There is a particular kind of task that survives forever: the manual step that only takes five minutes. It is too small to justify a ticket, too routine to complain about, and too frequent to ignore. So it stays, and it quietly costs more than anyone accounts for.
The arithmetic is worse than it looks
Five minutes, twice a week, is roughly nine hours a year. That alone is arguable. Automating it might take a day, and the payback is over a year out, which is why the ticket never gets written.
But the five minutes is not the real cost. The real cost is that it has to be done by a person who knows how, at a particular time, which means it blocks. It cannot happen while that person is on holiday. It is a step somebody has to remember, so occasionally it is forgotten, and the failure from forgetting it is usually much more expensive than nine hours. It is also a step that gets done slightly differently each time, so the system drifts.
And it is a context switch. The interruption costs more than the task, which is a well-worn observation about knowledge work in general and applies exactly here.
The good candidates
Not everything manual should be automated. The ones worth attacking share a few properties:
- It happens on a schedule or a predictable trigger, rather than in response to something novel each time.
- The steps are the same every time, or vary along one obvious axis.
- Getting it wrong has consequences, so the reliability gain matters as much as the time saved.
- Exactly one person knows how to do it. This one is worth prioritising regardless of frequency.
That last case is not really about toil, it is about a single point of failure that happens to be a human being.
Automate the decision last
The mistake I have made most often is trying to automate the whole thing at once, including the judgment call in the middle. That is where automation gets stuck, because the judgment is the part that resists specification.
It works better to split it. Automate the gathering and the doing, and leave the deciding to a person for a while. A script that assembles everything, shows you what it is about to do, and waits for you to confirm removes most of the cost immediately and takes an afternoon rather than a week. Later, when you have watched it be right fifty times in a row, the confirmation step can go.
Write down what you eliminated
Toil reduction is invisible work. Nobody notices the incident that did not happen or the step that no longer needs doing. If you do not keep a record, the effort reads as a quiet quarter.
A short list is enough: what the task was, how often it ran, and what it costs now. It makes the case for the next one, and it is a much better answer to "what did you get done" than a list of tickets.