Where AI actually helps a small business, and where it does not

Most conversations about AI in business are either breathless or dismissive. The useful answer sits in between: AI tools are very good at some specific jobs and poor at others, and the difference is predictable. This guide is a way to decide whether a task in your business is a good fit before you spend money on it.

The short answer

AI language models are good at work that involves reading, sorting, summarising and drafting text, where a person checks the result or where an occasional mistake is cheap to fix. They are a poor fit for tasks that must be right every time, such as calculating prices, giving legal or medical advice, or making decisions about people without review.

Tasks that usually suit AI

• Sorting incoming messages. Reading enquiries or support emails and tagging them by type and urgency, so the right person sees them first.
• Pulling details out of documents. Taking names, dates, order numbers and amounts from emails, forms or PDFs and putting them into a system.
• First drafts. Writing a draft reply, product description or report for a member of staff to edit and send.
• Answering from your own information. An assistant that answers common questions using your FAQs, policies or manuals, and passes anything it is unsure about to a person.
• Summaries. Turning long documents, meeting notes or customer feedback into a short overview.

Tasks that usually do not

• Anything with exact rules. Prices, stock levels, tax and eligibility checks are better handled by ordinary software, which is cheaper and gives the same answer every time.
• Final decisions about people. Hiring, credit or disciplinary decisions need human judgement and, under UK data protection law, often a right to human review.
• Facts the model has to remember. A model can state something wrong with total confidence. If an answer must be correct, it should come from your own data, not the model’s memory.

A simple test for any task

1. Is it mostly text? Emails, documents, notes and descriptions are where AI is strongest.
2. Is it repetitive? A task done dozens of times a week is worth automating; a task done twice a year is not.
3. What does a mistake cost? If a person reviews the output, or an error is easy to spot and fix, the risk is low.
4. Could a simple rule do it? If yes, use the rule. It will be cheaper and more reliable.

A task that passes the first three questions and fails the fourth is a good candidate.

How to try it without a big commitment

Start with a small pilot on one task, using a sample of your real data. Measure how often the AI gets it right compared with how the task is done today, and how much time a person still spends checking. Only then decide whether a full build is worthwhile. A pilot like this is usually a matter of weeks, not months.

Questions to ask about data and cost

• Which AI provider will process our data, and in which country?
• Will our data be used to train the provider’s models? (For business API use, the major providers say no by default, but check the current terms.)
• What exactly is sent to the AI: the whole document, or only the parts needed?
• What will it cost each month at our volume, and what limits stop that cost running away?
• What happens when the AI is unsure, or the provider has an outage?

The bottom line

AI is worth considering for repetitive, text-heavy work where a person stays in the loop. It is not a replacement for clear processes or ordinary software. Pick one task, pilot it on real data, and let the results decide.