The Complete Guide to Using Claude in Your Small UK Business

Claude is Anthropic's AI assistant, and for a UK small business it is at its most useful when you stop thinking of it as a chatbot and start treating it as a member of the team that you brief properly. That means giving it your business context in a `.md` file, packaging repeat tasks as Skills, connecting it to Gmail, Calendar and Drive, and — when you are ready — running it on a schedule to draft replies, summarise your week and refresh content while you sleep.

A quick word before we go any further

The AI landscape moves fast — sometimes weekly. Specific model names, exact usage limits and connector lists in this guide are correct as of September 2026, but the framework and workflows we describe will still apply in six months' time. When Anthropic renames a model or adds a new connector, the underlying approach — write your context down once, package repeat work as Skills, connect the tools you already use — does not change.

If you are still deciding *whether* to use Claude versus ChatGPT or Gemini, read our comparison first: ChatGPT vs Claude vs Gemini: which AI for UK small business. This post assumes you have already decided Claude is one of the tools you want to use, and now you want to use it *properly*.

Which Claude plan do you actually need?

For most UK small business owners, Claude Pro at $20 a month (roughly £19 including VAT) is the right starting point. The free tier is genuinely useful for casual work, and Max at $100 or $200 a month is only worth it if you use Claude Code heavily or run agents through the day. Team plans start at £20 per seat per month (annual billing) and are worth considering the moment you are sharing a single Pro login with anyone else.

Here is what each plan actually gets you as of September 2026:

Prices and included features are Anthropic's current published pricing on the Claude pricing page — check there before buying, because Anthropic revises this fairly often.

Our honest recommendation for a solo UK small business: start on Free for a week to get a feel for the writing style, upgrade to Pro as soon as you hit the free tier's message caps or want to build a Skill, only look at Max once you are actively hitting Pro's limits (do not buy it "just in case"), and move to Team the second person joins — sharing a Pro login is against Anthropic's terms and you lose per-user memory and Projects.

Data handling caveat that matters for UK businesses: the consumer Pro and Max plans are not designed for regulated personal client data (health, financial, legal case information). For that, you need Team, Enterprise, or Claude running via AWS Bedrock or Google Vertex in an EU/UK region with a proper Data Processing Addendum. If you are a solicitor, accountant, healthcare provider or anything similar, do not skip this — it is a genuine GDPR issue. We cover the shape of that decision in the ChatGPT vs Claude vs Gemini comparison.

How do you get your business context into Claude?

The single biggest thing that separates people who get magic from Claude and people who find it "quite good, I suppose" is whether they have written down their business context and given it to Claude to read. Everything else in this guide is built on top of that habit.

The tool for this is a plain markdown file — usually called something like `business-context.md` — that describes your business the way you would brief a new employee on their first day. You upload it once into a Claude Project, and every conversation you start inside that Project has that context loaded automatically. No more re-explaining who you are and what you do at the top of every chat.

What to put in your business-context.md file

Aim for 800–2,000 words. Structure it with clear H2 sections:

• About the business — what you sell, who buys it, how you make money, roughly how many customers you have
• Tone of voice — three or four adjectives, two or three example sentences that "sound like us", and a "we never say X" list. That list alone raises quality more than any other single instruction
• Products and services — a bullet list with price, what is included, and one line on who each is for
• Ideal customer — a short pen-portrait ("small independent café owner, 35–55, cares about quality suppliers, hates jargon, budget-conscious")
• Common questions and standard answers — the ten questions you get asked most often. This turns Claude into a passable version of your first-line support in one file
• Things we do not do — services or clients you turn away. Prevents Claude from confidently offering things you cannot deliver
• Useful links — your website, pricing page, key resources

How to load it into a Project

In Claude on the web, open the sidebar, click Projects, then New project. Paste your context file into Project instructions or upload it as a file into the Project's knowledge base. Add any other reference documents — your price list, an example proposal, your standard reply templates. Then start every business-related chat *inside* that Project.

