AI for small business: the uses and automations that really save time

AI for small business is worth paying for when it takes a task you already do every week off your hands: the first draft of an email, the research behind a web page, the admin nobody enjoys, the alert that tells you a prospect is waiting. It is a waste of money when you buy the tools first and look for a use afterwards.

I use AI every day in my work. Since it arrived, my delivery times have shrunk, and the time saved sometimes lets me look for extra solutions for a client and build something more complete than before. I also build software for my own use, not to sell, so I know what it is like to need something and build the solution yourself. It is satisfying.

This guide is the hub of the blog’s AI and automation section. It covers what AI can realistically do for a business of one to twenty people, where to start, what it costs at the time of writing (October 2026) and the rules on client data, with a link to the detailed guide at each step. If you want ChatGPT or Gemini to recommend your business, that is a different subject, covered in the guide to generative engine optimisation.

What AI for small business looks like in practice

Most of the time saved comes from five kinds of work. The table shows what AI or automation does in each, and what you need to start.

What is missing from the table matters as much: your prices, your strategy and the conversation with an unhappy client. AI can help you prepare for each of them, but the decision stays with you, and so does the responsibility for anything it gets wrong.

Start with one task you already do every day

My strongest advice on AI is the same as on marketing. It is easy to fall for every new shiny tool, and AI brings a new one every week; the noise is the point. What pays is improving the work you already do, which is why my rule is to automate a task you do every day, not one you might do some day. The marketing side of this is in my piece on shiny object syndrome.

Here is the method I suggest:

1. For one week, note every task you repeat that takes more than ten minutes: answering the same questions, writing quotes, copying enquiries into a spreadsheet, chasing call-backs.
2. Pick the one you do most often and can check easily. A first draft you will read anyway is a safe start; an email that goes to a client unread is not.
3. Time it once by hand, then once with AI, and write both figures down.
4. Use one tool for that task for a month, then decide: keep it, adjust it or drop it.
5. Only then add a second task or a second tool.

Lifetime deals are the classic trap. I have bought on AppSumo, but it is a bit like a casino: everything looks cheap, so you end up spending a lot, not everything is worth it, and some tools eventually stop working. My AppSumo review explains how to buy there without regrets.

None of this needs an agency. A free account with an assistant and an hour of your time are enough to test the first task.

Writing and research: where I use AI every day

I use Claude for writing and research. In my experience, ChatGPT tends to make things up a bit too often, while Claude, and its Fable 5.1 model in particular, has been extremely accurate. That is my experience, not a benchmark, so test both on your own work before paying: my Claude vs ChatGPT comparison shows how.

For copywriting and SEO, I use deep research to find what people complain about most in a niche, which is exactly what good copy answers, and which keywords keep coming back. Then I check those keywords in Semrush before writing a word. The method is in how to use Deep Research for copywriting and SEO.

On a website project, AI is where I bring the visual identity, the layout and the copy together, then rework and edit until I am satisfied, before the client sees anything. The catch is that a first draft from AI sounds like everyone else’s. What makes copy sound like you is information nobody else has: your customers’ words, your prices, the way you work. I explain how I brief and edit in how to make AI writing sound human.

AI can also build the website itself. I usually steer solo owners away from WordPress and Webflow, because both take a lot of learning; for a simple brochure site, making it with AI and hosting it on GitHub Pages is easier. The walkthrough is in how to build a website with AI.

Admin: the quiet time-saver

Admin is where a chat assistant saves time without anyone noticing. Typical uses:

• summarise a 30-page contract so you know which clauses to read closely;
• turn a voice note from a site visit into a to-do list;
• draft a quote from your notes and your price list;
• write a spreadsheet formula from a plain-English description;
• translate a client’s email, or your reply.

Claude can also create documents, spreadsheets and slides directly in the chat. Read what comes back, though: a summary can drop the one clause that matters, and a formula can be wrong in a way that looks right.

Which AI assistant to pay for, and what it costs

Both main assistants have a free plan that is enough to test a task. You pay when you hit the limits every week or need a feature the free plan lacks. At the time of writing (October 2026):

These are US prices. Both companies charge in local currency in many countries, with tax added or included depending on where you are, so check the amount in £ at checkout before you subscribe. Current prices are on Claude’s pricing page and ChatGPT’s pricing page.

The model I rate most, Fable 5.1, is Anthropic’s most capable. On Pro it is available through usage credits, billed at API rates on top of the subscription; the Max plans, from $100 a month, include it within weekly limits. Claude has no ads, and Anthropic said in February 2026 that it will stay that way. ChatGPT is the better choice if you want images made in the chat: Claude’s feature list does not include image generation.

Before paying for anything, check what you already have:

• Google Workspace: Google has bundled Gemini into its business plans since January 2025, and the entry plan gets fewer AI features than the higher ones.
• Microsoft 365: Microsoft’s assistant is Copilot; check what your plan includes before you buy a second assistant.
• European options: Mistral AI, a French company, makes the assistant Le Chat, and Infomaniak, whose services I recommend, says its assistant Euria processes and stores requests in its Swiss data centres without using them to train models.

Automations: let the routine run without you

An automation is a rule that does a task when something happens: a form is sent, so the enquiry lands in a spreadsheet and you get an email. Unlike a chat assistant, it runs without you, every time, and plenty of useful automations involve no AI at all.

The one to set up first is the lead alert, because speed wins clients. In a study of 1.25 million leads published in the Harvard Business Review, firms that tried to contact a lead within an hour were nearly seven times as likely to qualify it as those that tried even an hour later (The Short Life of Online Sales Leads, 2011). The study is old, but I would still rather call a prospect back in minutes than in hours.

