Human handoff for AI agents: when to pass the conversation to a person
Good AI agents know when to stop. How to design human handoff: the triggers, the summary a person needs, and how the conversation resumes afterwards.
By the Zoft team · · 2 min read
The quality of an AI agent is not only what it handles. It is also what it recognises it should not handle. A customer who is upset, a refund above a limit, a question nobody wrote an answer for: these need a person, and the way the agent passes them over decides whether the customer feels helped or abandoned.
When an agent should hand off
Write the triggers down. Do not leave it to the agent's mood. Common ones:
- The customer asks for a person. Always honour this.
- Strong emotion: anger, distress, repeated frustration.
- A decision above a limit you set, such as a refund amount or a contract change.
- Safety, legal or compliance topics.
- The agent is not confident, or has failed the same step twice.
- A request outside the agent's allowed tools or scope.
What the person receives
A handoff that makes the customer start again has failed. The person taking over should get:
- A short summary of what the customer wants and what has happened so far.
- The details already collected: name, contact, order or account, the issue.
- What the agent already tried, so the person does not repeat it.
- The reason for the handoff and how urgent it is.
- The full transcript, one click away.
Tell the customer what is happening
A short, honest line works: that you are bringing in a colleague, roughly how long it will take, and that they will not need to repeat themselves. If nobody is available, say so and offer a callback or a message, and make sure it is actually logged for follow-up.
Let the agent resume after the person acts
Handoff is not always a one-way door. Often the person makes a decision, such as approving an exception, and the agent can carry on with the rest: send the confirmation, update the record, close the ticket. Designing for that keeps people on the judgement and off the paperwork.
How Zoft handles handoff
Desk in Zoft covers tickets and human-in-the-loop handoff with agent resume: the person decides, and the agent continues. Handoff rules, permissions and approvals are part of the shared evals and controls, and every run keeps a record. The same rules apply whether the conversation started on a call or in chat.
Frequently asked questions
- What is human handoff in AI agents?
- Human handoff is when an AI agent passes a conversation to a person, with a summary and the details already collected, because the situation needs human judgement or the customer asked for one.
- When should an AI agent escalate to a human?
- When the customer asks for a person, when emotion is high, when a decision is above a set limit, on safety, legal or compliance topics, and when the agent is not confident or keeps failing the same step.
- What should a good handoff include?
- A short summary, the details already collected, what the agent already tried, the reason and urgency, and the full transcript, so the customer never has to repeat themselves.