Open last week's calendar and your sent folder. Scroll through both and count the things you did twice or more. Some will be obvious: the Monday status email, the weekly invoice reminder, the same three questions you answer for a sales rep every Thursday. Others will hide in the middle of longer tasks, like copying figures from a dashboard into a deck, or checking whether a customer replied before following up. Write the list down. Then pick the one that repeats most often, runs the same steps every time, and needs the least judgment from you. That is your first candidate for automation.
This sounds too simple to matter, but most people skip it. They start with a grand plan, connect five tools, and end up with an assistant that does a dozen things badly. Starting from your own record gives you a short, honest list grounded in what you already do.
Why repetition is the signal
A task you perform twice in a week is a candidate. A task you perform every Tuesday at 9:15 with the same inputs is a strong one. Repetition tells you two things: the work has a stable shape, and the cost of doing it by hand is recurring rather than occasional. Judgment is the other half of the test. If a task requires you to weigh a new situation each time, an AI tool can draft the work but should not finalize it. If the answer is the same whenever the inputs are the same, the task is a good fit for automation.
Look for tasks that move information from one place to another without changing its meaning. Pulling a list of accounts that have not replied in seven days, formatting it, and sending it to the right person is a clean example. Deciding whether a particular late payment deserves an exception is not.
Reading your own history honestly
The sent folder matters because it shows what you actually did, not what you intended to do. Many people believe their week is mostly strategic work. The sent folder often says otherwise. You may find that a large share of your messages are follow-ups, confirmations, or status updates that follow a predictable template. Those are the messages worth examining.
Be specific when you count. "Reports" is too vague. "The Thursday pipeline summary I send to three managers, built from the CRM export" is something you can automate. Specificity also reveals the inputs, and inputs determine whether automation is realistic. If the data lives in one system and the output goes to one place, you have a clear path. If the data comes from four tools and needs a manual reconciliation step each time, the task may be a poor first choice, even if it repeats.
Automating the first one
Once you have picked the task, write its steps in plain language, as if explaining them to a new colleague. Include where the data comes from, what you filter or check, what you produce, and where it goes. Notice the places where you hesitate. Those hesitations mark the points that still need a human yes. Keep them. A good automation runs the mechanical steps and stops at the decisions.
Then run it alongside your manual process for a week or two. Compare the outputs. You are looking for two things: that the automated version matches what you would have produced, and that its failures are visible rather than silent. A missing data source should produce a clear message, not an empty report that looks complete.
Only after the comparison holds should you stop doing the task by hand. Keep a short record of what the automation does and what it still asks you to approve. Revisit that record each quarter, because the repeats in your week will change as your work does.
The payoff is in the next list
The first automated task rarely saves the most time. Its real value is that it teaches you how to see your own work as a set of repeatable parts. The second list you write, next month, will be sharper. You will notice that the same kind of step appears across several tasks, and that one shared piece of automation can cover all of them.
Start small. Open the calendar and the sent folder, count the repeats, and automate the one that asks the least of your judgment. Everything else follows from that first honest count.