When a Human Node steps in

A Human Node steps in when a case needs human judgment within your policy. It is a money case your policy allows, and either the amount is above what you let the AI handle on its own, or the AI is not yet confident enough to carry it out alone. The case reaches the Human Node already prepared, and the Node validates the decision instead of starting from zero.

The conditions

A case goes to a Human Node when all four are true:

1. It moves money. After collecting the full context, the AI finds a real financial impact, such as a refund, store credit or a chargeback.
2. Your policy covers it and allows the action. The Policy Gate has checked the action against your Charter.
3. The amount is above what the AI can handle on its own, or the AI is not confident enough to carry it out alone. You set that amount for each money action, and above it a Human Node always confirms. See Limits on money actions.
4. The case is prepared. The AI has collected everything a person needs to decide.

The only other way a case reaches a Human Node is the customer's own choice: a customer who, after giving the full context, still wants a person. See When a customer asks for a person.

What never goes to a Human Node

• Informational requests, such as order status or tracking. The AI answers them.
• Actions that move no money, such as an address change or an order edit at the same price. When your policy allows them, the AI carries them out.
• A case no policy covers. A Human Node would have no rule to follow, so it goes to your team.
• An amount above your limit, or a kind of case you keep for yourself. Those go to your team.
• A request outside your policy. The AI says no itself and offers the alternatives.
• Tickets that arrive after your AI resolutions run out. They go to your helpdesk.

How the case arrives

Before passing a case on, the AI has already talked it through with the customer: what is not possible, what the real options are under your policy, and which one the customer chose and confirmed. The Human Node does not ask the customer again what they already confirmed.

The AI Manager, the AI that works beside the Human Node in the Console, hands over the case in plain language:

• which store it is, and who the customer is;
• what happened, and what the AI already did and agreed with the customer;
• which of your rules applies;
• why the case is being passed on.

The Human Node reads the case in seconds and decides: between the options the AI is genuinely torn between, each already within your policy, or by confirming or disagreeing with the single answer the AI is not confident enough to act on alone. The Node writes a reason, which is checked against the case. The AI Manager then writes the messages and carries out the action, after the Policy Gate checks it against your policy. Your limits still apply.

How confidence changes over time

At first, every money case that needs judgment goes to a Human Node. The AI's autonomy on a kind of case can only grow from validated successes on that kind of case, and a single recent failure sends it back to Human Nodes until the record is rebuilt. That autonomy stops at the amount you let the AI handle on its own: above it, a Human Node confirms whatever the record. Your limits never change because of this: confidence only decides who carries out an action your policy already allows.

If no Human Node picks up the case

Cases are shared among the Human Nodes on your brand, going to the one with the lightest load. If a case waits 30 minutes without any Human Node opening or answering it, it goes to your helpdesk. When capacity comes back, it returns to a Human Node, unless your team has already opened and answered it.

A case a Human Node worked on and could not resolve also goes to your helpdesk, and stays there.