AI for a small business: what it does and where to start

01

What is AI, in plain English?

AI (artificial intelligence) is software that learns patterns from large amounts of data and uses them to produce outputs such as text, images, predictions or recommendations. It follows no fixed script for each task, which is why it handles varied requests and also why it can be wrong.

The UK government's AI Playbook uses the OECD definition: a machine-based system that works out, from its input, how to generate outputs such as predictions, content, recommendations or decisions. People now usually mean a chatbot such as ChatGPT, Copilot or Gemini, one kind of AI.

02

How does a chatbot like ChatGPT produce its answers?

A large language model is trained on huge amounts of text, then predicts which words are most likely to fit your prompt. It writes fluent answers without checking them against a database of facts, so it can sound sure while being wrong.

The AI Playbook describes large language models as large neural networks trained on text. Given a prompt, one selects the response most likely to fit. It adds that they are not domain experts and not a substitute for professional advice.

Where a tool was built matters too. The British Chambers of Commerce (BCC) warns that models developed for other markets may give answers that do not fit UK rules, such as a dismissal process described under US law.

03

What is a hallucination, and why does it matter?

A hallucination is an AI answer that sounds plausible but is false or invented, such as a case, price or policy that does not exist. It matters because the tool delivers it as confidently as a correct answer.

The AI Playbook lists hallucination as a known limitation. In June 2025 the Divisional Court dealt with legal work containing made-up case citations. The judges said freely available tools such as ChatGPT are not capable of reliable legal research and may cite sources that do not exist. The critical safeguard, they said, is checking output against an authoritative source.

Hallucinations are common enough to plan for. In the Department for Business and Trade's Copilot pilot, 22% of survey respondents said they identified hallucinations, 43% said they did not and 11% were unsure.

04

What is the difference between AI, automation and an AI agent?

Automation follows fixed rules that you set. AI generates content or predictions from patterns it has learned. An agent is AI that can decide and act with limited human involvement. The more freedom the tool has, the more checking it needs.

The AI Playbook describes agentic AI as autonomous systems that can make decisions and perform actions with minimal human intervention, and advises caution about significant decisions.

Agents are still the least common type. In the Department for Science, Innovation and Technology (DSIT) study of businesses with five or more employees, 7% of AI adopters used agentic AI, against 85% using text generation.

05

What can AI do well for a small business, and what does it do badly?

AI is good at first drafts, summaries, rewording, sorting and answering routine questions from material you supply. It is poor at recalling facts reliably, professional judgement, and knowing your business unless you tell it.

Official data fits this picture. In the DSIT study, 85% of adopters used text generation and language processing. In the Department for Business and Trade's Copilot pilot, written tasks showed the largest time savings, while scheduling and generating images took extra time.

06

What does it cost to use AI in a small business?

Costs range from free chatbots to monthly plans per user, add-ons to software you already pay for, usage-based charges and custom builds. Prices change often, so check the supplier's page, and count staff time as a real cost.

We have not listed prices because they differ by supplier and change often. The bigger cost is usually people's time. In the DSIT study, the barriers cited most often were no identified need for AI (71%) and limited AI skills or knowledge (60%). The BCC adds that smaller firms often struggle with the cost and complexity of implementation.

07

Is it safe to put customer or business information into an AI tool?

Not without checking. The NCSC advises not to include sensitive information in queries to public AI tools, because the provider can see and store what you type. Use the least information needed and read the terms first.

The NCSC explains that queries to a public large language model are visible to the organisation that runs it, which stores them and will likely use them to develop its services. Stored queries could be hacked, leaked or made public. The blog dates from March 2023, so check current advice. You can often get what you need without names or account numbers.

08

What does UK data protection law say about using AI?

UK GDPR applies whenever personal data is involved, whether or not AI is used. You need a lawful basis, must keep people informed and their data accurate and secure, and still answer to the ICO. Using a supplier's tool does not remove those duties.

The Information Commissioner's Office (ICO) publishes worked answers. For a business using generative AI to draft client email replies that staff review, it says the organisation must show the tool is necessary and proportionate if it relies on legitimate interests, or may use consent, and that special category data needs an extra condition.

For a meeting-notes tool, the ICO says existing lawful bases can continue if the tool is not used for new purposes or to help make decisions about people, and the provider does not use the data for its own purposes. International transfer rules apply to providers outside the UK.

Decisions made only by software are separate. The Data (Use and Access) Act 2025 created a more permissive framework, but people must still be told about significant decisions, be able to challenge them and obtain human intervention. Check current ICO guidance, as changes commence in stages.

09

How do I start safely this week, and when is outside help worth it?

Pick one low-risk, repetitive task, write down what may not be entered, keep a person checking every output, and measure time saved and mistakes found over two weeks. Expand only if the result is clearly better.

1. 01
   Choose one task

   Pick something frequent, low-stakes and text-based, such as routine reply drafts. Avoid decisions about people.

2. 02
   Write the rules

   Set what may never be entered, who checks output, and which tool staff may use.

3. 03
   Start with non-sensitive materialFull step in the download ↓
4. 04
   Check every outputFull step in the download ↓
5. 05
   Time it honestlyFull step in the download ↓
6. 06
   Review after two weeksFull step in the download ↓

Outside help is worth considering when AI must connect to your own systems, handles customer data at scale, or nobody has time to test it. Reeve Consult offers AI consultancy that starts with a specific workflow. It promises no particular saving.