For a small business, artificial intelligence is most useful on tasks that are repetitive, language-heavy and tolerant of review: drafting and summarising text, answering common questions, categorising information and spotting patterns in data you already hold. Where accuracy is critical and nobody checks the output, it creates work rather than saving it.
Start from a task you already do badly or slowly.
Where it genuinely helps
Drafting routine written material such as quotes, descriptions and replies, which you then edit. Summarising long documents. Categorising and extracting information from unstructured text. Answering common customer questions as a first line, with a route to a person.
The pattern is that it produces a first version quickly and a human improves it. Used that way it saves real time on work that previously sat undone.
Where it fails
Anything requiring accuracy you cannot verify. These systems produce confident, plausible text that is sometimes wrong, and in professional, legal, medical or financial contexts that is a serious liability rather than an inconvenience.
They also reflect whatever they were trained on, which means output can be inappropriate or subtly misleading. Nothing should go to a customer unreviewed.
Responsibility does not transfer
If a tool gives a customer wrong information, mishandles personal data or produces a discriminatory outcome, that remains your responsibility. A vendor failure is not a defence to a regulator or a customer.
Be careful what you put into these systems. Customer information, contracts and confidential material entered into a third-party service may be stored or used further, and that carries obligations under the Protection of Personal Information Act, overseen by the Information Regulator.
Judge it on time saved, not novelty
Pick one task, use it for a month, and measure whether it saved hours or created review work. Most small businesses need no dedicated system, because the capability increasingly arrives inside software they already pay for.
Free business training covering digital tools is available through the Small Enterprise Development and Finance Agency, and any claims you make about AI in your marketing must be substantiable.
Frequently asked questions
What is AI genuinely useful for in a small business?
Drafting and summarising text, answering common questions as a first line, and categorising or extracting information you already hold.
Where does it fail?
Where accuracy matters and nobody checks. It produces confident text that is sometimes wrong, which is a liability in professional contexts.
Who is responsible if it gets something wrong?
You are. A vendor failure is not a defence to a customer or a regulator.
Is it safe to enter customer data?
Be cautious. Information entered into a third-party service may be stored or reused, which carries data protection obligations.
How should I judge whether it is worth it?
Use it on one task for a month and measure hours saved against review work created.
Further reading
Originally published in January 2018. Updated September 2026 into guidance on where AI is useful in a small business and where it is not.
