Most people pick the task that annoys them most. That is understandable and usually wrong, because the most irritating job is often rare, complicated, or the one that genuinely needs judgement. The jobs worth automating first are the boring ones nobody complains about, because they happen so often that nobody notices the total.
The test
A task is a good first candidate when all four of these are true.
It happens often
Daily beats weekly, weekly beats monthly. Frequency is what turns a small saving into a real one, and it is also what makes the thing worth maintaining.
It follows the same steps every time
If you can explain it to a new starter as a sequence, it can be automated. If your explanation contains "it depends" more than once, the judgement is the job, and that part should stay with a person.
It has a clear input and a clear output
"An email arrives, the details go into the CRM" is clear. "We manage customer relationships" is not a task, it is a department.
Getting it wrong is recoverable
Start where a mistake is embarrassing rather than expensive. You want the first automation to build confidence, not to be the one that pays a supplier twice.
Where these usually hide
Across the businesses we talk to, the same four keep coming up.
- Re-entry. The same information typed into two or three systems because they do not talk to each other.
- Chasing. Quotes, renewals and unpaid invoices that go quiet because nobody had time to follow up.
- Answering the same question. The twenty queries that make up most of your inbox, each answered from scratch.
- Assembling reports. Someone pulling figures out of several places into a spreadsheet every week.
None of them are dramatic. That is exactly why they survive: no single instance is bad enough to fix, so the total never gets counted.
What to leave alone, at least for now
- Anything where the rules genuinely change case by case. You will spend the budget encoding exceptions.
- Anything that happens a handful of times a year. Annoying is not the same as expensive.
- Anything nobody can explain. If the process is not understood, automating it just makes the confusion faster.
- Anything where a wrong answer reaches a customer with no human in between, until you trust the rest.
That last one is a sequencing point rather than a permanent rule. Customer-facing automation is fine once you have seen the system behave for a few months. It is a poor place to begin.
A useful exercise
For one week, ask everyone to note any moment they think "a computer should be doing this". Do not analyse it, just collect it. At the end of the week the list will have obvious repeats, and those repeats are your shortlist.
It works because it captures the small daily friction that never makes it into a meeting. People will not raise a two-minute task in a planning session. They will write it down when it happens for the fourth time that morning.
Then be honest about the numbers
Take the top two or three, estimate the hours each consumes weekly, and multiply out to a year. Some will obviously justify a build. Some will not, and knowing which is which is worth the exercise on its own.
If you want a second opinion, describe the job to us and we will tell you honestly whether it is worth automating, including when the answer is no.