[object Object]

The Breeze Customer Agent will resolve 30-50% of inbound tickets without human intervention. The story everyone hears is the deflection rate. The story that matters is what happens to the other 50-70% when the bot hands off. That handoff design is the difference between a great customer experience and a worse one than you started with.

Define the handoff trigger explicitly

Don’t let the agent decide on vibes. Configure explicit triggers:

Handoff if:
- User explicitly requests human ("agent", "human", "person")
- Sentiment score < -0.3 across two consecutive messages
- Topic = "billing dispute" OR "cancellation"
- Conversation length > 8 turns without resolution
- Confidence score < 0.6 on the agent's last response

Without these, the agent stays in the conversation past the point of value.

Pass full context, not just a transcript

The human picking up the conversation needs the agent’s understanding, not just the message history. Surface:

  • Detected intent
  • Extracted entities (account ID, product, error code)
  • The agent’s attempted resolutions
  • The reason for handoff

Pipe this into a breeze_handoff_context text property on the ticket. The agent reads it in under 10 seconds and skips the “let me catch up” phase.

Don’t make the customer repeat themselves

The single biggest source of frustration. The human agent must open with context: “I see you’ve been trying to update your billing address and the form errored. Let me look at that with you.” Train your agents to read the handoff context first.

The handoff message matters

The bot’s handoff message should set expectations clearly:

Bad: "Connecting you to an agent."
Good: "I'm bringing in Sarah from billing — she'll have your account history and the error you saw. Estimated wait: 2 minutes."

The named agent and time estimate change the customer’s tolerance.

SLA on the handoff queue

A handed-off conversation must connect to a human within 90 seconds during business hours. Above that, the customer’s session ends and you’ve lost the recovery opportunity. Configure routing to skip OOO agents and prioritize handoffs over net-new tickets.

Track handoff resolution rate separately

A handed-off ticket isn’t a deflection failure; it’s a triaged ticket. Track:

Bot-resolved (no handoff): 42%
Bot-handed-off, human-resolved: 51%
Bot-handed-off, customer-abandoned: 7%

The 7% abandoned is the metric to drive down. That’s where your handoff design is failing.

Fall back gracefully when humans are unavailable

After-hours handoff should not say “we’ll get back to you.” Configure a structured intake: collect the issue type, urgency, contact preference, and SLA-bound callback time.

Train the agent on your handoff conversations

Pipe handoff transcripts (with PII scrubbed) back into the agent’s training corpus monthly. The patterns the human had to fix become things the agent learns to handle directly.

What to do this week

Audit your last 50 handoffs. Categorize by reason. The top three reasons become this quarter’s agent training priorities. Fix the handoff message and queue SLA before adding any new agent capabilities.

[object Object]
Share