A UK SME guide to Claude AI, from better prompts to Cowork automation, with the plan, permission and data residency decisions a business should make at each level.
The fastest way to get value from Claude in a business is to work up through six levels of use, from asking good questions to letting it run tasks on its own, and to do it on a business plan with access set deliberately at each step. Most beginner guides stop around level three, and a business has its decisions to make from level four onwards, when Claude starts reading your inbox and your SharePoint.
This guide sets out those six levels in the order most teams adopt them, with the decisions we'd ask a UK business owner to make at each one. It draws on how we use Claude at Inflection Point and on what we see when we roll it out for clients.
What is Claude, and who makes it?
Claude is an AI assistant built by Anthropic, a US AI company. You type requests in plain English and it writes, analyses, summarises and builds things for you, much like ChatGPT. It runs in a web browser and in apps for Mac, Windows, iPhone and Android, and Anthropic's download page confirms that conversations, projects, memory and preferences sync between them.
In practice, Claude use in a business tends to grow through six levels:
- Talking to Claude and shaping its answers.
- Giving it your world: files, images and the live web.
- Making real outputs such as spreadsheets, documents and small web pages.
- Making Claude remember you, through Projects and memory.
- Connecting the apps you already use, and installing skills.
- Letting Claude work while you're away, with Cowork and scheduled tasks.
Levels one to three carry little risk. From level four onwards Claude keeps your business context and can act inside your systems, so the account it runs on and the access it holds start to matter.
Why should a business use a Claude business plan?
Company information belongs on a Claude business plan (Team or Enterprise), because the consumer plans handle your data differently. Plenty of people start on a free account and upgrade to Pro when they hit the usage limit. That's sound advice for someone paying for their own account.
Under Anthropic's consumer terms, people on the Free, Pro and Max plans choose whether their chats and coding sessions are used to train Anthropic's models. If they allow it, Anthropic keeps that data for five years; if they don't, the period is 30 days. Commercial products work differently. Anthropic states that Claude for Work (Team and Enterprise) and the API aren't used for model training by default. On a personal account, each member of staff makes that call for themselves, usually on a sign-up screen.
On a business plan the organisation owns the account and sets the data controls, so company information should only go there.
How common is personal AI use at work?
A Microsoft UK survey run by Censuswide of 2,003 UK employees found that 71% had used unapproved consumer AI tools at work, and 51% did so every week. We've written about shadow AI in small businesses and about staff paying for AI themselves, and the pattern we see in client sites matches those numbers.
I was talking to the managing director of a client business about AI, and we went through what their team was already using. Several people were on personal AI accounts, pasting in client documents and emails, and none of it appeared anywhere the business could see. The MD nearly fell off their chair. Within a few weeks we'd moved everyone onto business accounts, connected Claude to their Microsoft 365 tenant and SharePoint with sensible permissions, and given staff the tool they'd been reaching for anyway, on accounts the business controls.
What does Claude cost in the UK?
Anthropic's UK pricing page, viewed from a UK location, shows these prices including 20% VAT:
Plan Annual billing Monthly billing Free £0 £0 Pro (individual) £15 a month (£180 a year) £18 a month Max (individual) from £90 a month from £90 a month Team, standard seat £18 per seat a month £22.50 per seat a month Team, premium seat £90 per seat a month £113 per seat a month Enterprise Quoted Quoted
Those UK figures come from the Claude pricing page. US articles quote dollar prices before tax, so they won't match these figures. Anthropic changes the page by location and updates prices, so check it on the day you buy. A Team standard seat on annual billing is £15 a month before VAT, which matters when you compare it with Microsoft's ex-VAT pricing further down.
Level one: how do you get useful answers from Claude?
You get useful answers by giving Claude context, a clear task and the format you want. "Write a follow-up email to a client" gets something generic. Explain who the client is, what was agreed in the meeting and what you need from them next, then ask for under 150 words in a friendly, direct tone, and you'll get an email you could send after a quick read. Both answers came from the same model.
The next habit is to reply and refine. Treat the first answer as a draft and correct it in the same chat ("shorter, and put the deadline in the first line") instead of starting again. The same question won't produce the same answer twice, and that's normal for this kind of tool.
Claude offers several models. Anthropic's model settings guide lists Haiku, Sonnet, Opus and Fable, plus an effort setting that controls how long Claude thinks before answering. Haiku is the fast, light option. Sonnet is the everyday default, and Opus and Fable are for long contracts and multi-step analysis. Higher models and higher effort use more of your allowance. Anthropic's usage guide explains that all plans have a session limit that resets after five hours, plus weekly limits, and that consumption depends on the model, features and effort level you use.
