AI for small businesses: where to start (and what to avoid)

If you run a small business or a startup, you’re probably being told daily that you “need an AI strategy”. Some of that is genuinely good advice. A lot of it is people selling something. Here’s how I’d think about it.

Start with the problem, not the technology

The best AI projects start with a boring question: “Where do we waste the most time or money on work that’s repetitive and rules-based?” Not “how can we use AI?”.

AI is excellent at tasks involving language and documents: reading, summarising, drafting, sorting, extracting and answering questions. It’s much less reliable at tasks requiring perfect accuracy, real-world judgement or accountability. Keep that in mind and most of the decisions get easier.

Seven sensible places to start

1. Triaging and drafting replies to enquiries and support emails. AI reads incoming messages, categorises them and drafts a response for a person to review and send.
2. Pulling data out of documents. Invoices, contracts, application forms, order confirmations: extracting the key fields into your system instead of re-typing them.
3. Summarising calls and meetings. Turning a recording or transcript into notes and action items, ready to drop into your CRM.
4. An internal “ask the handbook” assistant. Staff ask questions about policies, procedures or product details and get answers drawn from your own documents, with links to the source.
5. Customer-facing question answering. The same idea for customers, on your website or inside your product, for common questions. Needs more care and testing than the internal version.
6. Reporting in plain English. Asking questions of your own data (“which customers haven’t ordered in 90 days?”) without waiting for someone to build a report.
7. First drafts of repetitive content. Product descriptions, proposals, job adverts: drafted by AI, finished and approved by a human.

Notice what these have in common: a human stays in the loop, the task is tedious rather than creative, and the benefit is easy to measure (hours saved, speed of response).

Three traps to avoid

1. Starting with a big, impressive project. A six-month “AI transformation” is a great way to spend a lot and learn very little. Start with something small, prove it works and expand.

2. Treating a demo as a product. It’s easy to build something that looks magical on five examples and falls apart on real data. Always test on your own messy, real-world examples before committing.

3. Automating high-stakes decisions without oversight. Anything involving money, legal commitments, health or people’s livelihoods needs a human reviewing the output. AI can be confidently wrong.

Keeping your data safe

This is the part many businesses skip, and it matters:

• Know what you’re putting where. Free consumer AI tools often have weaker privacy terms than business plans or API access. Check the provider’s terms on how they use and retain your data.
• Be careful with personal data. If you’re handling customers’ or staff members’ personal information, UK GDPR still applies, and you need a lawful basis and appropriate safeguards.
• Write a short staff policy. What’s fine to paste into AI tools, and what isn’t? One page is plenty.
• Choose tools deliberately. The best model for your task isn’t always the most famous one, and you don’t want to be locked into a single vendor.

How do you know it’s working?

Before you start, decide how you’ll measure success: hours saved per week, response times, error rates, customer satisfaction. After a few weeks, compare. If it isn’t clearly better, change it or drop it. That’s not failure; that’s the whole point of starting small.

Sometimes the answer is “you don’t need AI”

I mean that. Plenty of problems are better solved with a clearer process, a spreadsheet or a simple automation between two tools you already use. A good adviser will tell you that, even though it means less work for them.

Where I can help

I help startups and SMEs find the places where AI will genuinely pay off, prototype them quickly on real data and build the good ones properly, with safeguards. That might start with a short AI readiness workshop, or as part of a wider fractional CTO arrangement. If you’d like to talk it through, get in touch.