Bot-to-human handover: the checklist that saves tickets

A bot-to-human handover works when it fires on an explicit list of signals and carries the conversation’s context with it. Without those two things, automation doesn’t save time — it creates a new way to lose customers, quietly, at the seam between two systems.

The most expensive version looks like this: the bot failed to understand three times, the customer wrote «give me a person», got another automated reply, and left. Technically, everything worked.

Which signals should silence the bot?

Each has to fire without any attempt to «help one more time»:

1. 1

   A direct request

   «Can I speak to someone?» — actioned immediately, no clarifying, no another-try offer.

2. 2

   A second failed attempt

   The question went unrecognised twice — further automation only worsens the impression.

3. 3

   A problem or complaint

   A defect, a delay, a refund. Here the cost of a mistake is a public review.

4. 4

   Haggling over price

   The bot has no authority to concede and can't judge when a concession is earned.

The fifth signal is a non-standard request: a custom order, wholesale terms, unusual delivery. The sixth is anything involving health, children or safety — that category goes to a human regardless of how simple the question looks.

What should you pass along?

Your operator won’t read thirty messages before replying. So the handover is the conversation compressed to three lines:

1. What the customer wants — the specific product, task, parameters.
2. What the bot already said — especially if it quoted a price, a date or a condition. That will have to be honoured.
3. Where you stopped — the last question and whose turn it is.

The third is the most important and the most frequently lost. An operator who doesn’t know the customer already gave their size asks for it again — and the conversation rewinds to the start.

How do you avoid losing people at the seam?

Three rules, each removing a distinct way to lose a ticket.

The operator’s first message demonstrates context. Not «hello, how can I help?» but «hi — I see you’re after a 38, we have it, I’m setting one aside». The customer sees nothing was dropped.

A pause is stated as a number. If no operator is available, the bot names a specific time: «we’ll reply by 10am». «A manager will be in touch shortly» contains no commitment and therefore reassures nobody.

Take the contact before the pause, not after. If the answer will come tomorrow, ask where it’s best to write while the conversation is still live.

How do you check it is working?

One slice: conversations that ended within five minutes of a handover. If that’s a noticeable share, the problem is neither bot quality nor operator quality — it’s the seam.

Two typical causes. A long pause between «let me get a manager» and the actual reply. Or lost context: the operator started over, and the customer concluded nobody was listening.

Pre-launch checklist

1. Six handover signals written as an explicit list in the flow.
2. A direct request for a person is actioned without clarifying questions.
3. Attempt limit: handover after the second unrecognised reply.
4. Three lines of context travel with the conversation.
5. The operator’s first message references that context.
6. Out of hours, the bot states a specific response time.
7. The contact is taken before the pause.
8. Weekly review of conversations that ended after a handover.