If your enquiries come through a Google Form, this costs nothing: Google Apps Script can email you the prospect’s name and phone number the moment the form is sent, log the enquiry and remind you if it is still marked New an hour later. The tutorial on automating lead follow-up for free gives the script and every step. A free Google account can send emails to 100 recipients a day from a script, which is plenty for alerts.

When the job means passing data between apps that do not talk to each other, such as your website form, your CRM and your invoicing tool, you need a connector tool. The three names you will meet are Zapier, the easiest; Make, which does the same work for less; and n8n, which bills each complete run rather than each step and can be self-hosted for free. Zapier’s free plan covers 100 tasks a month and Make’s 1,000 credits a month. Their prices and limits are compared in Zapier vs Make vs n8n.

AI fits inside an automation as one step: reading an enquiry and sorting it (quote request, complaint, supplier), drafting a reply for you to approve, or pulling the name, address and type of job out of an email. That step calls an AI model through its API, which is the next section.

Two practical points before you build anything. An automation that reads your enquiries needs access to the mailbox they arrive in: it is one of the things I ask clients for when I build one. And every automation can break without telling you, so give each one an owner and check once a month that it still runs.

APIs explained in two paragraphs

An API is the door that lets software talk to an AI model directly, without the chat window. When an automation, a website form or an internal app needs AI, it sends text through the API and gets the answer back. You pay per token, a token being a piece of a word; Anthropic counts a million tokens as roughly 555,000 words. The API is billed separately: a Claude Pro or ChatGPT Plus subscription does not include it.

The small models cost very little. At the time of writing, Anthropic charges $0.10 per million input tokens and $0.50 per million output tokens for Claude Haiku 5.5 (on prompts up to 100,000 tokens), and $4 and $20 for Claude Opus 5.5, according to its API price list. By my calculation, sorting 1,000 enquiries of about 500 words each with Haiku 5.5 costs less than 20 US cents. To get a key, set a spending limit and send a first request, follow the Claude API tutorial or the OpenAI API key guide.

Custom internal tools: when a spreadsheet is no longer enough

Sometimes no off-the-shelf tool fits, and that is when building one makes sense. The client version of this story is Pilotech, an air conditioning installer near Paris. I look after their SEO, cold emailing and websites, and at some point I was bringing them more prospects than the team could handle. They asked me for an app to see everything clearly and never forget to call a prospect back or answer them.

I custom-coded an internal tool that manages prospects, job sites, communication with the employees, the calendar and accounting analytics, with AI and phone integrations. How it came about, and the signs that you need one too, are in bespoke CRM for a small business. Most businesses do not need to go that far: if your process is simply enquiry, quote, won or lost, a spreadsheet with automatic reminders will serve you better.

You do not always need code. Claude’s Artifacts let you build a small interactive tool, such as a quote calculator or a checklist, in any conversation and on any plan, as long as code execution and file creation is switched on in Settings > Capabilities (it is by default). An artifact can even call Claude itself with no API key, each person’s usage counting against their own plan, and anyone you share it with needs a Claude account. AI app builders go further. What is realistic without a developer, and where it stops, is in how to build an app with AI. Before an AI-built app holds real data, check its Supabase or Firebase rules.

Client data, GDPR and the EU AI Act

The rules are simpler than they sound, and they decide which plan you need. This is general information, not legal advice.

Personal data

The moment you paste a client’s name, email address or phone number into an AI tool, data protection law applies: the GDPR in France, the UK GDPR in the UK. Three habits cover most cases:

1. On a personal plan, switch off model training: in ChatGPT, open Settings > Data controls and turn off Improve the model for everyone; in Claude, make the same choice in the privacy settings.
2. Anonymise what you can. A complaint with the name removed is as useful to the AI as one with the name.
3. Keep clients’ personal data for business plans and the API, which are not used for training by default. OpenAI signs a data processing agreement for ChatGPT Business, Enterprise and its API on request, and Anthropic’s commercial terms include its data processing addendum.

What each plan allows, in detail, is in ChatGPT and GDPR: what client data you can share.

AI literacy under the EU AI Act

Article 4 of the EU AI Act, on AI literacy, has applied since 2 February 2025 to businesses that use AI at work. An amendment that came into force in July 2026, Regulation (EU) 2026/1744, softened it: you must take measures to support the AI literacy of your staff and of anyone using AI on your behalf, but no particular level is required, and the text imposes no course or certificate. For a small team, a sensible start is a one-page AI rule that you go through with everyone: which tools are approved, what data never goes in, and that a person checks every output before it reaches a client. The European Commission’s page on AI talent, skills and literacy is the official starting point. Fraudsters use AI too: my guide to voice clones and deepfake calls sets out the rule that protects your team.

The AI Act is EU law. A UK business is concerned mainly when it serves the EU market, for example when the output of its AI tools is used for clients in France; at home, the UK GDPR is what governs client data in AI tools.

Your first 30 days with AI: a checklist

1. Week 1: list the tasks you repeat every week and pick one you can check easily.
2. Open free accounts with Claude and ChatGPT, switch off model training on both, and run the same task through each.
3. Week 2: write your one-page AI rule and share it with everyone who works with you.
4. Set up one automation, starting with the lead alert if enquiries come through your website.
5. Weeks 3 and 4: use the tool daily and note the time it saves.
6. Day 30: keep, adjust or drop. Pay for a plan only if you hit the free limits every week.

When to hand it over

You can do everything above yourself. Bring in a developer when a tool stores clients’ personal data, connects to your phone system or your accounts, or would bring your work to a halt if it broke. If you commission one, expect to pay a deposit: I ask for 30% or 50% upfront and the rest at delivery.

For the rest of my daily toolkit, from design to email verification, see the small business marketing tools I use every day.