Which tasks should a small business automate first?

The worst first automation is the most interesting one. The best first automation is the most boring one — the task everyone in the business knows is tedious, that happens the same way every time, and that nobody has owned long enough to fix.

Here is the test we use to choose, followed by the candidates that usually pass it, the ones to avoid, and a scoring sheet you can copy.

The five-question test

1. Does it happen often? Daily or several times a week is ideal. Something that happens twice a quarter rarely repays the effort of automating it.
2. Does it follow a stable pattern? If the steps are the same nine times out of ten, good. If every case is different, it is not an automation candidate yet.
3. Does the information it needs already exist somewhere — a form, an inbox, a spreadsheet, a system? Automation moves and reshapes information; it cannot invent what nobody recorded.
4. Does someone own it? A task with no owner has nobody to design the exception path, nobody to notice when it breaks, and nobody to say whether it is still needed.
5. What is the cost of being wrong? Low cost means the automation can run on its own with a periodic check. High cost means a person reviews every output, which is still worth it, but changes the design.

A task that passes the first four is a candidate. The fifth question decides how it gets built.

Candidates by function

These are the tasks that pass the test most often in the small businesses we see. Your own list will differ, but it will probably overlap.

Customer communications

• Contact form submissions that need to be sorted, assigned, and acknowledged before anyone follows up.
• Appointment confirmations, reminders, and rescheduling messages currently sent by hand.
• Replies to the same handful of questions that arrive every day, drafted from approved language for a person to check and send.

Documents and data

• Intake forms copied into two or three systems before work can start.
• Attachments renamed and filed by rules that live in one person's head.
• Exported spreadsheets cleaned up before every report.

Sales and proposals

• Lead details moved from a form into the sales record and a follow-up queue.
• Proposal sections assembled from approved language rather than from the last proposal someone can find.
• Reminders when a quote has gone quiet.

Scheduling

• Appointment requests checked against availability and business rules before a person confirms.
• Calendar changes that notify the right people without a manager forwarding them.

Knowledge

• The internal questions that every new hire asks in the first month, answered from maintained procedures.
• Checklists surfaced at the moment a task starts, instead of buried in a folder.

Finance and admin

• Bills and receipts prepared for entry into the accounting system, with missing details flagged.
• Monthly reporting assembled from exports, with a person writing the narrative.
• Approvals tracked so a stalled item has a named owner.

What not to automate first

• The rare exception. It is the most memorable case, which is why people bring it up, and the worst place to start.
• Anything that needs fresh judgment from a senior person every time.
• A process you are about to change. Automating it locks in the current version.
• A task with no owner. See question four.
• A broken process. Automating a bad process makes it faster and worse. Fix the process, then automate the fixed version.

Rule or AI?

Once a task passes the test, ask one more question: can you write the rule down completely? "When a form arrives with service type X, create a job in the system and email the customer template Y" is a rule, and plain automation handles it. "Read the email and work out what they actually want" is judgment, and that is where AI earns its place — usually with a person reviewing the result until you trust it. Adding AI to a task that a rule could handle makes the system slower to build, harder to test, and harder to explain. We wrote about that trade-off in automation versus AI automation.

How to run the first one

1. Pick one task. Resist the urge to do three.
2. Write down the current steps, including the workarounds, by asking the person who does them.
3. Decide what happens to the exceptions — where they go, who sees them, what they see.
4. Build the smallest version that handles the normal path and routes everything else to a person.
5. Watch it for a few weeks. Keep the old way available during that period.
6. Only then choose the second task.

A scoring sheet you can copy

Score each candidate from 1 to 3 on the first four questions, add them up, and note the cost of being wrong separately as low, medium, or high. The highest total with a low or medium cost is your first automation. The highest total with a high cost is a good second one, built with review built in.

• Frequency: 1 = monthly or less, 2 = weekly, 3 = daily or more.
• Stability: 1 = every case different, 2 = mostly the same, 3 = the same every time.
• Information available: 1 = in someone's head, 2 = partly recorded, 3 = already in a system or form.
• Owner: 1 = nobody, 2 = informal, 3 = a named person.
• Cost of being wrong: low / medium / high.