Anthropic's Projects documentation covers the mechanical detail — up to 5 Projects on the free tier, unlimited on Pro and above, with paid plans automatically enabling retrieval so your knowledge base can be far larger than the raw context window allows.

Callout: how SME Shack uses this ourselves. Our own workflow is built around this exact pattern. We keep a `CLAUDE.md` file at the root of every project — one for the website, one for internal ops, one for each client — capturing the brand, tone, technical stack and rules. Claude Code reads it automatically at the start of every session, so when we write a blog post like this one, Claude already knows we are a UK agency, we write in British English, we do not use hype language, and every post follows our AEO structure template. That is the difference between three drafts and one.

What are Claude Skills, and why do they matter?

Claude Skills are packaged, reusable instructions that Claude loads automatically when a task matches. Think of them as "recipes" — instead of pasting the same 400-word prompt every time you write a newsletter, you package that prompt (plus templates, examples, and any reference files) as a Skill once, and Claude picks it up whenever you say "write this week's newsletter". They live as folders on disk (or as uploaded zips in claude.ai) and are technically just a `SKILL.md` file with a bit of YAML frontmatter.

Skills are documented in full on Anthropic's Agent Skills documentation. The important idea to grasp is progressive disclosure — Claude only loads the name and description of every Skill at startup (about 100 tokens each), then only pulls in the full instructions if your request matches. So you can have 30 Skills installed with no meaningful cost — Claude just uses the right one when it needs it.

Why Skills beat one-off prompts: the newsletter you send in three weeks follows the same structure as this week's; if someone else joins the business they can trigger your Skills without knowing your prompt engineering; every time you tweak the instructions to fix an issue, every future run inherits the fix; and a Skill can bundle template files, examples, and reference documents that Claude only reads when needed.

A Skill you should build in your first week

Here is a realistic first Skill for a small UK service business — "draft-customer-reply":

``` draft-customer-reply/ SKILL.md examples/ good-reply-refund.md good-reply-quote-request.md good-reply-complaint.md ```

Inside `SKILL.md`:

```

name: draft-customer-reply description: Draft a reply to a customer email in our house tone. Use when the user pastes a customer email and asks for a response, or says "draft a reply".

When drafting a reply to a customer email:

1. Read the customer's email carefully — identify what they actually want, not just what they said
2. Match our house tone: warm, plain-spoken, British English, no exclamation marks
3. Start with a one-sentence acknowledgement of their situation before answering
4. Answer the actual question in 1–3 short paragraphs
5. End with a clear next step or question
6. Sign off "Best, [name]" — do not add job title
7. Never promise something we do not offer (see business-context.md)
8. If the email involves a refund, complaint, or legal question, mark the draft as REQUIRES REVIEW at the top

See the /examples folder for the tone we aim for. ```

That is a working Skill. Zip the folder, upload it via Settings > Capabilities > Skills in claude.ai (Pro plan and above), and from now on, every time you paste a customer email into a Project chat, Claude will draft a reply in your voice with the safeguards built in.

Other Skills a UK small business should build early: a weekly newsletter Skill (structure, length, subject line rules), a blog post outline Skill (your structure, tone, SEO rules), a quote/proposal Skill (standard sections, pricing bands, terms), a job advert Skill, and a meeting notes Skill. Start with two. Keep them tight. Improve them monthly. Within three months you will have a small library that saves you an hour or two every week.

What is Claude Code, and do you need it?

Claude Code is Anthropic's AI coding tool that lives in your terminal, not in a browser tab. It reads your files, edits them, runs commands, uses git, and generally works like a very fast junior engineer that never gets tired. It is included with Claude Pro and Max plans, and is documented at claude.com/product/claude-code.

