Guides · AI for small business

How to use AI in a small business: a plain-English start

No hype and no jargon. The handful of jobs AI is genuinely good at, the ones it's still rubbish at, the two-hour exercise that shows you where to start, and how to ask so you stop getting waffle back.

By Paul Rudland · AI Confidence Coach · Suffolk

Most small businesses don't have an AI problem. They have a where-do-my-hours-go problem — and AI happens to be the cheapest tool ever invented for fixing it. So this doesn't start with the technology. It starts with your week.

Everything below can be done by one person with an ordinary laptop, a paid AI subscription costing about £17 a month, and an afternoon. No procurement, no consultants, no integration project.

Start with the boring job, not the interesting one

The instinct is to point AI at something clever. Don't. Point it at the boring, repetitive, weekly thing nobody wants to do — the one that always slips to Friday evening. That's where the money is. Interesting problems make good conference talks; boring problems make good margins.

The two-hour audit

You can't automate what you haven't named. Before you touch a single tool, do this. It's the highest-return exercise here and the one most people skip.

List the week. Write down every recurring task in your business — not projects, tasks. "Chase overdue invoices." "Write the job advert." "Reply to enquiry emails." Aim for thirty lines. Most owners get to forty and are quietly horrified.

Score each one. Give every task a mark out of five for how often it happens and how similar it is each time. The tasks that score high on both — frequent and samey — are your candidates.

Be honest about each candidate. Write one of three words next to it:

• Delete — nobody's read this report in a year. Automating it would be automating waste.
• Delegate — a person could do this; the only reason they don't is nobody's written it down. That's a checklist, not an AI job.
• Draft — the work is genuinely yours, but ninety per cent of it is getting to a first version. This is AI's territory, and it's enormous.

Only the Draft pile goes forward. If thirty tasks yield six, that's a good result — six jobs done in a quarter of the time is roughly a day a week back.

The jobs AI is genuinely good at

For a business of five to fifty people, ranked roughly by return — and all of it needs no technical work whatsoever:

1. First drafts of anything written — proposals, adverts, listings, policies, tender responses, letters. Three to six hours a week for most people.
2. Turning a call or meeting into notes, actions and a follow-up email — before the kettle's boiled.
3. Reading long documents and answering questions about them — contracts, leases, surveys, reports.
4. Triaging and drafting replies to inbound enquiries — so a lead never sits unanswered overnight.
5. Writing the checklist, SOP or induction pack that lives in someone's head.
6. Interrogating your own numbers — ask questions of a sales export instead of building a pivot table.

And the jobs it's still rubbish at

Just as important, so you stop wasting evenings on the wrong ones. Never hand AI the final decision about a person — hiring, dismissal, credit, a tenancy. Never trust it with your specific numbers from memory (paste the data in — it cannot recall your revenue and won't tell you it's guessing unless you ask). And never let it touch the relationship: nobody ever renewed a contract because the follow-up email was well structured. Automate the admin around the relationship, never the relationship itself.

How to ask, so you stop getting waffle

Most disappointment with AI is a briefing failure, not a capability failure. You wouldn't expect a sharp new graduate to do good work from "write something about our company." Brief it the way you'd brief them on their first morning — they're quick, well-read, eager, and know nothing about you.

Give it four things: context (who you are, who it's for, what you're trying to achieve), limits (length, tone, what to avoid), an example (one thing you've written that you like — this single move improves the output more than any other), and a specific ask ("a 300-word advert", not "help with recruitment").

One rule before you scale it up

Your first three wins should be things that, if the output is wrong, cost you an apology rather than a client, a fine, or a person. Confidence is built on small, safe wins. Start there and earn your way up to the risky stuff. And before anyone types anything, agree in writing which tools you use, on which paid plans, and what must never be pasted in — personal data, anything under NDA, client information on a consumer account.

Based in Suffolk, working across the East of England. More on AI training near you →