Should a voice agent aim for fewer transfers to humans, or better ones?

Transfer rate on its own is a weak goal, because it is easy to push down in bad ways: the agent keeps a frustrated caller on the line, or ends calls that should have reached a person. What you actually care about is whether the caller's problem got solved, so measure that first and treat transfer rate as a side number.

Decide up front which calls must always go to a human: complaints, cancellations, payment disputes, anything involving safety or legal risk, and any caller who asks for a person. Those transfers are the agent working correctly, not failing. Then look at the rest and find the call types where the agent hands over without need, and fix those one by one.

A good transfer is mostly about what the human receives. Pass a short summary with who called, what they want, what the agent already checked and why it is handing over, so the caller never has to repeat themselves. Callers forgive a transfer; they do not forgive starting again from zero.

Review a sample of calls every week in both directions: calls that were transferred but did not need to be, and calls that stayed with the agent but should have been transferred. The second group is the one a low transfer rate hides.

Listings mentioned

• Customer Escalation Brief Skill · skill by mohitagw15856
  A structured brief with context, timeline, root cause and next steps, a good template for what a human should receive on handover.
• LLM evaluation · skill by wshobson
  Sets up automated metrics plus human review for an AI system, so you can track resolution quality instead of only transfer rate.

Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.