Almost every small business we walk into says the same thing: everything is manual, and everything is urgent. That is not a priority problem, it is a measurement problem. Nobody has counted the hours, so the loudest task wins.
Start with the task that repeats, not the task that hurts
The most painful job is usually the one you have already learned to survive. The better target is the job that happens most often, in the same shape, with the same inputs every time. Quoting. Booking. Sending the same six answers to the same six questions. Repetition is what automation is good at; judgement is not, so leave the judgement alone for now.
A rough way to find it: for one week, write down every task you do more than three times a day and how long each one takes. Multiply by four weeks. The number that comes out is usually between eight and twenty hours a month disappearing into work that never needed you specifically.
Then check whether the steps are actually stable
Automation assumes the process does not change shape every week. If the way you quote depends on the customer, the weather and your mood, the fix is to settle the process first — with a template or a short checklist — and automate the second version. Automating chaos just makes the chaos faster.
What not to start with
Do not start with the thing that would embarrass you most if it went wrong. Pricing decisions, credit approvals, anything that touches a customer's money or a legal commitment: keep a human on the trigger until the automation has earned it. In practice that means starting with intake and follow-up, which are high-volume, low-risk, and where slow replies cost the most.
Load-shedding is worth planning around here. A local automation that runs on a laptop dies at 8pm; one that runs in the cloud keeps answering while the office is dark. That difference alone is usually worth the first project.