When Claude Code matters for a small business owner. For most non-technical owners it does not — at least not directly. It is a developer tool. But there are three cases where it is worth the learning curve even if you are not technical: (1) you run a website and want to maintain it yourself — small copy changes, new pages, pricing tweaks — on a plain-text codebase like a Next.js site; (2) you want to build simple internal tools like a quote calculator or a data cleanup script from a plain-English description; (3) you want to run scheduled automations locally, which use Claude Code as the runner.

Our honest take: if you have never used a terminal, do not start with Claude Code. Get comfortable with the web app first. Once you find yourself thinking "I wish Claude could edit my website directly" or "I wish this ran every morning by itself", that is the signal Claude Code is worth learning. Anthropic's Claude Code docs are a good starting point.

Second callout on our own use: SME Shack builds and maintains almost every one of our client sites through Claude Code. Every blog post on our site — including this one — was drafted with Claude Code invoking a Skill that knows our house style, SEO rules, and publishing workflow. It is the operational spine.

Which Claude connectors matter for a small business?

Connectors are official integrations that let Claude read from and write to the tools you already use. Claude currently ships with connectors for Google Workspace (Gmail, Google Calendar, Google Drive), Microsoft 365, GitHub, 1Password, and various third-party services, plus the ability to add custom connectors through MCP (Model Context Protocol). All the Google connectors are available on the free tier — you do not need to pay to try them. Anthropic's Google Workspace connectors documentation has the full setup guide.

What each Google connector actually does

• Gmail — Claude can search your inbox in natural language ("show me all emails from Sarah about the March invoice"), draft replies, and send them after you approve each send. It never sends without asking
• Google Calendar — read your calendar, create events, find free time in the week, RSVP to invites, set up recurring meetings, add Google Meet links automatically
• Google Drive — search files, read Docs, Sheets, Slides and PDFs, and save Claude-generated files back to your Drive as new documents

Real small business examples. Gmail: "Look at my inbox for anything overnight. Sort into quote requests, customer questions, spam, other. For the quote requests, draft a friendly reply asking for the details we need to price it. Do not send — just draft." Calendar: "Look at next week — how many meetings, any back-to-back with no prep gap, what I appear to be missing prep for, and my longest free block." Drive: "Read our proposal template in Drive plus this client brief. Draft a filled-in proposal, save it back to Drive in /Proposals as a new Doc."

Microsoft 365 and other connectors. If you are on Microsoft 365 rather than Google Workspace, the Microsoft 365 connector gives Claude the same shape of access to Outlook, Word, Excel and OneDrive. GitHub is useful if you develop software or manage a website through a repo. 1Password is useful for anyone whose day involves looking up credentials. Anthropic also publishes a Connectors Directory of third-party integrations, and you can add custom connectors through the Model Context Protocol standard.

Two rules that matter. Read every action before approving it — Claude is very good, not infallible. And on the consumer Pro plan, do not connect accounts containing regulated client data. For law firms, accountants and healthcare providers, that means keeping client-matter email out of consumer-tier connectors and using Team or Enterprise with the right data handling agreement.

Can Claude do things on a schedule, without me asking?

Yes. As of 2026, Claude supports scheduled tasks — jobs that run automatically on a cron-style schedule, do their work, and hand you the result. These are what turn Claude from "a tool I open" into "a colleague who works in the background". There are three ways to run them, and the right one depends on how technical you want to get.

Option 1: Claude Code /loop. From inside a Claude Code session, type `/loop 30m` followed by a prompt and Claude re-runs it every 30 minutes for as long as the session is open. Handy for babysitting a job, useless for anything that needs to survive your laptop closing.

Option 2: Desktop scheduled tasks. Claude Code on desktop can register tasks that fire on a real cron schedule — every weekday at 9 AM, every hour between 9 and 5. Each run starts a fresh session with full access to your files, Skills and connectors. Runs on your machine, so it needs to be on (or a small always-on machine like a Mac mini or a cheap cloud VM).

Option 3: Cloud Routines. Anthropic's newest option, introduced in April 2026 (still in research preview). A Routine packages a prompt, repositories and connectors into a reusable automation that runs on Anthropic's servers. Triggered by schedule, API call, or GitHub webhook, with daily run caps of 5–25 depending on plan tier. Survives your laptop being off. Details on the Claude Code scheduled tasks docs.

