Practical AI for small business: where to start (and where not to).

Every week, a small business owner asks me some version of: "should we be doing more with AI?" The honest answer is usually yes, but not in the way the headlines suggest. "Doing more with AI" doesn't mean adopting a 12-tool stack. It means looking at the five things you do every week that are repetitive, low-judgement and time-consuming — and removing the friction from one of them.

Start with the boring stuff

The most useful AI work I do with small businesses is almost never glamorous. It's not chatbots, not content factories, not autonomous agents. It's email drafts, meeting notes, document templates, quote follow-ups, simple data tidying. The boring stuff. The reason: the boring stuff is high-frequency, low-stakes, and easy to verify. That's exactly the shape of a problem AI is genuinely good at.

Headline-grabbing AI use cases tend to be high-stakes and hard to verify. Those are exactly the problems where AI underperforms quietly — and the failure mode is invisible until it isn't.

The three workflows worth starting with

If you want a sensible starting map, I'd suggest these three — roughly in order of effort to value:

1. First-draft email and document responses

Common, repetitive replies — quote requests, booking confirmations, FAQ responses — are an easy win. Use AI to generate a first draft based on a template plus the specific request. You still send it. But you spend two minutes editing instead of fifteen drafting from scratch. Saves hours per week. Low risk.

2. Meeting and call summaries

If you talk to clients regularly, recording (with consent) and auto-summarising your calls is one of the highest-leverage moves available right now. The follow-up email becomes 90% faster, action items don't get lost, and you build a searchable history of what was actually agreed. Tools like Otter, Fathom and Granola handle this well for small teams.

3. Internal "what did we decide?" search

If your team uses Google Workspace or Microsoft 365, a basic AI assistant over your own documents will save more time than any productivity hack you've tried this decade. The question "where's that file from last March?" becomes a sentence in plain English. This is genuinely transformative for businesses with more than two or three people.

Where AI is the wrong tool

There are a few use cases I steer clients away from until they have more experience and tighter guardrails:

• Public-facing chatbots without supervision. The downside (an AI making confident promises on your behalf) is bigger than the upside. Start internal.
• Numerical reasoning. AI is unreliable with arithmetic, unit conversions and structured calculations. Use real software for real maths.
• Legal, medical, financial advice. The fluency of the output makes it easy to mistake confidence for correctness. Don't.
• Bulk content generation for SEO. Search engines are getting better at detecting it, and your brand is getting worse for being associated with it.

The two questions to ask before any AI rollout

Whenever a client wants to introduce a new AI workflow, I ask two questions before anything else:

1. What happens when it gets it wrong? If the failure mode is "Tim sends a slightly awkward email," fine. If it's "we promise a customer something we can't deliver," that's a different conversation.
2. How will we know it's wrong? A workflow you can't audit is a workflow you can't trust. Every AI step should have a human checkpoint or an obvious signal of failure.

Those two questions filter out about 80% of bad ideas before they cost anyone anything.

What "doing AI well" actually looks like

A year from now, the businesses that benefit most from AI won't be the ones with the most tools. They'll be the ones where staff have quietly built confidence using AI on small, well-defined tasks — and have started to recognise where it helps and where it doesn't. That experience compounds. Toolchain choices don't.

So if you take one thing from this: pick one repetitive, low-risk task this month. Improve it with AI assistance. Pay attention to where it surprises you (good or bad). Then pick the next one. That's the unglamorous path, and it's the one that actually works.