Skip to navigation

Conversation History

6MM AI Customer Support · Tenant guide

View as Markdown

Use conversation history to recover context, review service quality and investigate a specific consultation. Start with a conversation number or a narrow time range instead of browsing every record.

  1. Find the session — Choose filters and the correct date basis.
  2. Read the full context — Review messages, agent and visitor details.
  3. Follow up — Improve an answer or continue the unresolved case.

Find an earlier conversation

Open Conversation History. Use the available user/session search, conversation type, rating, agent, category, language and time filters. Open the details to read the messages and session information.

The user-email list column prefers the contact email and otherwise uses the visitor email. The detail view retains both fields so agents can distinguish their sources. An empty value means that field was not collected; it does not prove a customer has no email account.

Review service quality

Use the conversation, agent and rating together to understand the case. A reply shown as an AI message may come from a configured answer or matching stage rather than a large-model call. Compare the content with the current knowledge and use a flow test when investigating reply behavior.

Deletion is available only with the relevant permissions. Read the confirmation before deleting one or multiple records. MCP history queries use the authorized scope, and deletion requires confirmation; do not assume full tenant history is available to every agent.

Read the fields without mixing identities

FieldInterpretation
Conversation numberIdentifies the consultation you are reviewing. Use it to avoid confusing separate sessions.
Visitor nameA readable display label, not proof of a verified exchange account.
Entry type and user IDHelp distinguish anonymous access from an integrated login identity. Missing identity data should not be guessed.
User emailUses the contact email first, then the visitor email; details keep the two sources separate.
Language and IP locationUseful context for support, not proof of the user’s identity or exact physical location.
RatingFeedback that exists for that session. No rating is not the same as a negative rating.

Example: review a low-rated consultation

  1. Filter by date, agent and rating as appropriate, then open the complete record.
  2. Read what the visitor asked before evaluating the answer. Check whether a handoff or language misunderstanding affected the conversation.
  3. If the answer was incomplete, review the corresponding knowledge content and bot configuration. A poor outcome does not always mean the model itself failed.
  4. Keep the session number with your internal follow-up so the team can return to the same evidence.

Common questions

The session may not have collected a contact email or received an email from the entry integration. Review both fields in details before concluding the data is missing unexpectedly.

The page can filter by start time or close time. A conversation that starts on one day and closes on another belongs to different results under those two choices.

Continue with