Three scheduled jobs a small business should consider

1. Hourly Gmail triage (weekdays 9–5). Every hour, checks the inbox for anything new. Categorises by intent (new enquiry / existing customer / spam / noise), drafts a reply to any new enquiry using your standard first-response Skill, and messages you a one-line summary: "3 new emails, 1 needs your attention: quote request from Cotswold Cakes — draft ready to review". You now respond within an hour without checking email every 15 minutes.

2. Monday morning weekly summary. Runs at 7 AM every Monday. Reads last week's and this week's calendars, your inbox for anything unresolved, and any relevant Drive documents. Emails you what happened last week, what is coming this week, what you have not yet responded to, and what your one biggest priority should be. Ten minutes of your Monday morning, automated.

3. Monthly comparison-post refresh. The one we run for SME Shack. Once a month, a routine takes each of our top 3–5 comparison blog posts (like Square vs SumUp vs Zettle), re-checks every price and feature on each vendor's current pricing page, and if anything has changed, drafts an update pull request for us to review. This is exactly the "refresh top posts every 3–6 months" habit that keeps content ranking, done at scale without us remembering to do it.

Do not build your first scheduled job in your first week with Claude. Get the manual workflow working first, package it as a Skill, run it manually for a month, and only then automate it.

What is Claude bad at?

Any honest guide to Claude has to include the things it is not good at, because assuming AI can do something it cannot is where small businesses lose money.

Real-time information without connectors. Base Claude does not know what the weather is today, what your competitor announced this morning, or what your bank balance is. Web search helps, connectors help more, but do not assume Claude has today's information unless you have connected it to today's information.

Autonomous financial decisions. Do not let Claude approve invoices, transfer money, or make binding commercial commitments unattended. Draft, categorise, and prepare — yes. Send, sign, and pay — no. Always keep a human in the loop for anything with money attached.

True understanding of physical goods. If you run a shop, restaurant, or physical service business, Claude cannot see your stock, smell your kitchen, or hear that the coffee machine is making a funny noise. It is a superb assistant for the digital parts of the business (email, marketing, admin, planning) and useless for the parts that require hands and eyes on the floor.

Regulated client-personal-data work on consumer plans. For solicitors, accountants, medical practices, financial advisers — the consumer Pro and Max plans are not the right home for actual client-matter data. The tool is fine; the plan is wrong. You need Team, Enterprise, or Claude via AWS Bedrock / Google Vertex with a DPA and the right regional data handling. Do not skip this.

Anything that requires being wrong being catastrophic. Claude is very good but occasionally wrong. If a mistake would be catastrophic (a legal filing deadline, a medication dose, a safety-critical calculation), Claude should draft and a human should verify. Never remove the verification step.

High-stakes negotiation and judgement calls. Do not let Claude negotiate a contract, settle a dispute, or handle a genuinely upset customer alone. It is fine at drafting the reply — not at reading the room.

The 5-step "get started with Claude" plan for a solo business owner

If you have got this far and want to actually do this, here is the plan we would give you if you asked us at a pub.

Day 1: Sign up and write your business-context.md. Sign up for Claude Pro at claude.com. Open a new Project called "Everyday work". In a plain text editor, write your `business-context.md` — 800 to 2,000 words on your business, tone, products, customers, common questions, things you do not do. Paste it into the Project's instructions or upload it as a file.

Day 2: Do your real work inside the Project. Do not test Claude with parlour tricks. Do the actual work you would do that day *inside the Project* — draft an email, write a social post, summarise a meeting, plan the week.

Day 3: Connect Gmail and Calendar. Turn them on from Settings > Connectors. Try three things: "summarise emails from this week I have not replied to", "what does my next week look like — anything I am under-prepared for?", and "draft a reply to the [Jane Smith] email". Approve every action; do not let Claude send automatically yet.

