top of the administration panel to read a conversation in French, English, Italian, Spanish, Portuguese, or German. Each language has a flag to help you find it quickly.

The messages show an automatic translation once it is ready. Select Original text to return to the conversation’s original wording at any time. Translation only changes your reading view; saved messages and the conversation seen by the user stay unchanged. Opening another conversation restores the original text.

AGO keeps each translated message, so reopening a conversation in a language you already chose shows the translation right away. Only new or edited messages are translated again.

If translation fails, the original text remains visible. Click Retry to try again: messages already translated are kept, and only the remaining ones are translated.

Disable or reactivate a conversation

Staff members can click Disable conversation in the Actions menu to make its content inaccessible to users in the portal and embedded chat, including through existing shared links. Users see a message explaining that administrators disabled access. The conversation remains in their history with a “Conversation disabled” label. They cannot continue or create a new copy of the disabled conversation.

Administrators retain access to review the content. Click Reactivate conversation to restore access. Disabling does not delete the conversation or change its evaluation. Copies created before disabling remain separate conversations and can be disabled individually.

Manual Evaluation

Assess conversation quality using the three labeled rating buttons:

Button	Color	Use When
Good	Green	The agent handled the conversation well
Mixed	Yellow	Average handling, neither good nor problematic
Poor	Red	The conversation had issues or was handled poorly

Click a button to set the evaluation. Click again to clear it (returns to “Not Evaluated”).

Evaluation Comments

Add context to your evaluation:

1. Click the Admin comment row or its preview
2. Enter your comment explaining the evaluation
3. Click Save to store the comment, or Cancel to discard your edits

Comments help other reviewers understand why a conversation was rated a certain way and are useful for training and quality improvement.

LLM-Based Evaluation

When the Thread Evaluator is enabled, you can trigger automated evaluation:

Running an Evaluation

Clear any existing manual rating by clicking the selected rating button again. Then, open Actions → Run Evaluation to request an AI assessment. Conversations that have already been evaluated are skipped.

CX Score Display

When a CX score is generated, it appears in the Co-pilot evaluation summary. Expand the row to see:

• CX Score Gauge: A 0-100 score representing overall quality
• Dimension Breakdown: Visual bars showing scores for each dimension:
  • Resolution Quality
  • Response Coherence
  • Tone
  • Efficiency
  • Customer Effort

Some conversations are marked as evaluated without scores, including conversations without a meaningful user/assistant exchange and short exchanges requesting human support.

Review Flags

The LLM evaluation may flag conversations for human review:

Badge	Meaning
Review (Yellow)	The LLM recommends human verification
Validated (Green)	No review flag was set; a CX score may be unavailable

Improvement Suggestions

If the LLM identifies issues, it provides specific suggestions for how the agent could have responded better. Use these insights to improve agent instructions or knowledge base content.

LLM evaluation is only available when the Thread Evaluator feature is enabled in your platform configuration.

Reviewer Assignment

Assign conversations to team members for review or follow-up:

1. Click the current value beside Reviewer
2. Select a configured reviewer from the list
3. The assignment is saved automatically

Use reviewer assignments to:

• Distribute quality review workload
• Track who reviewed which conversations
• Filter conversations by reviewer in the list view

Admin Status

Click the current value beside Status to set the conversation’s status.

Apply custom status labels to track conversation states:

1. Click the current value beside Status
2. Select a status from your configured options
3. The status badge appears with its assigned color

Example statuses:

• “Under Investigation”
• “Escalated to Engineering”
• “Needs Follow-up”
• “Resolved”

Configure admin statuses on the Conversation Labels page to match your workflow.

Tags

Conversation Tags

Add descriptive tags to categorize conversations:

1. Click the + beside Topics
2. Search the existing topics
3. Select a topic to add it
4. Click the X on any tag to remove it

When topic analysis is enabled, choose Actions → Run Topic Analysis to classify the conversation automatically.

Tags help you:

• Categorize conversations by topic
• Track issue trends
• Filter conversations in list view

Admin Tags

Staff members can use admin tags for internal categorization:

1. Click the + beside Admin Tags
2. Search and select tags
3. Admin tags display with color coding

Use admin tags for internal categorization like:

• “Training Example”
• “Bug Report”
• “Feature Request”
• “VIP Customer”

User Feedback Section

If users submitted feedback about the conversation, it appears under User Feedback.

Feedback Display

For each feedback entry, you’ll see:

• Issue Type Badges: Inaccurate, Incomplete, Information Not Found, Technical Issue
• Additional Feedback: Free-text comments from the user

The section also shows thumbs up/down counts for the conversation.

Managing Feedback

Click on a feedback entry to edit its Internal Note, then click Save.

Tools Called

In each agent answer, just above the reply text, a Tools called bar lists every tool the agent ran to produce it, with the tool names shown next to the count. Open it to see each call with its status and, when it was measured, how long the call took. Expand a call to read the arguments it received and the result it returned.

The bar only appears for staff members, and only on answers where at least one tool ran. It lists every call, including the ones the conversation itself draws nothing for — so an answer that looks like it did no work no longer hides the tools behind it.

Superadmins can open Actions → Full Thread Debug View, with the prompts, token usage and timing breakdown behind each answer.

Conversation Details

The top of the administration panel shows the customer, creation date, linked ticket and translation controls.

Where the conversation came from, and where it went

Expand Sharing & copies to trace a conversation’s lineage, which is what explains a transcript containing messages its owner never wrote:

• Copied from: the original conversation, linked at the branch message when that message is available
• Copies: the conversations started from this one, each openable
• Started from a shared link: this conversation was copied through a share link — the copy was taken by someone who may never have had access to the source
• Public links: the share links cut from this conversation. Each row opens the shared page, links to the message the link stops at, and shows how many conversation copies were started from it

While the conversation is enabled, the Public links list identifies extracts readable by any signed-in user holding the URL. The message each link stops at tells you how much it includes. Disabling the conversation blocks these links; reactivating restores access. Deleting the conversation revokes every link made from it.

Workflow Example

Here’s a typical workflow for reviewing a flagged conversation:

1. Open the conversation from the list (filtered by “Needs Review”)
2. Read the full conversation on the left side
3. Review the LLM evaluation and CX Score
4. Set manual evaluation (Good/Mixed/Poor) with a comment
5. Add tags to categorize the issue
6. Assign reviewer if follow-up is needed
7. Set status if additional action is required
8. Use navigation arrows to move to the next conversation

Best Practices

1. Evaluate consistently: Use the same criteria across all reviewers
2. Add comments: Explain your reasoning, especially for Poor ratings
3. Use tags strategically: Don’t over-tag; focus on meaningful categories
4. Check LLM suggestions: Use improvement suggestions to improve your agent
5. Track status: Keep admin status updated as conversations progress
6. Review feedback promptly: User feedback often reveals important issues

Related Features

• Conversations - Find conversations to review
• Thread Evaluator - Configure automated evaluation
• Topic Analyzer - Set up automatic topic analysis
• CX Score - Understand the quality scoring system
• Feedback Analytics - Analyze feedback trends
• Sharing a Conversation - How the public links listed in the sidebar are created