What small businesses should automate first (and what to leave alone)
A practical way to pick your first automation: find the task that is repetitive, rule-based and high-volume — and be honest about the ones that are not.
Most small businesses that ask about AI start from the wrong end: they pick a tool and then look for somewhere to put it. The order that actually works is the reverse — find the task that is quietly eating your week, then decide whether software can do it reliably.
Three questions that identify a good first candidate
A task is worth automating when all three answers are yes:
- Is it repetitive? You or someone on your team does it more than a few times a week, in roughly the same way each time.
- Is it rule-based? A new hire could be taught it in a short session, because the decisions follow from the information in front of you.
- Is it high-volume enough to matter? Saving four minutes on something you do twice a month is not worth a project.
If a task fails any one of these, it is usually a bad first project — not because it is impossible, but because it will take longer to build than it saves, and you will conclude that automation "doesn't work here".
The usual winners
In practice, the first automation for a small business is almost always one of these:
- First-response to enquiries. Every hour an enquiry sits unanswered, the chance of it converting drops. An assistant that replies immediately with a useful answer — and flags the serious ones for a human — pays for itself quickly.
- Appointment booking. Especially calls that arrive outside working hours, which currently go nowhere at all.
- Data moving between tools. Forms into the CRM, orders into the spreadsheet, invoices into accounting. Nobody enjoys this and nobody notices when it is done well.
What to leave alone, for now
Be careful with anything where a wrong answer is expensive and hard to reverse: pricing quotes on complex jobs, anything legally binding, medical or financial advice, or final decisions about people. These can be assisted by AI — a draft for a human to approve — but should not be handed over outright.
The same goes for tasks where the rules live only in one person's head and change constantly. Automate that and you have built something that will be wrong within a month.
Start with one thing, measure it
Pick a single workflow. Write down what it costs you today — hours per week, enquiries missed, whatever the honest number is. Build the smallest version that works. Then compare. If the number did not move, you have learned something cheaply. If it did, you now have a case for the next one.