Where to start with AI automation in a small business
Most small businesses should not start with an AI assistant. Start with one repeated task that eats hours every week, map it, and automate only that.
Published 2026-10-15 · 5 min read

Start with one task that repeats every week
The fastest way to waste money on AI automation is to start with the question 'what could AI do for us?' That question produces a long list of ideas, and none of them gets finished.
Start with the opposite question: which task does someone on your team do the same way, again and again, and dislike doing? Lead qualification, copying enquiries from one inbox into a spreadsheet, chasing unpaid invoices, and writing the same follow-up email are typical examples.
Write the steps down before you touch any tool
Take that one task and write its steps in plain order. What triggers it? What information does it need? Where does the result go? Who handles the cases that do not fit the usual pattern?
The exception handling is the part most people skip, and it is where automations fail. If the answer to 'who handles the strange case?' is 'nobody', the automation will either guess or stall. Name a person for every exception before anything is built.
Use the simplest tool that can run those steps
Many repeated business tasks can be handled by an existing form, a shared inbox rule, or a spreadsheet with one connected step. A custom model is only worth it when the input is messy text that rules cannot classify.
When a language model is needed, keep its job narrow: read one message, extract the fields you need, or draft one reply for a person to check.
Keep a person in the loop for anything a customer sees
Automated drafts are fine. Automated sending to customers, without a check, is how a small business loses trust. Decide in advance which outputs go out automatically, which need a person to approve, and which are never automated.
Measure hours saved and mistakes made
Before the automation starts, record how long the task takes each week and how often it goes wrong. Two weeks after it starts, measure again. If the time saved is small or the mistakes went up, fix the steps or stop.
Once one task works and has been measured, pick the next one. Most small businesses end up with five to ten small automations rather than one large system.

