The rule
The first thing you automate should not be the biggest process in your company. It should not be the weird workflow that only Brenda understands. It should not be the one that requires seven approvals, three legacy tools, and a prayer.
Start with the 5-minute rule: choose something you do for about five minutes, at least ten times a week, that follows mostly the same steps every time. Five minutes sounds small. That is the point. Small work has fewer edge cases, fewer politics, and a much higher chance of surviving contact with Monday morning.
Ten times a week turns those five minutes into more than 40 hours a year. If two people do it, you have a week of work hiding in plain sight. Not glamorous. Very useful.
What counts
A good first automation has three ingredients: repetition, a clear input, and a clear decision. An email arrives and needs a label. A form response arrives and needs a follow-up. A call ends and needs notes. A receipt lands in an inbox and needs to be filed.
A bad first automation sounds impressive in a meeting. It has twelve exceptions, no owner, unclear data, and a customer-visible failure mode that would make everyone sweaty. Leave that one alone until you have built confidence on quieter work.
Get the weekly field note
One practical AI automation idea, one cost note, and one thing we tried that did not work.
Four examples
A dental clinic we worked with had a receptionist copying appointment requests from email into a scheduling system. Each one took four to six minutes. The steps were the same: identify patient, requested time, service, insurance note, and callback number. We built a draft queue that extracted the fields and prepared the scheduling note. The receptionist still approved it. The clinic saved 6.5 hours a week without changing how patients contacted them.
A bookkeeper received bank statement reminders from twelve clients. The old process was to search, nudge, wait, search again, then mark the spreadsheet. We automated the reminder and status update. The bookkeeper did not need an AI strategist. She needed a machine that remembered who owed what on Thursday.
A roofing company had a more physical version of the same problem. After site visits, estimators texted rough measurements to the office. Someone turned those into a quote starter. We used a form, a transcript summary, and a pricing table to create a first draft. Nobody trusted it blindly. They did trust it to remove the blank page.
A Shopify store had support emails asking where orders were. The team answered with the same steps every time: find order, check carrier, paste tracking, soften tone. The automation retrieved the status and drafted the reply. Humans handled the angry ones. That line matters.
What to ignore first
Do not begin with strategy decks, brand voice generators, or anything called an AI command center. If your first automation needs a steering committee, it is not your first automation.
Also ignore tasks you do once a month. They may be annoying, but the payback is slow. Frequency beats irritation. One five-minute task done fifty times is better than one two-hour task done quarterly.
How to start
Open a note and write down every small repetitive task you touch this week. Do not judge it yet. On Friday, circle the ones that happen ten or more times. For each, write the input, the decision, the output, and the place where a human should approve it.
That is your first automation brief. No mythology required. If you can describe the steps clearly, an AI workflow can probably help. If you cannot describe the steps, the automation is not ready. Fix the process before you ask software to perform it in public.