What should an SME actually automate first?

Where the time actually goes

In a 2025 OECD survey of more than 5,000 small and medium-sized enterprises across seven countries, about 31% said generative AI was in use in their business. Of those, only 29% used it in their core activities. The rest used it at the edges, mostly for simple, one-off tasks: a rewritten email here, a social post there.

That isn't wasted effort, but it isn't where the time goes either. The hours that drain a small team are spent on the same tasks every week: answering the same customer questions, copying figures from one document into another, chasing the information needed to finish a report.

So where should you start? Not with the most exciting idea, and not with whichever tool is in the headlines. Start with the most boring task that passes four tests.

The four-question filter

Before committing to any AI initiative, evaluate the candidate task against these four practical criteria:

• 1. Does it happen every week, or more often? A task done twice a year will never repay the effort of automating it. A task done fifty times a week will. Frequency is where the savings come from.
• 2. Could someone write down the rules? If a capable new hire could learn the task from a one-page guide, it's a good candidate. If it depends on judgment that even your best people struggle to explain, leave it for later.
• 3. Does the information already exist somewhere? AI works best when the answer is already in your documents, inbox, website or spreadsheets, just hard to find or slow to assemble. If the information doesn't exist yet, you have a data problem to solve first, not an AI one.
• 4. Is a mistake cheap to catch? Your first project should be one where a human can easily review the output, and where an occasional error costs minutes, not money or reputation.

Where the first wins usually are

Across commercial businesses and operational teams, the same four high-leverage areas emerge consistently:

• Repeat customer questions: Opening hours, delivery times, how to book, and what's included. An assistant that answers only from your own website and documents, and hands off to a person when it isn't sure, can take a real load off a small team.
• Turning documents into data: Invoices, application forms, delivery notes, field reports. Pulling the same ten fields out of every document is slow, error-prone work, and a well-built system can do the first pass for a person to check.
• Drafting routine replies and summaries: Quote requests, follow-ups, meeting notes, weekly updates. AI drafts and a person approves. The time saved is in the blank page, not the final decision.
• Finding internal information: "Where's the policy on refunds?" "Which template do we use for this?" When knowledge sits in scattered folders and in people's heads, a search assistant grounded in your own documents saves a lot of interruptions.

What to leave until later

Some tasks look tempting on paper but make hazardous first projects:

• Anything legally binding, such as contracts, compliance sign-offs or medical advice
• Anything that moves money without a human check
• Anything that depends on information you haven't gathered or cleaned up yet
• Anything where nobody on the team would own the result

Not everything needs AI

This is the part most AI advice skips. When you look closely at a time-consuming task, the fix is often simpler than AI: a shared template, a better form, a spreadsheet rule, or a clear checklist.

Use AI where language, ambiguity or volume make simple software fall short. Use plain tools everywhere else. They're cheaper, easier to maintain and easier to trust.

How to size the opportunity

Before spending anything on software or consulting, run this straightforward calculation:

• Illustration (hypothetical): A small clinic's front desk answers around 40 calls and messages a day, and roughly half are about opening hours and booking.
• That task passes all four tests: it's frequent, rule-based, the answers are already on the website, and a wrong answer can be caught by a person.
• An assistant that answers those questions from the clinic's own information, and passes anything else to staff, frees real hours every week without touching anything sensitive.

A one-page checklist

Before you choose your first AI project, verify that:

• The task happens at least weekly
• The steps could be written down
• The information it needs already exists
• A person can easily review the output
• An error would be cheap to fix
• Someone on your team will own it
• You've estimated the hours it costs today
• You've checked whether a simpler, non-AI fix would do

The takeaway

The businesses getting real value from AI aren't the ones with the most tools. They picked one recurring, well-understood task, did it properly, and measured the result. Start small, prove it works, then build from there.

Not sure which task to start with? Our AI Opportunity Review maps your workflows, identifies the three to five places where AI would genuinely help, and ranks them by value, risk and effort. Write to us at ada@blueberryia.com.