What AI can actually do for a 20-person business.
Forget the demos. Here are the three jobs AI does well in a small business, what it costs to try, and a one-week plan to find your first win.
Daine MawerFounder, Plaincraft
3 min read
The short answer
In a business of around 20 people, AI is most useful for three jobs: writing first drafts, reading documents and pulling out details, and answering routine questions. The best place to start is the task your team repeats most often each week. Most first projects run on tools you already pay for.
Start with the work, not the tool
Most owners I speak to start in the wrong place. They try a chatbot, get a mediocre answer and decide AI isn't for them. Or they buy licences for everyone and wait for something to happen.
A better question is simpler: what does my team do every week that's repetitive, rule-based and boring? That list is where AI pays back. The tool comes second.
The three jobs AI does well
Almost every useful AI project in a small business is a version of one of these.
It writes.
First drafts of quotes, proposals, emails and reports in your tone. A person checks and sends.
It reads.
It pulls details out of invoices, contracts, CVs and forms, then puts them where they need to go.
It answers.
It replies to routine questions from customers or staff using your own price lists, policies and FAQs.
What it costs to try
Less than most people expect. If you already use Microsoft 365 or Google Workspace, you probably have access to capable AI tools today. The first project is usually about setting them up properly, not buying something new.
The real cost is time: someone has to describe the task clearly, test the output and decide what “good enough” looks like. That's the part most businesses skip.
If a task can be explained to a new hire in five minutes, it's a good candidate for AI.
Where it goes wrong
- No one owns it. Licences get bought, nobody is responsible for making them useful.
- Client data goes in unchecked. Under POPIA, you need to know where that information ends up.
- It's used for judgement calls. Hiring, pricing strategy and difficult conversations stay human.
A one-week plan to find your first win
- MonAsk each person to list the three tasks they repeat most.
- TueCircle the ones that involve writing, reading documents or answering questions.
- Wed–ThuPick one. Try it with the AI tools you already have, using made-up data.
- FriCompare: was it faster, and was it good enough? Decide whether to keep going.
If that week gives you a clear yes, you've found your first project. If it's a muddle, that's normal too, and it's exactly what a readiness audit is for.
Written by Daine Mawer
Founder of Plaincraft, an independent AI consultancy for South African SMEs, with 17 years of building websites and digital products behind it. AI is part of how I work every day, so I know where it earns its keep and where it doesn't.
More about me
