Customers judge a support agent most by what happens when it hands them to a person. If they have to repeat their name, their problem and what they have already tried, everything the agent did before counts for nothing. A good handover carries the case across intact, so the person picks up exactly where the agent stopped.
Why is the handover the moment that counts?
Because it is the moment the customer's patience is lowest and their expectations are highest. A customer reaches a handover because the agent could not finish the job: the request was complex, needed authority, or involved something sensitive. They have already spent time explaining. Now they expect the person to know what they said.
When the person does not, the customer experiences the whole exchange as wasted. They do not separate "the agent was helpful for five minutes" from "then I had to start again". They remember the restart. It colours how they feel about the agent, the firm and the next time they are offered a chat.
The reverse is also true. A handover where the person opens with "I can see you've been trying to change the delivery address on an order that's already shipped, and the agent couldn't amend it; let me sort that out" makes the agent look good in hindsight. The customer's earlier time was not wasted. It was the first half of the job.
What must travel with the case?
Everything the person would otherwise have to ask for. A handover note template makes this consistent, but the essentials are:
- Who the customer is, with any identity checks already completed and how.
- What they want, in one or two sentences in plain words, not the agent's internal category.
- What has happened so far: what the agent checked, found and tried.
- Why the agent stopped: the specific reason, such as a bound it hit, a trigger it detected or a question it could not answer.
- What the customer has been told, including any timescale or next step.
- Anything sensitive: signs of vulnerability, frustration, a mention of a complaint.
- Links to the records the person will need, already opened.
Item four is often missing and is the one that saves the most time. "Escalated: customer request" tells the person nothing. "Escalated: refund of 450 requested, above the agent's limit of 200; order and payment confirmed" lets them act in seconds. The method for producing these consistently is in how to summarise a case for handover with an AI agent.
What goes wrong in handovers?
| Failure | What the customer experiences | Cause |
|---|---|---|
| Context lost | Asked to explain again from the start | No handover note, or the person's tool does not show it |
| Identity rechecked | Asked security questions a second time | Checks done by the agent not recorded in a form staff trust |
| Wrong team | Transferred again after waiting | Agent routed on a surface reading of the request |
| Expectation mismatch | Told something different from what the agent said | The person did not see what the agent promised |
| Long silent wait | Left in a queue with no update | No estimate given, or no update during the wait |
| Cold start | Person opens with "How can I help?" | Note exists but was not read |
Several of these are not the agent's fault alone. A good note is worthless if the person's screen does not display it, and a perfect route is undermined if the receiving team does not trust the agent's identity checks. Handover is a joint design between the agent and the people receiving the work.
When should the agent hand over?
That is a separate question, covered in when an AI support agent should escalate to a person, but it affects handover quality. An agent that hands over late, after several failed attempts, delivers a more frustrated customer. One that hands over early, as soon as it detects a trigger, delivers a customer who still has patience left. Early, well-documented handovers are almost always better than late ones.
What should the customer be told during the handover?
What is happening, who will help and roughly how long it will take. Something like: "I'm passing this to a colleague in our billing team who can approve the refund. I've shared everything you've told me, so you won't need to repeat it. The current wait is around ten minutes." Then, if the wait runs longer, an update.
The agent should not promise what the person will do, only that the person will look at it. "A colleague will approve your refund" commits the firm to an outcome the agent does not control. "A colleague who can approve refunds will review it" does not. The principles in how a support agent should hand a conversation to a person cover the wording in more depth.
For live chat, where the customer is waiting in real time, the mechanics matter even more. The practical steps are in how to hand a live chat over to a person with an AI agent.
How do you measure handover quality?
Directly, not through overall satisfaction. Useful measures include:
- Repeat-yourself rate: in a sample of handed-over conversations, how often the person asked for information the agent had already collected.
- Time to first useful response: how long after handover the person said something that moved the case forward, not just a greeting.
- Re-transfer rate: how often the case had to move again after the first handover.
- Customer feedback on the handover specifically: one question asking whether they had to repeat themselves.
The first measure is the most telling and the cheapest to collect. A reviewer reads a sample of handovers and marks each one. If people routinely ask for details already in the note, either the note is not good enough or it is not being read, and you can find out which.
What should you do differently?
Treat the handover as a feature you design and test, not a fallback. Write a required set of fields for every handover note. Make sure the receiving team's tools display it prominently. Include handover quality in certification, scoring whether the note is complete and accurate. And measure repeat-yourself rate every week. An agent that hands over well is one customers are willing to try again. That is part of what makes the AI support agent that knows when to say no worth having.
Questions people ask
Should the person read the whole conversation before responding?
They should be able to, but the note should make it unnecessary for routine cases. If people routinely have to read the full transcript, the note is too thin.
Can the customer see the handover note?
There is often value in showing a short version, so the customer knows what was passed on. Sensitive internal observations should stay in the staff view.
What if no person is available?
Tell the customer plainly, give a realistic time for a reply, and make sure the case and note are waiting when someone is. Never leave the customer in an open chat with nobody coming.