A warm transfer is a handover in which the customer is passed to another agent or person together with the context of their conversation, such as who they are, what they want and what has been tried, so the receiving person can carry on without the customer repeating themselves. For AI agents, it is how a good refusal ends.
Why do warm transfers matter once agents act?
AI agents hand over constantly. Some cases are outside their role, some need a person by policy, and some customers simply ask. Each handover is a moment where the experience can collapse: the customer waits in a queue, reaches someone who knows nothing, and has to explain their problem again. Customers judge the agent by that moment as much as by anything it said.
A warm transfer makes the handover useful. The person receiving it sees the summary, the verification status, the intent, any vulnerability indicators and the reason the agent stopped. It also makes the agent's decision to stop visible and reviewable.
How does a warm transfer work in practice?
When an agent hands over, it should pass:
- Who the customer is, and how they were verified.
- What they want, in one or two sentences.
- What the agent checked or did, with sources.
- Why it is handing over: outside role, policy requires a person, customer request, uncertainty, or a vulnerability indicator.
- Anything said to the customer about what happens next.
In voice channels, a warm transfer may also mean the person joins the call and is briefed before the customer is connected. Guidance on letting an AI agent transfer calls to a person covers the voice case.
An example from an insurer's claims line
Take a home insurer whose voice agent takes a call about storm damage. The customer mentions that part of the ceiling has collapsed and an elderly parent lives with them. The agent recognises an urgent claim with a possible vulnerability and transfers.
The claims handler receives the policy number, the verified identity, the address, a summary of the damage, the fact that a vulnerable person is in the house, and a note that the agent told the customer someone would arrange emergency help. The handler opens with: "I can see the ceiling has come down and your mother is with you. Let's get someone out today." The customer repeats nothing.
How is a warm transfer different from a cold transfer?
A cold transfer moves the customer to another queue or person with no context. The receiver starts from nothing, and the customer starts again. It is quicker for the agent and worse for everyone else.
Cold transfers are especially damaging after an AI agent, because customers often already doubt the agent understood them. Being asked to repeat everything confirms it. Whether an AI agent should be allowed to transfer a call at all comes down partly to whether it can transfer warmly.
Which terms sit next to warm transfer?
- Escalation: handing a case to a person or more senior level, which a warm transfer carries out.
- Handover note: the context the agent passes with the transfer.
- "Transfer" also means moving money, a quite different capability; whether an AI agent can send an international transfer is a question about payments, not handovers.
Where Pryme fits
In Pryme, a held verdict produces a warm transfer by design: the draft, the sources and the bound that stopped the agent go to a named person together. A support agent that knows when to say no is tested in Attest on whether it hands over at the right moment, and the handover is filed.
Questions people ask
Should the agent tell the customer what it passed on?
Yes, briefly: what the person will see and what happens next. It reassures the customer that they will not have to start again.
What if no one is available to take a warm transfer?
The agent should say so plainly, give a realistic timeframe, and offer a callback or a case reference, with the context saved so the person who picks it up has it.