Most conversations about AI in small business start in the wrong place: with the tool. "Should we use ChatGPT?" "Do we need an AI chatbot?" The better question is simpler and less exciting: where does work get stuck, repeated, or dropped?

AI is not a strategy. It is a very capable assistant that is good at specific kinds of work — reading, sorting, drafting, summarising, answering repeat questions. Pointed at the right problem, it pays for itself in weeks. Pointed at the wrong one, it becomes another subscription nobody uses.

Step 1: List the work nobody enjoys

Ask your team one question: what do you do every week that feels like a waste of your skills? The answers are usually the same handful of things:

  • Replying to the same enquiry questions again and again
  • Copying details from forms or WhatsApp into a sheet or CRM
  • Writing quotes, follow-ups and proposals from scratch
  • Compiling a weekly or monthly report

That list is your AI shortlist.

Step 2: Put a number on each item

For each task, estimate two things: hours per week, and what happens when it's done late or badly. Slow enquiry replies, for example, often cost more in lost customers than in staff time. A task that takes two hours a week but loses one customer a month is worth more than one that takes five hours and loses nothing.

Step 3: Choose the boring, contained one

Your first AI project should be small, frequent and easy to check. Good first projects:

  • Enquiry first-response: an assistant drafts an instant, accurate reply using your FAQs and prices; a person approves or edits.
  • Data capture: details from emails or forms are extracted and added to your sheet or CRM automatically.
  • Proposal drafting: a first draft built from your templates and the client's brief, finished by a human.

Notice the pattern: AI does the first 80%, a person checks the last 20%. That keeps quality high and risk low while you learn.

Step 4: Measure, then expand

Track the before and after: response time, hours saved, errors, conversion rate. If the numbers move, take the next item on the list. If they don't, you've learned cheaply.

What to avoid

Don't start with a customer-facing bot that answers anything without review. Don't automate a broken process — fix the process first, then automate it. And don't buy five tools before you've proven one use case.

The businesses that get real value from AI aren't the ones with the most tools. They're the ones that picked one real problem and solved it properly.