Where a small business should start with AI.

Start with one job: the one your staff repeat every week that takes the most hours. Build AI into that job, inside accounts your company owns, with a person checking every result before it reaches a customer. Then teach the person who does the job, and check a month later that they still use it. That’s the whole method. Everything else is detail.

Here is the detail.

Pick the job

A good first job for AI has four things going for it:

• It repeats. At least weekly, ideally daily.
• The inputs are clear. A form, an email, a call note, a price list.
• The output is a draft. Something a person can check in a few minutes.
• A mistake gets caught. Nobody outside the company sees the draft before a person has.

In small businesses the same few jobs keep coming up:

1. Quotes and follow-ups. The client’s details go in; a follow-up email and a draft quote come out, built from your own price rules.
2. Weekly content. Topics and first drafts for your posts or newsletter, in your voice, laid out for each platform.
3. Notes into actions. Meeting notes, emails and spreadsheets turned into a summary, a to-do list and the next step.
4. The same questions, answered. Inside the company first: “What’s our policy on…?” answered from your own documents.

If one of these takes someone five hours a week, start there.

What not to start with

• Anything customers see unchecked. A chatbot that answers customers on its own can come later, once you’ve written down the answers it needs and tested it on real questions. Start where a mistake stays inside the company.
• A company-wide AI strategy. A plan for everything usually changes nothing. One job that works teaches you more than ten pages about AI.
• Buying a tool first. Know the job, then pick the tool. Most small businesses need nothing beyond ChatGPT or Claude to begin.

Build it in six steps

1. Gather the material. Price rules, product or service details, the questions you get asked most, and three examples of past work you were proud of. AI copies what you give it, so give it your best.
2. Write the job down. What goes in, what comes out, who checks it, where it’s saved. Half a page.
3. Set it up in your own account. In ChatGPT or Claude that usually means a project with instructions and your reference files. Business plans cost about US$20–30 per person a month, paid to the platform directly.
4. Test it on last month’s real work. Ten cases are enough to see where it fails. When it gets something wrong, fix the material, not just that answer.
5. Teach the person who does the job. On their own work, with a one-page guide: how to start, what to paste in, what to check, what to do when it’s wrong.
6. Check after a month. Is it still used? Where does the draft still need heavy editing? Update the material.

How you know it worked

Keep the measures simple. How long does the job take now, compared with before? How often does the draft need more than a light edit? And the one that matters most: a month later, does the person still use it without being reminded?

If the answer to the last one is no, the problem is rarely the AI. It’s usually the material, the training, or a job that wasn’t a good fit to begin with.

A word on your data

Set it up in accounts the company owns, not someone’s personal login. On the business plans, OpenAI and Anthropic don’t use your conversations to train their models by default; on personal plans, check the data setting. And give staff one simple rule: no customer’s personal information goes into a tool the company hasn’t set up.

When the first job works, pick the next one. That is how AI becomes part of a business: one finished job at a time. The AI page shows three examples of how I set these up, and how to teach your staff to use AI covers step five in detail.