Freshchat’s AI Summary uses Freddy to condense a chat conversation into a few sentences plus action items. For long support conversations passed between agents, it saves real time. For quick exchanges, it generates more text than the original. Configure it for the right moments.
When summaries help
The AI summary earns its keep when:
- A conversation crosses 20+ messages.
- A conversation handed off between agents (continuity needed).
- A conversation escalated to tier 2 or to ticket.
- An agent returns to a conversation after hours away.
Below 20 messages, reading the original is faster.
Triggering generation
Configure summary generation:
- On conversation transfer (auto, summary attached as note).
- On conversation close (auto, summary stored on the conversation record).
- On agent demand (manual button).
Avoid summary on every message. The AI cost and latency are not worth it for ongoing exchanges.
Privacy review
Summaries are generated by an AI model that processes the conversation content. Confirm with your data protection officer:
- Where the model processes (region, residency).
- Whether content is retained by the model provider for training.
- Whether your data processing agreement covers it.
For regulated industries (healthcare, financial), document the LLM usage in your data flow diagrams.
Who sees the summary
By default, summaries are visible to all agents on the conversation. For sensitive conversations (HR, security incidents), restrict summary visibility to specific roles via the conversation visibility settings.
Customers do not see the summary. It is internal only.
Action items
The AI extracts action items as a list at the end of the summary. Quality varies; the model sometimes invents action items not in the conversation. Always validate before acting on AI-generated items.
For high-stakes conversations, treat the summary as a draft. Edit before relying on it.
Integration with tickets
When a chat converts to a ticket, the summary becomes the ticket description. This saves the agent re-typing context. Customize the format: summary text, conversation snippet (last 5 messages), original first message.
The ticket description format is set in the chat-to-ticket conversion settings.
Editing the summary
Agents can edit the AI summary. The edit is logged. For audit purposes (regulated industries), retain the original AI summary alongside the edit. The full activity log on the conversation captures both.
Multi-language
The summary generates in the conversation’s primary language. For multi-language conversations (customer in Spanish, agent in English), the summary defaults to the agent’s language.
For multi-region teams, configure the summary language explicitly per agent profile. Otherwise summaries arrive in unexpected languages.
Reporting
Track:
- Summaries generated per week.
- Edit rate (how often agents modify the AI output).
- Conversations transferred without summary (gap in coverage).
A high edit rate (over 40 percent) suggests the AI summaries are unreliable for your conversation patterns. Pause auto-generation and switch to manual.
What to do this week
Enable summary on transfer for one team and measure agent feedback after 7 days. If they find it useful and edit rate is reasonable, expand to other teams. If not, save the AI cost and stick with manual handoff notes.
