You decide what to automate first by crossing two questions per task: how much time it eats (frequency times minutes) and how clear its rule is (whether you can explain it without saying "it depends"). The first candidate is the one that scores high on both: lots of time and a clear rule. It's almost never the flashy automation from the adverts; it's usually that boring thing someone does five times a day without thinking.
The usual mistake is to start with the eye-catching stuff —a chatbot, a report with charts— when right next to it someone is copying data from an email into a spreadsheet for half an hour every morning. Automate first what hurts every day and has fixed rules. The flashy stuff later, if it's even needed.
Why not automate the first thing that comes to mind?
Because the first thing that comes to mind is usually the most visible, not the heaviest. The task that annoys you in a meeting isn't necessarily the one that eats the most hours over the month. And automating something that's barely done is spending effort on something that won't be noticed.
There's another trap: choosing the hardest task "to get it off my plate once and for all". Hard-to-automate tasks are usually hard because they're full of exceptions and human judgement. Starting there is the fastest way for the first automation to go wrong and for no one to want to hear about a second.
Which tasks are good candidates?
The ones with these signs. The more a task gathers, the better a candidate it is:
- Frequent. Done several times a day, or at least every day. What's done once a year isn't worth automating.
- Boring and mechanical. No thinking required: copy, paste, forward, rename. If whoever does it could do it with their eyes closed, that's a good sign.
- With clear rules. It can be written as "if this happens, do that", with no "it depends" or "depending on the day". Exceptions are the enemy.
- With data that already exists digitally. If the information is in an email, a spreadsheet or a system, it can move on its own. If it's in someone's head or on paper, you first have to get it out of there.
The textbook example: someone receives orders by email and types them into a spreadsheet. Frequent, mechanical, clear rules, and the data already digital. That gets automated before anything else.
How do you decide, step by step?
Six steps, and the first two are just looking:
- Note the repetitive tasks of one week. In a normal week, write down every manual, repeated task, without filtering. You'll filter later.
- Measure how much time each one eats. Times done by minutes taken. Ten minutes five times a day is nearly four hours a week. That number sorts the list.
- Score how clear the rule is. Can it be explained without saying "it depends"? Fixed rules with no exceptions = easy. Every case different, needing judgement = hard.
- Cross time with ease. Sort by both at once. The first candidate is the one that eats the most time and has the clearest rule.
- Automate one task, end to end. From start to finish before touching the next. Half a task is worse than none: it forces you to remember the other half.
- Measure it and move to the next. Check it saves what you claimed and broke nothing. With that, go back to the list and repeat.
You don't need software to do this. It's done with a notebook and a week of attention. The deciding part matters more than the programming part, and it costs no money.
How do I tell a good candidate from a bad one?
At a glance, by its traits:
| Trait | Good candidate | Bad candidate (for now) |
|---|---|---|
| Frequency | Several times a day | Once a month or less |
| Rule | "If this happens, do that" | "Depends on the case" |
| Human judgement | None: it's mechanical | You have to judge each time |
| Data | Already digital | On paper or in someone's head |
| Stability | The process has been steady for a while | Changing every month |
A task that falls entirely in the left column is gold. One in the right isn't impossible to automate ever; it's just not first today. Sometimes stabilising the process or getting the data digital is enough to move it to the left.
How much does it save and how do you work it out?
The sum is the usual one and you can do it yourself. Take the task, measure how much time it eats a week and multiply it by what that hour costs. If a task takes four hours a week, that's over two hundred a year: nearly six weeks of one person's work on a single repeated thing.
To that saving, add what you can't see: the type-by-hand errors that stop slipping in, and the fact that the task gets done the same even when the person is on holiday. Subtract what it costs to build and maintain the automation. If the number comes out clearly in favour and the rule is stable, go ahead.
When NOT to automate (yet)?
- When the process still changes every month. Automating something that hasn't settled is building on sand: you build it and have to redo it. Stabilise first; automate second.
- When every case needs judgement. If doing the task requires judging, negotiating or interpreting, automating it fully usually goes wrong. Sometimes you automate the mechanical part and leave the decision to the person.
- When it's barely done. A five-minute task once a month doesn't justify building anything. Doing it by hand is cheaper than maintaining the automation.
What mistakes do people usually make?
Almost all come from choosing with the gut instead of the clock.
- Starting with the flashy stuff. A chatbot while someone copies orders by hand every morning. What shines isn't what saves most.
- Automating everything at once. Five processes in parallel, none finished. One whole is worth more than five half-done.
- Not measuring first. Without knowing how much time a task eats, you automate by feel and then can't tell if it paid off.
- Forgetting the exceptions. The automation works with the normal case and breaks on the odd one no one warned about. Before automating, ask "and when does this not happen this way?".
Frequently asked questions
Where do I start if I don't know what to automate?
By looking, not deciding. For one week, note every manual, repeated task and how much time it takes. The first candidate almost always jumps out on its own: it's that boring thing someone does several times a day without thinking. You don't need an audit; a notebook and a week of attention is enough.
Is it better to automate the biggest task or the easiest?
Neither on its own. A good first automation crosses both: it eats a fair amount of time and has a clear rule with no exceptions. Automating the biggest one but full of odd cases gets stuck; automating the easiest one that's barely done doesn't show. Look for the crossing point.
How much should a task cost me for automating it to pay off?
There's no magic figure, but a simple rule helps: if a task eats more than two or three hours a week steadily and its rule is clear, automating it almost always pays back in a few months. Below one hour a week, it usually works out better to leave it by hand unless it's trivial to automate.
When is it NOT a good idea to automate a task?
When every case is different and needs human judgement, when it's done very rarely, or when the process is still changing and hasn't settled. Automating a process that's going to change next month is building on sand: better wait for it to settle.
Start with the boring, repeated stuff, not the flashy stuff
To decide what to automate first, cross two things per task: how much time it eats and how clear its rule is. The first candidate is the one that scores high on both, and it's almost always mechanical and unglamorous. You choose it with a notebook and a week of attention, not with a tool.
Automate one whole task, measure it, and only then move to the next. And leave out what still changes every month, what needs judgement each time, and what's barely done. The best first automation is the one no one misses, because it was never interesting, only tedious.
Not sure where to start automating?
If you sense time is being lost on repeated tasks but don't know which to tackle first, we can look at your week and tell you which one pays off best. It might be a single thing that a couple of days sorts out. Tell me which task eats more time than it should.
See the automation service Let's talk about your case