THE PAYOFF

The One Automation We Actually Kept Running

2026-09-19

Setting up automations is easy to enjoy for its own sake. Watching a workflow fire on schedule feels productive even when it isn't saving you much. So before recommending any of the automation recipes on this site, we actually ran a few of them on our own work for a couple of weeks, to see which ones survived contact with a real week.

Most didn't. A social listening setup that pinged us every time a keyword appeared got muted within four days, not because it was broken, but because most of what it flagged wasn't worth acting on. A weekly numbers report was accurate but redundant with a dashboard we already checked. Automating something you weren't going to do consistently by hand doesn't make it more valuable, it just makes the noise arrive on a schedule.

A third one deserves its own mention because it looked the most promising on paper and turned out to be the biggest waste of setup time: an AI summary of every article we published, generated automatically and posted to a personal notes app. The idea was to build a searchable archive of our own thinking. In practice, we never once searched it, because writing the article in the first place was already the thinking, and a summary of your own writing tells you nothing you didn't already know when you wrote it.

The one that stuck was the simplest: sorting incoming email into categories before we ever opened the inbox. Not because it's clever, but because it removes a decision we were making dozens of times a day anyway. Nothing about it required trusting the AI's judgment on anything important. It just did the sorting we'd have done manually, slightly faster, and left the actual decisions to us.

What made that one different from the ones that got muted is worth naming specifically: it operated on volume, not judgment. Sorting a hundred emails into three buckets is a low-stakes, repetitive, mechanical task, and getting one email miscategorized costs almost nothing to fix. The automations that failed were all asking the AI to make a judgment call about what mattered, which is a much harder and much riskier thing to hand off, and the cost of a bad judgment call is a lot higher than the cost of a misfiled email.

That's become our rough filter for whether an automation is worth setting up: does it remove a repetitive decision you were already making, or does it add a new decision you now have to review? The first kind earns its place. The second kind usually gets muted by the end of the week, however good it looked in the demo.

If you're deciding where to start, start with volume over judgment every time. Look for the task you do the same way, the same number of times, every single day, regardless of context. That's the one worth automating first, because there's almost no way to get it wrong badly enough to matter, and the time savings compound the same way every day without you having to think about whether today's version of the task needs a different answer than yesterday's.

It's also worth being honest about the sunk cost that almost kept the failed automations running longer than they should have. Once you've spent an afternoon wiring something up, there's a real pull to keep it alive just to justify the setup time, even after it's clearly not earning its keep. The discipline that actually matters isn't building the automation well. It's being willing to kill the ones that aren't pulling their weight, on schedule, before they quietly become one more dashboard nobody checks.