Useful things to do with automation

Most automation advice is either too abstract to act on or so specific that it only applies to the author’s stack. This is the middle version: a list of flows we have watched teams build repeatedly, each small enough to set up in an afternoon.

Two rules first, because they matter more than the list.

Automate the boring half, not the judgement half. A flow that drafts a reply and holds it for a human is durable. A flow that sends the reply is a support incident waiting for a bad week.

Every flow needs an off switch someone can find. If pausing an automation requires knowing who built it, it is not a tool — it is a dependency.

Around publishing

Re-queue the evergreen. Tag a handful of posts as evergreen and let a scheduled trigger drop one into an empty slot when the calendar has a gap. It keeps the rhythm without anyone writing filler on a Friday.

Cross-post with a delay. Publishing the same thing everywhere at the same minute reads as a bot. A fifteen-to-ninety minute stagger, randomised per channel, reads as a person catching up.

Hold anything with a link to a draft page. A simple filter that checks the destination returns a 200 before the post can leave the queue. This one catches an embarrassing mistake roughly once a quarter.

Around replies

Route by sentiment, not by keyword. Keyword rules break the first time somebody is sarcastic. Sentiment plus a priority split gets the angry ones to a human quickly and lets the rest wait for the daily sweep.

Escalate on the second message. A first message is a question. A second message from the same person within an hour is a problem. That is a two-node flow and it changes how support feels from the outside.

Draft, do not send. Generate a suggested reply, attach it to the thread, and let whoever is on duty press the button. Response times drop; the tone stays yours.

Around reporting

Weekly digest to the people who never open the dashboard. Most of your colleagues will never log in. A Monday email with five numbers and one chart keeps them informed anyway.

Flag the outliers. A flow that compares each post against the median of the last thirty and pings the channel when something is three times better is more useful than any dashboard. Outliers are where the next month’s content plan comes from.

Close the loop on campaigns. When a campaign’s end date passes, have the automation assemble the rollup, attach the export and open a task for whoever owns the retro. Retros that are scheduled happen; retros that are remembered do not.

A worked example

The re-queue flow is the one most teams start with, and it is small enough to write out:

trigger:
  type: schedule
  cron: '0 9 * * 1-5'
filters:
  - calendar.slots_empty_today > 0
  - library.tag == 'evergreen'
  - post.last_published_days_ago > 90
actions:
  - queue_post:
      slot: next_empty
      channels: [beacon, tidal]
  - notify:
      channel: '#marketing'
      message: 'Filled today's gap with {{post.title}}'

The last_published_days_ago filter is the important line. Without it the flow will cheerfully republish the same three posts forever, which is worse than an empty slot.

When to stop

A good automation programme plateaus. Once the boring half of the work is covered, additional flows start adding coordination cost rather than removing effort — you end up maintaining a small distributed system whose only job is to save ten minutes a week.

Our rule of thumb: if you cannot explain a flow to a new teammate in one sentence, it should probably be a checklist instead.