Skip to content

Conversations

Conversations is the operator's view of what users actually experienced. Use it to handle work that needs a person, inspect Agent behavior in context, and preserve important new judgment as cases.

Conversation queue with participant, Channel, Agent response, and selected thread

Find the conversation that needs attention

Use the list filters to focus the queue:

  • Agent: Limit the list to one Agent.
  • External User ID: Trace one customer or participant.
  • Feedback: Find conversations with a helpful or improve signal.
  • Source: Separate internal Live Test or Copilot activity from published Channels.
  • Human in the loop: Show conversations currently handled by a person or in AI mode.
  • Ordering: Review newest or oldest conversations first.

Cards marked Needs human reply should usually be handled before general quality review. Combine Source and Human-in-the-loop filters to focus on a particular operational queue.

Review the complete job, not one sentence

Open the conversation and ask:

  • What was the user trying to accomplish?
  • What context did the Agent have at that point?
  • What did the Agent answer or do?
  • Did the result meet the expert's quality and boundary decisions?
  • Does someone need a reply now?
  • Is this a one-off issue or a pattern worth preserving as a case?

Feedback on a specific message is useful evidence, but read the surrounding conversation before deciding what to change.

Decide the next action

Situation Next action
The user is waiting for a person Send a reviewed human reply through the original Channel when available
The operator wants help drafting the reply Generate an AI draft, review it, apply it to the composer, and edit before sending
The current Agent behavior is wrong Use Add Case, define the expected behavior, and diagnose before changing the Agent
The request is outside the approved scope Handle it with the current fallback and consider it for a later version
The conversation is acceptable and low risk Leave it unchanged and continue reviewing the queue

→ Reply to Conversations

Save important judgment with Add Case

Use Add Case when you want the same situation to remain checkable after the Agent changes.

Before creating the case:

  • Keep the conversation context required to understand the task.
  • Make sure Input represents the behavior you want to rerun.
  • Write a Standard that states what must happen and what must not happen.
  • Mark the case as release-critical in your process when it must not regress.

After saving the case, make the smallest relevant change and run the affected cases before publishing a new version.

When end-user identity is masked

If the workspace has Mask end-user identity enabled, Conversations shows a stable pseudonym instead of the raw identity.

  • Select the pseudonym from the external-user filter to review one user's conversations.
  • Free-text searches for the raw external user ID do not match masked records.
  • Conversation content remains available for review, while raw identity and profile fields stay hidden.

A practical operating rhythm

  1. Handle conversations waiting for a human reply.
  2. Review high-consequence or feedback-marked conversations.
  3. Save patterns that must be protected.
  4. Make focused changes.
  5. Verify the affected cases before publishing again.