When a business first decides to automate something, the choice is almost always emotional: whichever task the loudest person in the room hates most. That instinct is not worthless — irritation tracks repetition — but it misses the two things that actually determine payback.
The two numbers that matter
The first is frequency × duration. A task that takes forty minutes but happens twice a year is a bad candidate. A task that takes four minutes but happens thirty times a day is an excellent one. Multiply them out over a year before you do anything else.
The second is how rule-based it is. If a human has to exercise judgement halfway through, automation gets expensive fast, because you are no longer moving data — you are encoding decisions. Start with the processes where every step has a defensible "if this, then that".
A quick scoring exercise
List every repetitive task in the business. For each one, write down how many times a week it happens, how many minutes it takes, and whether a new hire could be taught the rules in a single sitting. The tasks that score high on all three are where you start.
What good candidates look like in practice
Copying data between two systems that do not talk to each other
Sending the same report to the same people on the same schedule
Creating a record in one place every time something happens in another
Chasing people for status updates
None of these are glamorous. All of them pay for themselves quickly, and each one buys you the credibility to attempt something harder next.
