[Cloud Chat] Ticket Transfer to Level 2 (L2)

1) What is L2 Transfer?

It is the process (handover) where ClaudIA stops automated assistance and redirects the conversation to a human (team or agent) in the helpdesk.

Why it happens: to ensure that complex, sensitive, or frustrating cases are handled by people, with context and actions that AI cannot perform.

ClaudIA can transfer automatically based on various triggers. For each trigger, we explain what happens and what you can do:

What it is: ClaudIA assesses the confidence score of responses found in the knowledge base. If the score is repeatedly low, it indicates that the AI does not have a reliable answer.
How it works (internal step-by-step):
1. ClaudIA searches for relevant content in the database.
2. Calculates a confidence score for each content piece.
3. If the score falls below the configured threshold, it increments a counter (count_limit).
4. When the counter reaches the set limit, it triggers a handover.
What the admin should do: review the Low Confidence Threshold and Count Limit, improve Level 1 (responses) content, and test in staging.

What it is: the customer asks to speak with a human (e.g., "I want to speak with a representative").
How it works (internal step-by-step):
1. The intent detector recognizes predefined phrases/intents.
2. ClaudIA performs an immediate handover.
What the admin/training should do: adjust the list of phrases/intents that trigger handover; test variations (e.g., "agent", "human support", "real person").

What it is: ClaudIA accumulates signs of dissatisfaction (words, repeated complaints, tone) and generates a frustration score.
How it works: when the score exceeds the limit, a transfer is triggered to prevent worsening the experience.
What to do: review the frustration threshold; train containment messages before handover; monitor frequent causes.

What it is: when a conversation exceeds a configured number of messages (e.g., 30), a transfer is triggered to avoid loops and loss of context.
What to do: adjust the Message Limit to your flow or enable partial routing.

What it is: some topics in the knowledge base can be intentionally marked as N2 — always forwarded to a human.
What to do: review the list of contents marked as N2 and ensure only truly complex cases are marked.

What it is: an agent or supervisor can interrupt ClaudIA and assume the ticket.
What to do: train agents to use the “Take Over” button correctly and ensure ClaudIA properly exits the interaction.

Trigger detected (one of the above).
Message to customer — ClaudIA sends a standard (configurable) message informing that a transfer will occur.
- Note: if enabled, transfer may happen without sending a message.
Create an internal note with: reason for handover, scores (if applicable), last conversation snippets, and content attempted.
Apply tags that describe the subject.
Forward the ticket to the configured L2 group (or the group defined by tags).
L2 agent assumes: opens the ticket, reads the note/transcript, and continues the support.

Verify Tagger: confirm the tag engine is active and with correct rules.
Map tags → groups: in the integration panel, define that billing goes to Billing Team, technical to Technical Support, etc.
Test mapping: create test conversations that trigger each tag and confirm the ticket reaches the correct group.
If NO_TAG appears:
- Check if the conversation triggered any content.
- If not, review the Tagger rules/expressions or add fallback tags.

Create ClaudIA account/agent in the helpdesk.
Create the N2 group(s) in the helpdesk (e.g., N2 Support, Billing).
Configure integration / webhook between Cloud Chat and your helpdesk.
Map tags → groups in the integration panel.
Adjust triggers in ClaudIA’s ClaudIA panel:
- Low Confidence Threshold (e.g., 0.5–0.6 as a starting point)
- Count Limit (e.g., 3 attempts)
- Message Limit (e.g., 30)
- Frustration Threshold (e.g., 0.7)
- Enable/disable "handover without message" according to policy.
Customize default messages (user message and internal note template).
Test in staging: simulate low confidence, customer request, frustration, and long conversation.
Publish / monitor: after deployment, track handover rates, NO_TAG instances, and most frequent reasons.

Open the ticket and read the internal note immediately.
Read the latest messages (partial transcript provided by ClaudIA).
Check applied tags — helps prioritize and route.
Take over the conversation (change owner/status as per your process).
Initial message to the customer (suggestion):
- "Hello [Name], this is [Your Name] from [Team]. I will continue the support that started with ClaudIA regarding [subject]. Can you confirm [data/latest action]?"
Record internal actions and update the ticket with resolutions, attachments, or notes.
Close the ticket when resolved.

Transfer does not reach the team:
1. Check if webhook/integration is active.
2. Verify if the N2 group exists and is active in the helpdesk.
3. Test with a staging handover and review logs.
Ticket without tags (NO_TAG):
1. See if ClaudIA triggered any content.
2. Review Tagger rules and add fallback tag if needed.
Too many handovers due to low confidence:
1. Increase Count Limit or adjust Low Confidence Threshold.
2. Improve/add more comprehensive Level 1 content.
ClaudIA continues responding after "Take Over":
1. Check if "Take Over" was registered in the helpdesk.
2. If necessary, restart manual handover and have the team monitor the ticket.

Customer: "Why was I charged twice?"
ClaudIA: attempts answers with scores 0.28, 0.31, 0.29 → Count Limit reached.
Action: ClaudIA sends a handover message, creates a note: Low confidence, tags the ticket as billing, and forwards to Billing Team.
L2 Agent: opens ticket, reviews note, responds explaining next steps.

Customer sends repeated messages with an aggressive tone.
ClaudIA detects high frustration score → handover.
Note: Frustration detected (score 0.82).
Agent: quickly takes over, apologizes for the experience, and suggests next steps.