How to spot a candidate task for automation

Not every task is suitable for automation. A good candidate meets three criteria: it is repetitive (regularly, not once a year), it has clear rules (you can verbalize when to do what), and the input is predictable (data comes in in a similar form). The selection mechanics are like a sieve: you run the task through these three questions, and if at least one is a firm “no”, the task is not yet for the conveyor. Leverage: Make a list of what you do with your hands in a week, and next to each item note the frequency and 'is there a hard and fast rule'. Insider: the most delicious candidates are not the longest ones, but the most frequent and boring ones, where you copy and paste on autopilot. Second trick: if a task is almost suitable, but requires a solution “by eye” - do not throw it away, but take the controversial part to a person, and give the rest to the machine. A typical beginner's mistake: jumping into a bright, complex problem ('let the AI ​​do the talking') where there are no stable rules. The result is a fragile script that breaks down in every non-standard case and requires constant nannying.