Handing off to people before live transfer exists

Let's start with the limitation: zlicx agents can't yet transfer a live call to a person. It's next on our roadmap, but it hasn't shipped, so please don't plan a pilot that depends on it. What you can build today is a callback handoff. Done well, it serves many cases better than a transfer to a queue nobody's watching.

Decide what must go to a person

Write this list before you write the prompt. A typical one:

• The caller asks for a person
• Anything involving money beyond what's been approved
• A complaint or an upset caller
• Requests the agent can't complete after one clear attempt
• Anything that sounds urgent or involves safety

For true emergencies, the prompt should give your approved instruction straight away, such as the right number to call. An agent's job there is to direct the caller, not to assess the situation.

Tell the caller the truth

The closing line matters most. Compare these:

• Bad: "I'm connecting you now." (It isn't.)
• Bad: "Someone will call you right away." (Can you guarantee that at 8:50pm?)
• Good: "I can't connect you directly, but I've noted this for our team. They call back during office hours, from 10 to 6."

Only promise what your team will really do. Keep the hours in the prompt accurate, or pass them in as a dynamic variable.

Collect a useful summary

A callback goes better when the person doesn't have to start from scratch. Use post-call analysis to pull structured fields from the conversation:

• Why they called, in one line
• What has already been tried
• Best number and time to call back
• A priority you've defined, with clear rules for each level

The transcript and recording are in the call log for anyone who needs the full story.

Make someone own the queue

This is where handoffs usually fail. A post-call webhook can send the record to your helpdesk, CRM or a team channel. After that, it's a people problem:

1. Name the owner for each working shift.
2. Set a target for callback time, and measure against it.
3. Create a daily check for requests that haven't been resolved.
4. Have a plan for when your webhook endpoint is down. Test it by turning it off.

Check that it works

Place test calls asking for a person, then follow each one through. Did the request arrive? Did someone act? How long did it take? Track what fraction of handoffs were called back within your target. That number tells you more about caller experience than almost any AI metric.

The AI answering service page shows how this fits an inbound line. Reviewing these calls also has a small cost, which you can size with the pricing formula. When live transfer ships, we'll say so plainly. If it's essential for you, tell us.