Day 4: Build your first Skill. Pick the one email or document you write most often. Build a `SKILL.md` with a good description and clear step-by-step instructions, plus a `/examples` folder with 2–3 examples of good outputs. Zip it, upload it in Settings > Capabilities > Skills. Test it three times, refine after each test.

Day 5: Build your second Skill and start a habit. Add one more Skill. Then commit to two habits: every time you type a long prompt more than once, turn it into a Skill; every time Claude gets something wrong you had to correct, add that correction to the relevant Skill or `business-context.md`.

By the end of week one, you should have: one Project, one context file, two Skills, two active connectors, and about an hour a week of work already off your plate. That is the honest floor for what a solo owner should expect from a first week with Claude. It grows from there.

How can SME Shack help?

If you would rather not spend a week working this out, or you want someone to help you design the workflow properly — including which Skills are worth building, how to structure your context file for your specific business, and where scheduled automations will pay for themselves — that is exactly what our AI consulting service is for. Half a day of proper setup usually saves a solo owner five to ten hours a month, every month.

We also help teams that are further along — designing Skills libraries, setting up scheduled Cloud Routines, and building the small internal tools (quote calculators, onboarding flows, admin dashboards) that Claude Code can now build in an afternoon.

If you are also thinking about how AI search itself will find your business, see how to get ChatGPT to cite your small business — a sister post in this series covering the SEO / AEO side of the same story.

Frequently asked questions

Q: Is Claude Pro at $20 a month really enough for a solo UK business?

A: For almost everyone, yes. Pro gives you unlimited Projects, custom Skills, connectors, longer context windows, and Claude Code. Only upgrade to Max if you use Claude Code heavily throughout the day or run several long agentic sessions. Start on Pro, upgrade only when you actually hit the usage cap in a way that hurts.

Q: How safe is my business data with Claude?

A: For general work — marketing, admin, planning, drafting — Pro is fine. For anything covered by GDPR that involves other people's personal data (client records, health, financial details), you need Team or Enterprise with a proper Data Processing Addendum, or Claude via AWS Bedrock or Google Vertex in an EU/UK region. Do not paste other people's regulated personal data into a consumer Pro account.

Q: Will Claude replace staff in my business?

A: Realistically, no. What it replaces is the boring 30% of many jobs — drafting, summarising, categorising, "get me started" work. The same person can do more meaningful work, or a smaller team can serve more customers. Businesses that assume AI is a direct staff replacement tend to make bad hires (or bad fires); businesses that use it to amplify existing people grow faster.

Q: How long does it take to learn Claude properly?

A: The web app itself: one hour. Business context in and first Skill built: one to two days. Full Skills library and scheduled automations: one to two months of steady use. You get value from day one — you do not need to master everything before it starts helping.

Q: What is the difference between a Project and a Skill?

A: A Project is a workspace with persistent context — "who I am and what documents I care about". A Skill is a packaged recipe for a repeatable task — "how to do this specific thing". Projects set the scene once; Skills run the same job the same way every time. Most useful workflows use both.

Q: Do I need to know how to code to use Claude Code?

A: No, but you do need to be comfortable typing commands into a terminal and installing software from a web page. If neither bothers you, Claude Code can maintain a website, build small tools, and run automations without you writing code — you describe what you want in English, Claude writes it, you review. If terminals put you off, stick to the web app for now.

Q: What if Claude gets something factually wrong?

A: It happens. The mitigation is process: use web search or connectors for factual answers, have a human review before anything goes to a customer, treat output as a first draft. For fast-moving facts (prices, model names, features), always verify against the source before publishing.

Q: Will this guide age badly in six months?

A: The specific model names, usage limits and connector list will drift. The framework — write context down once, package repeat work as Skills, connect the tools you already use, keep a human in the loop, automate only after proving the manual workflow — will not. Before making any decision, re-check current pricing and features on Anthropic's own site.