Discord Support Escalation: When a Bot Should Hand Off to a Human

What escalation is for

An escalation is a handoff. It says: this ticket has stopped being something automation can finish, and here is what a person needs to know.

Two failure modes, both common:

The second is more visible. The first does more damage, because it destroys the signal quietly.

When should a bot escalate?

The rule that holds up: when the next step requires a decision or an action only a human can take.

That covers issuing a refund, changing someone's account, making a judgement call on a rules dispute, or anything where the answer is "it depends on what we decide".

It does not cover:

AI Ticket Bot's system prompt forbids all three of those failure patterns explicitly: escalating on frustration, escalating before searching, and escalating as a safe default.

Priority, and how to keep it honest

Priority is only useful if the levels mean different things, and the test for each one is what waiting would cost, not what the ticket is about:

Above High sits Urgent, for harm that is happening right now: a threat to someone's safety, a raid or a takeover in progress, a live scam or leak, money being lost, or something broken for many people at once. Be strict about both raised levels. If "high" starts including "member is impatient", the level stops meaning anything within a week.

Which role should get pinged?

One. Exactly one, and one you chose.

AI Ticket Bot resolves it in this order: the category's responsible staff role if you set one, otherwise the panel's, otherwise nobody gets pinged. There are no hard-coded pings. Nothing is going to invent an "@billing" role you never created.

The AI never mentions individual users and never pings everyone. Both are deliberate: a bot with the ability to ping everyone will eventually do it at 3am.

Set the role per category. That is the whole point of categories, and a billing escalation reaching your general moderators helps nobody.

What a good escalation message contains

The human arriving should not have to read the whole thread. A useful handoff carries:

Ours posts a short summary embed with the priority. The test is whether a staff member can decide in five seconds whether this is theirs to pick up.

After the escalation

Once a ticket escalates, the AI should get out of the way. On AI Ticket Bot the AI is disabled on that ticket, and it also stops when a staff member sends a message or when the ticket is claimed.

That last part matters more than it sounds. Nothing is worse than a bot cheerfully interjecting while a staff member is handling a delicate conversation.

Escalation paths worth setting up

That last one is the metric that tells you whether escalation is healthy. Near zero is suspicious rather than excellent: some tickets genuinely need people. The analytics guide covers how to read it alongside the AI resolution share so the pair tells you something true.