Where Should a Small Business Start With AI?

Most small business owners we speak to have already tried AI. Someone opened a chat tool, wrote a few emails with it, felt mildly impressed, and then went back to the same working week as before. Nothing was wrong with the tool. The problem is that a tool is not a system.

If you want AI to make a real difference in a service business, the starting point is not "which tool should we use". It is "which piece of work should stop being done by hand".

Start with where your time actually goes

Before choosing anything, spend one week recording where the hours go. It does not need to be elaborate: a shared note or a spreadsheet with three columns is enough:

• What the task was
• Roughly how long it took
• Whether it happens again in a predictable way

At the end of the week you usually see something uncomfortable and useful at the same time: a surprising share of the week goes to work that repeats. Quoting. Chasing. Re-typing. Formatting. Summarising the same information for different people.

That list is your actual starting point.

Look for tasks with four qualities

Not every repetitive task is a good first candidate. The ones worth starting with usually share four qualities.

It repeats often enough to matter. Something you do twice a year will not pay back the effort of setting it up. Something you do five times a week will.

The steps are agreed. If two people in the business do it differently and both think they are right, you have a process problem first. Automating an unclear process only makes the confusion faster.

The inputs are available. The information the task needs must exist somewhere accessible: an inbox, a form, a spreadsheet, a CRM. If the key detail only exists in someone's memory, that has to be solved first.

A mistake is recoverable. Good first projects are ones where a wrong output gets caught in review, not ones where a wrong output goes straight to a client or a bank account.

Pick one. Not five.

The most common failure we see is scope. A business decides to "implement AI", opens six fronts at once, and finishes none of them properly. Six half-built workflows create more work than they remove, because now everything needs checking and nothing is documented.

One workflow, built properly, is worth more than five experiments. Properly means: it works with your real data, someone owns it, the steps are written down, and the team knows what to check.

Decide where a person stays in the loop

Every workflow needs an answer to a simple question: what happens when the output is wrong?

For a first project, the safest answer is usually that a person approves before anything leaves the business. Drafted, not sent. Prepared, not submitted. That single decision removes most of the risk of getting started, and you can loosen it later once you have seen how the workflow behaves with real inputs.

Measure something before you build

Write down one number before you start. Hours spent per week on the task. Average time to send a quote. Number of enquiries that go unanswered for more than a day.

It does not need to be precise. It needs to exist, because otherwise you will have no honest way of knowing whether the work helped. A rough number recorded beforehand is far more useful than a confident number invented afterwards.

A reasonable first 30 days

If you want a concrete sequence:

1. Week one: record where the time goes.
2. Week two: choose one repeating task that meets the four qualities above, and write out its steps as they actually happen.
3. Week three: build or commission the smallest useful version, with a human approval step.
4. Week four: run it on real work, note what breaks, and write the documentation.

At the end of that month you will not have transformed the business. You will have one reliable system, a written process, and a much better sense of what is worth doing second. That is a considerably stronger position than owning six tool subscriptions.

Where this goes next

Once one workflow is stable and documented, the next choice becomes much easier, because you now know what your business is actually like to automate: how clean the data is, how much review the team wants, how quickly people adopt something new.

If you would like a structured version of this exercise, our AI Opportunity Audit does it with you and hands over a prioritised plan. If you would rather get a quick indication first, the AI Opportunity Score takes about three minutes.