A workflow's trigger decides when it runs and, more than anything else, whether it's useful. The steps are the easy part.
On conversation started — someone opens a new conversation. The most-used trigger, and where greetings and menus belong.
On ticket created — a ticket comes into existence, whatever created it.
On customer comment — the customer writes again on an existing conversation.
On AI question — someone asks something through the AI chatbar.
Triggers exist for an agent comment and an agent note, and for status changes: a ticket being closed, reopened, or moved to a different status.
Closing is the useful one here — it's where a satisfaction survey belongs, and where a follow-up can be scheduled.
Two triggers fire when nothing happens, and they're the ones most worth setting up because nobody notices what they catch.
On no customer response — you replied and heard nothing back. For a polite nudge, or for closing a thread that's evidently finished.
On no agent reply — a customer wrote, and nobody answered. This is the safety net for the conversation that slipped past everyone.
There's an AI-based trigger, which fires on what a conversation is about rather than on a mechanical event, and a pipeline trigger for when an entry is added. And no automatic trigger for workflows that only ever run because something else sends someone into them — a menu item, or another workflow.
A trigger on its own is usually too broad. Set the channels it applies to and, where it matters, the audience — a workflow enabled on every channel for everyone will surprise you somewhere you weren't thinking about.
Then publish it. Until then, it doesn't run, which is also the fastest way to stop one that's misbehaving: unpublish rather than unpick it.
