The right altitude for a small business

Most small-business AI advice is written for companies with engineering teams and venture funding. If that's not you, ignore it. You don't need an 'AI strategy.' You need to take one or two annoying, time-consuming tasks off your plate using tools that already exist. This guide is the unglamorous, useful version.

1. Find your repetitive, low-stakes tasks

The best first AI use case has three properties: it's repetitive (you do it often), it's time-consuming (it eats hours you'd rather spend elsewhere), and it's forgiving (a mistake is annoying, not catastrophic). Look for those.

Common candidates for a small business:

• Drafting first versions of emails, quotes or proposals.
• Summarising long documents, threads or notes.
• Writing and tidying social posts and product descriptions.
• Answering routine, repetitive customer questions.
• Cleaning up and organising messy text data — addresses, lists, notes.
• Turning rough bullet points into polished copy.

Notice these are all drafting and summarising tasks where you review the output. That's exactly where AI is most reliable and lowest-risk today.

2. Use off-the-shelf tools first

Resist the urge to build something. For the vast majority of small businesses, the right answer is an existing product:

• A good general AI assistant handles drafting, summarising and brainstorming.
• Many tools you already pay for — email, documents, your CRM, design apps — now have AI features built in. Use those before adding anything new.
• For specific jobs (transcription, image editing, scheduling) there are focused tools that do one thing well.

Building custom AI makes sense only when you have a genuinely unique, repeated need that no product serves — and that's rarer than vendors would have you believe.

3. Start with one task and learn it properly

Don't roll out AI across the whole business at once. Pick a single task, use it daily for a couple of weeks, and learn:

• How to instruct it to get good results (clear, specific requests — see any prompting guide).
• Where it's reliable and where it slips up.
• How much time it actually saves once you account for reviewing its output.

This builds real intuition cheaply. Once one task is working smoothly, add the next. Compounding small wins beats a grand rollout that overwhelms everyone and gets abandoned.

4. Always keep a human in the loop for anything that matters

The single most important habit: review AI output before it goes anywhere consequential. AI tools are excellent drafters and unreliable final authorities. They can state wrong facts confidently, miss context, or strike the wrong tone. For anything customer-facing, contractual, financial or reputational, a human reads it before it ships. Treat AI as a fast junior assistant whose work you always check — not an autopilot.

This isn't a temporary precaution; it's the operating model. The time you save is in the drafting, not in skipping the review.

5. Mind privacy and your data

A few sensible guardrails:

• Don't paste genuinely sensitive information — customer personal data, confidential contracts, passwords — into tools without understanding how that data is handled. Check the tool's data policy.
• Be cautious with anything regulated — health, financial or legal data carries obligations regardless of the tool.
• When in doubt, anonymise or generalise the input. You can often get the same useful output without the sensitive specifics.

6. Measure whether it's actually helping

After a few weeks, ask the honest question: is this saving real time or just feeling modern? Look at the actual hours reclaimed and the quality of output after your review. Keep what genuinely helps; drop what doesn't. The goal is leverage on your real work, not adopting AI for its own sake.

The realistic promise

Used this way, AI won't transform your business overnight, and anyone promising that is selling something. What it will do — reliably, this month, for almost no cost — is take a handful of repetitive drafting and summarising chores off your plate, freeing hours for the work only you can do. Start with one task, keep a human checking the output, and let the wins compound. That's the whole playbook.