Half the Things I Wanted to Automate Should Have Been a Checklist

In healthcare the checklist is not a productivity technique, it is infrastructure, and I came to office-style automation with that background. What struck me immediately is how rarely the checklist is considered as an option. The framing is usually manual versus automated, with manual meaning “from memory,” and that skips over the thing that solves most of the problem for almost none of the cost.

Here is the comparison I now run. A checklist takes four minutes to write, has no dependencies, cannot break when a vendor changes an API, works when you are logged out, works for a colleague covering your shift, and — this is the part that is underrated — is self-documenting, so the reasoning is visible to whoever reads it next. An automation is faster to execute and has none of those properties.

So the question is whether execution speed is the binding constraint, and for a surprising share of tasks it is not. The reason a monthly report is late is almost never that typing it takes too long. It is that step four gets forgotten, or that the person doing it is not sure of the order, or that it was nobody’s job this month. A checklist fixes all three. An automation fixes none of them unless it also runs on a schedule and tells someone when it does not.

The rule I have settled on: automate when the task is frequent and identical, checklist when it is infrequent or variable. Frequent and identical means a script pays back its maintenance quickly and its edge cases are rare by construction. Infrequent means you will not remember how the automation works when it breaks, and variable means it will break.

Our roster export is a good example of getting this wrong first. I automated it. It ran monthly, which is infrequent, and the input format varied slightly depending on who had entered the shifts, which is variable. It broke in month five, I could not remember how it worked, and fixing it took an afternoon — longer than the task had taken by hand all year. It is now a nine-item checklist that takes eleven minutes a month and has never failed.

Conversely, filing incoming documents is frequent and identical, and that one is automated and has run without attention for two years.

The hybrid that works better than either, and which I would suggest people try before writing a script: a checklist where two or three of the steps are automated, and the checklist itself tells you to run them. This keeps the human in the loop as the scheduler and the error handler, which is the role humans are actually good at, while removing the tedious parts. It also degrades gracefully — if the script breaks, the checklist step is still there, and the worst case is a slow month rather than a silent gap.

The other property of checklists that has no automation equivalent is that they can be handed to someone else. I have covered for colleagues and been covered for, and a checklist is the only artefact that makes that possible without a conversation. Every automation I have built is, in practice, unusable by anyone but me, and I have started treating that as a real cost rather than an acceptable one.

The four minutes is the thing I would emphasise if I were only allowed one point. Most of the tasks people intend to automate never get automated, because writing the script requires a block of time that does not appear. The checklist version requires four minutes and can be written while the task is happening, which means it actually gets made.