Keep a human handover for conversations involving risk, distress, exceptions, negotiation, complex judgement or a clear request for a person. The handover should preserve context, explain what happens next and make it easy for the team to take responsibility.
The choice is not “AI or humans”. Useful systems decide which part of a journey benefits from speed and consistency, and which part needs experience, discretion or a genuine relationship.
Handover when the customer asks
If someone clearly asks for a person, the assistant should not keep defending its own usefulness. It can gather a small amount of routing information if appropriate, but the route out should be visible and honest.
Handover when emotion changes the conversation
Complaints, distress and frustration need careful handling. AI may acknowledge the situation and collect essential context, but an upset customer should not be trapped in a loop. A person needs enough information to continue without asking them to repeat everything.
Handover when the answer carries consequence
Some questions involve contractual decisions, unusual pricing, safety, legal interpretation or regulated advice. The assistant can explain the normal route or say what information is needed, but it should not invent authority it does not have.
- The visitor explicitly requests a person.
- The conversation shows distress, anger or repeated misunderstanding.
- The request falls outside approved knowledge or normal policy.
- A decision needs negotiation, discretion or accountable judgement.
- The assistant cannot verify that a promised action succeeded.
- Identity, privacy or security checks require a protected process.
Design a warm handover, not a dead end
“Someone will contact you” is not enough if nobody owns the task. A useful handover states what the assistant can do now, what the person can expect and which route is available. That may be live transfer, a call booking, an enquiry with context or a named follow-up task.
Pass context, not just contact details
The team should be able to see why the visitor got in touch, what they have already said and what the assistant attempted. Good context reduces repetition. It also helps the person recognise when the automated conversation went wrong and respond appropriately.
Let people own the exceptions
Automation is strongest around repeatable work. Humans remain valuable where the business wants flexibility: unusual requests, long-standing relationships, commercial judgement and the moments when a customer needs to feel genuinely heard.
Review handovers as a source of improvement
A high number of handovers is not automatically a failure. It may show where customers need judgement. It may also reveal missing knowledge, an unclear policy or a broken action. Review the reasons, then decide deliberately whether the AI, the process or the human response should change.