Why didn't my automation run?

Every run is recorded, so this is answerable rather than a guess. Open the automation and switch to the Runs tab: if there is a run for the response you are thinking of, the automation started and something inside it went wrong. If there is no run at all, it never started.

There is no run

• •It is off. Publishing does not turn an automation on — check the switch in the header.
• •You published after the response arrived. An automation only acts on things that happen after it goes live.
• •The trigger is watching a different form, list, or property than you think.
• •The trigger has conditions and this response does not meet them. The test panel tells you this outright: "Nothing would run: the conditions on the trigger do not match this response."
• •The response was blocked as spam, so it never counted as a submission.

There is a run, and it failed

Open it. The canvas shows the path it took and stops at the step that failed, with the error next to it. The usual causes are an app you connected that has since been disconnected, an email address that does not exist, or a step that is missing something it needs.

There is a run, and it is still waiting

Waiting is a normal state. A run sitting on a Wait step is doing exactly what you asked; a run sitting on an approval is waiting for a person. The run shows what it is waiting for and, if you set one, when it gives up.

Fixing and re-running

Once you have fixed the cause you have two ways to run it again. Retry carries on from where it stopped, on the same version — right after a temporary problem. Replay starts a new run from the beginning on the version running now, and lets you correct the answers first — right after you have changed the automation itself. You can also replay everything matching a filter, for the case where one broken step took out an afternoon's responses.

Things worth checking now and then

The filter row above your automations grows two more chips when there is something to say. "Needs review" collects automations that are failing or half-configured. "Unused" collects ones that are on but have not run in a long time — usually a trigger pointing at a form nobody uses any more. Neither appears while there is nothing in it.