Closing conversations and analysis

Use AI settings to configure inactivity closure and conversation analysis. Introduction, clarification, unavailable-operator and handoff messages now live in Rules. Each site has its own rules and channel scope.

Close abandoned conversationsCopy Anchor Link

Go to AI settings → Close conversations. Close automatically is off by default in new projects. If enabled, set the Inactivity period (hours) to a whole number between 1 and 8760. A new message resets the timer. A short timeout may be appropriate for a quick pre-sale query but not for a refund case that takes several days.

Enlarge

Prepare the localized closing message in the same section. When you enable automatic closure, publish the default-language wording first. A staff member can also close a thread manually under Conversations, and scheduled support hours are configured elsewhere.

What conversation analysis actually doesCopy Anchor Link

Under AI settings → Conversation analysis, choose Enable automatic analysis when your plan allows it. Analysis summarizes a conversation, assesses sentiment and suggests priority and tags. It does not reply to the customer. You can analyze after closure, during an active conversation or manually. Turning automatic analysis off leaves manual analysis available, subject to plan limits.

Enlarge

For active conversations, the frequency choices are Off, Economy, Balanced and Frequent. Economy runs after at least five new messages and fifteen minutes from the previous successful analysis. Balanced requires three messages and five minutes. Frequent requires one new message and one minute. Each successful analysis uses one credit when an allowance applies.

Apply high-confidence priority automatically and Apply high-confidence tags automatically use recommendations with at least 70% confidence. If your plan has a fixed monthly allowance, reserve some credits for manual analysis by setting an automatic-analysis budget.

Review AI activityCopy Anchor Link

Open AI settings → Activity when you need to investigate what happened. Available traces show the feature, status, provider, model, processing time and who initiated the request. Filter by feature and result, then compare a trace with the related conversation. This is different from the Audit log, which records administrative events. If activity is empty, check whether analysis was enabled, permitted by the plan and triggered by qualifying conversations.