Field NotesNext step
How do you find which tasks to automate first?
Track a week of real work, ask one question about every task, and rank what’s left by time. A simple audit beats guessing.
Short answer
A task audit is a one-week log of every recurring task, how long it takes, and whether it needs a specific person’s judgment or just a reliable process. Tasks that repeat, follow the same steps, and need no judgment are your first automation candidates. Rank them by weekly time and start with the biggest one.
Where should you start when deciding what to automate?
Start with a written record of where time actually goes, not a tool. You can’t automate what you haven’t identified, and memory is a poor guide to how a week is really spent.
Most owners know they spend too much time on low-value work. Few have it written down. A feeling of being overwhelmed can’t be turned into a system. A list of tasks with times next to them can.
How do you run a time audit for automation?
For one week, log every task longer than about ten minutes: what it was, how long it took, and whether it needed you specifically. Do it as you go, not from memory at the end of the day.
- Set up a simple sheet with three columns: Task, Time spent, Needs a specific person?
- Log every task over about ten minutes for one full week.
- Ask everyone on the team to do the same, so you see the whole operation, not just your own calendar.
- At the end of the week, group similar tasks and total the time for each group.
What question decides whether a task can move to a system?
“Does this need a specific person’s judgment, or just a capable process?” If a process could do it the same way every time, it’s a candidate for a system.
Be honest with this one. The most misleading tasks are the ones that feel important because you’ve always done them yourself. That feeling isn’t data.
What makes a task a good first candidate for automation?
A good first candidate repeats every week, follows the same steps, needs little judgment, and has a clear way to tell when it was done right.
- It repeats. Weekly or daily, not once a quarter.
- The steps don’t change. A trained person would do it the same way nearly every time.
- It needs little judgment. No pricing, no exceptions, no relationship calls.
- “Done right” is obvious. You can check the result quickly.
Common examples are compiling a weekly report, answering the same status questions, scheduling meetings, sorting inbound inquiries, and drafting first versions of routine messages.
How do you decide which task to automate first?
Rank the candidates by the time they take each week, then start with the one at the top. One system that works beats five that are half built.
| Task | Needs judgment? | Owner after the audit |
|---|---|---|
| Compiling the weekly report | Pulling numbers: no. Reading them: yes. | System compiles, person interprets |
| Answering “what’s the status?” | No | Agent from approved answers |
| Scheduling meetings | No | System |
| Drafting routine replies | Some | Agent drafts, person approves |
| Client strategy session | Yes | Person |
What should you leave off the automation list?
Automate the process, never the judgment. Keep strategy, real client conversations, and final approval of anything that goes out under your name with a person.
A system can compile the data. It can’t decide what the data means for this client, in this market, right now. Your systems should surface work for a person to review, not route around them. For a closer look, see Which client conversations should never be handled by AI?
How do you build the first automation once you’ve picked a task?
Map its inputs, outputs, and decision points. Build that one system. Then run it alongside the manual process for a week before you switch over.
- audit
Log a week of tasks and mark which ones need a specific person.
- rank
Sort the candidates by weekly time.
- design
Map the top task: what comes in, what goes out, and where decisions happen.
- build
Build one system for the top task only.
- parallel
Run it next to the manual process for a week and compare before you rely on it.
How do you keep automated tasks working over time?
Review each system on a schedule, quarterly at minimum, and check whether its output still matches what you need. Systems drift when the business changes and the instructions don’t.
The tool matters less than the discipline to build the system and then look after it. For who should own that work, see Who maintains an AI system after it’s built?
More questions
How long should a time audit take?
One working week is enough to capture most recurring work. Shorter than that and you miss tasks that only happen on certain days.
Should I automate the easiest task or the biggest one?
Start with the one that takes the most time each week, as long as it meets the candidate test. If it’s too complex to map clearly, pick the next one down.
Do I need AI to automate these tasks?
Not always. Scheduling, reminders, and moving form data into your CRM are plain automation. Use AI when the task involves reading or writing that follows a pattern.
What if every task feels like it needs me?
Split the task into steps. Gathering information, formatting, and sending often don’t need you, even when the decision in the middle does.
How many tasks should I automate at once?
One. Get it working and measured before you start the next.