Things our automations never do alone

Automation is great at the dull parts of running a community. It is also very good at doing the wrong thing quickly. So alongside every helper we've added to this network, we keep a short list of things software is never allowed to finish by itself.

Our "a person clicks approve" list

• Removing a member or banning an account. A bot can flag a pattern. A moderator decides.
• Deleting content. Automations can hide a post pending review. Only a person deletes.
• Anything involving money. Refunds, membership changes outside the normal checkout, discount codes.
• Messages that speak for the community. Announcements, apologies, replies to complaints. A draft is fine. Sending is human.
• Anything in a health or patient support group. Those groups get no automated replies at all.

Making the loop fast

The trick is to make approval quick, or people start working around it. Flagged items land in one queue in Moderators HQ with the reason, the evidence and two buttons. Most take under a minute to handle. The moderators asked for a one-line note on why each item was flagged, and that turned out to be the most useful part of the whole thing.

The list will change as we learn. But I'd rather start strict and loosen it than explain to a member why a script removed them from a group they'd been part of for three years.