Handoff rules: which topics your AI should never answer

Start with what it does not touch

Handoff rules are the part of the configuration that pays off most and that fewest people write.

Almost everybody starts with the opposite: they list what the AI has to answer. That is the easy half and also the less important one.

The reason is simple. A good answer arriving on its own saves you time. A bad answer on the wrong topic costs you a customer.

The five kinds worth always escalating

This list works in almost any industry.

1. Anything with anger in it. A complaint with a sharp tone does not get an automated reply, even if the reply would be correct. What the person is asking for is somebody to take responsibility.
2. Commercial exceptions. Off-list discounts, special terms, conditions that are not the usual ones. That is judgement, and judgement belongs to your people.
3. Amounts in dispute. If the customer says they were charged wrong, it goes to a person. Always.
4. Anything needing coordination with a third party. A supplier, a technician, a visit. The AI cannot commit somebody else's calendar.
5. Anything with legal or health consequences. If your industry has this, you already know what it is.

Start with these five escalating and most of the serious problems do not happen.

How to build your own list

The five above are the floor. What is missing is what is specific to your business, and there is a fast way to find it.

Sit down with whoever handles support and ask them one question: what was the last conversation where you had to think before answering?

Repeat it five times. The answers are your list. Nobody could deduce them from a document, because they are exactly the things written in none.

That exercise has an extra benefit worth taking: it makes whoever handles support part of the configuration instead of receiving it finished.

What has to happen when it escalates

Escalating is not transferring and vanishing. Four things have to happen for a handoff to be useful.

• The reason gets recorded. Why it escalated, written down. It is the one action where keeping the reason, not just the fact, is worth it.
• Somebody finds out. By email with a direct link to the conversation, not by hoping somebody checks the inbox.
• The customer does not repeat everything. Whoever picks it up has to see what already happened.
• It is clear who owns it. An escalated conversation with no owner is a lost conversation.

The first is the most skipped and the most useful later. When you review what escalated and why, patterns appear: if the same reason repeats thirty times, that one probably could have been resolved on its own.

Escalate too much at first

Advice that runs against the instinct to optimise.

The first few weeks it is worth escalating far more than necessary. The cost of escalating too much is somebody reading a conversation that could have resolved itself. The cost of escalating too little is a mistreated customer.

They are not comparable, and that is why the handoff list shrinks over time, never the other way round.

A figure from one of our customers

A building management company we serve receives complaints from owners over WhatsApp: leaks, elevators, noise.

It uses handoff intensively and with clear boundaries. Anything requiring coordination with a supplier or sending somebody to the building leaves the automated path and goes to a person. Close to 40% of conversations ends up with the team.

That number did not fall over time, and it does not have to. In that industry the 40% is the real work: it is what happens when somebody has to be moved in the physical world.

Conclusions

• Defining what it does not answer pays off more than defining what it does.
• Anger, exceptions, disputed amounts, third-party coordination and legal topics always escalate.
• Your specific list comes from asking whoever handles support what made them hesitate.
• Escalating without recording the reason wastes half the value of escalating.
• Start by escalating too much: the list shrinks on evidence, not on optimism.