Not every annoyance deserves software. Some tasks are rare. Some are pure judgment. Some are already handled by a tool you ignore. Automating those creates a project and leaves the real pain in place.
Look for repetition with a cost
A good candidate happens often enough that people sigh, and a miss has a cost you can describe. A missed call can lose a job. A late quote can lose a job. A sloppy intake means a second visit. A checklist that lives in texts means the office decides without the facts. If you cannot name the cost, wait.
Check that you can explain “good”
If you can explain the task the way you would to a new hire, it can be designed. What does a finished note include? What must never be promised? What does the next person need to see? Vague hope (“be more responsive”) is not a spec. A specific handoff is.
Separate capture from commitment
The tasks that go well are the ones that capture and prepare, then stop. Capture the caller’s request. Prepare the quote brief. Collect the intake. Record the inspection. The commitment — the price, the time, the exception — stays with a person unless you have a narrow case where the answer is always the same.
Prefer one path
The first version should be one path you can watch. All of your operations is not a first version. One service, one kind of call, one form. If that path saves real time or catches real misses, you have earned a second path. If it does not, you have learned cheaply.
If you want a short self-check before you write, use “Is AI right for my business?” It does not store an answer. It only helps you show up to the contact form with a clearer note.
Related: what we can build, a short self-check, and how a project runs.