What is AI-to-human handover?
AI-to-human handover (also called escalation, live agent transfer or AI handoff) is the moment a customer moves from an automated conversation to a person. It happens in every channel where AI talks to customers: an AI voice assistant transferring a call, a chatbot passing a chat to a live agent, or an AI-triaged email landing in an agent’s queue.
Handover is where many automated experiences succeed or fail. A smooth handover makes the AI feel like a helpful first step. A poor one, where the customer waits in a queue and then repeats everything, makes the whole experience feel worse than if they’d reached a person in the first place.
Why handover matters
No AI handles every conversation well. Even a strong deployment will meet requests it wasn’t designed for, callers it can’t understand, and situations where a person is simply the right answer. What matters is what happens next.
Good handover:
- Protects customers who need a person, especially vulnerable customers and anyone reporting a hazard.
- Saves time because agents start with context rather than from scratch.
- Builds trust in the automated channel, because customers know they won’t be trapped.
- Improves the AI, because handover reasons show where it needs work.
When to hand over
Handover triggers usually fall into a few groups:
- Customer request. The customer asks for a person. This should almost always be honoured.
- Low confidence. The AI isn’t sure what the customer wants, or has misunderstood more than once.
- Out of scope. The request is something the AI isn’t set up to handle.
- Risk. Safety issues, signs of distress or vulnerability, or regulated topics such as hardship and medical dependency.
- Emotion. Sentiment analysis detects growing frustration or anger.
- Value. Some organisations route high-value or complex conversations, such as complaints or cancellations, straight to people.
What good handover looks like
The best handovers work like a warm transfer between two colleagues:
- Context travels with the customer. The agent sees who the customer is, why they called, the details captured and what the AI has already said or done, ideally as a short summary.
- No second queue if it can be avoided. Priority handovers, such as safety calls, should go to the front.
- The customer is told what’s happening. A simple “I’ll put you through to one of the team, and I’ll pass on what you’ve told me” sets expectations.
- Routing is accurate. The handover goes to an agent with the right skills, using skills-based routing.
Common mistakes
- Hiding the exit. Making it hard to reach a person drives repeat calls and complaints, and inflates containment figures misleadingly.
- Losing the context. Transferring the call but not the information is the most common complaint about AI handover.
- Not staffing the human side. Handover only works if there are people available to take the conversation, including during a surge.
- Never reviewing handovers. Each one is a signal. Patterns in why the AI handed over show what to improve next.