Once your team agrees how to brief Claude for proposals and client emails, put those briefs in shared Project instructions (level four) so everyone starts from the same one.
Level two: can Claude read files and search the web safely?
Claude reads PDFs, Word documents, spreadsheets, images, screenshots and handwritten notes, and it can search the web and show you its sources. Upload three supplier quotes and it'll put the line items side by side in a table and show which is cheapest per unit. Photograph the whiteboard after a planning meeting and it'll type up the actions with an owner against each.
Web search returns answers with links you can check, and you should check them whenever the facts matter. The paid Research feature goes further: Anthropic's Research guide describes it running many searches and returning a cited report. It's available on Pro, Max, Team and Enterprise, and it needs web search switched on.
At this level you need to decide what staff may upload. A supplier quote is harmless. A spreadsheet of client names and salaries is personal data under UK GDPR, and the ICO's guidance on AI and data protection applies to how you process it. A simple rule of thumb for staff ("client personal data only goes into the company Claude account, never a personal one") prevents most of the problems we see.
Level three: can Claude create real documents and spreadsheets?
Yes. Claude produces downloadable Excel spreadsheets with working formulas, Word documents, PowerPoint decks and PDFs, which Anthropic's file creation guide confirms. Ask for a project budget tracker with columns for cost, sell price and a calculated margin, and Claude returns a working Excel file you can download and share.
It also builds interactive pages beside the chat, which Anthropic calls artifacts. In a business that might be a first draft of a quote calculator or an internal request form, built in minutes and then refined by your team.
Check the output before it's used for anything that matters. A spreadsheet with plausible numbers and a broken formula is harder to catch than an obviously wrong one, because it's less likely to be checked.
Level four: how do Claude Projects and memory work for a team?
A Project is a shared workspace where every chat can see the same files and instructions. Anthropic's Projects guide describes a project knowledge base and custom instructions that apply to every conversation inside it. Add your service catalogue, pricing rules and a note on tone of voice to a sales Project, and a request such as "draft a proposal for a 30-person accountancy firm moving to Microsoft 365" starts from your own material instead of a blank page.
At Inflection Point we manage customers in shared Claude Projects. Each client's agreements and account history sit in one place, and anyone on the team who picks up that account starts with the same background. On a Team or Enterprise plan the Project belongs to the business, so it stays put when an account manager moves on, whereas personal chats leave with the person.
Memory works differently. Claude notices details across your conversations and remembers them, and Anthropic's memory guide confirms you can view, edit or delete what it holds. Incognito chats aren't saved to history or memory. Claude can also import memory from other AI providers, which Anthropic labels experimental.
Our advice to clients is to keep shared knowledge in Projects and treat memory as personal. Ask staff to review what memory has stored once in a while, especially after client work, and to use incognito chats for anything sensitive that doesn't need to persist.
Level five: what can Claude connectors and skills access?
Connectors let Claude read from and act inside other apps, so each one needs only the access its task requires. Over-granting is easy. Connecting Google Calendar, for example, brings up a permissions screen covering view, edit, share and delete. Tick "select all" to answer a question like "which afternoon next week is free?" and Claude now holds the right to delete events, which that question never needed. Marketing and CRM platforms can offer full read and write access in a single click in the same way.
Connectors are built on the Model Context Protocol, which Anthropic introduced in November 2024 as an open standard for connecting AI assistants to the systems where data lives. Ready-made connectors cover common tools, and custom connectors can point at any remote MCP server, with Free accounts limited to one.
How does Claude connect to Microsoft 365 and SharePoint?
Most of our clients run Microsoft 365, so this is the connector we deploy most. Anthropic's Microsoft 365 connector searches SharePoint, OneDrive, Outlook and Teams. On Team and Enterprise plans, the Claude organisation owner enables it and a Microsoft Entra ID Global Administrator grants one-time consent. It uses delegated permissions, so each person's Claude sees only what that person can already open, and it's read-only by default. Write actions such as sending email, posting Teams messages and editing files are optional extras that need further consent.
If a SharePoint site is open to "Everyone except external users", Claude will find what's on it for anyone who asks, exactly as Microsoft 365 Copilot would. Tidying permissions before switching on any AI search is the same job whichever assistant you choose.
Google Workspace businesses have an equivalent. Anthropic's Google Workspace connectors cover Gmail, Google Calendar and Google Drive.
What does least privilege look like with Claude connectors?
Least privilege means giving each connector only the access a task needs, and adding more when a real task demands it. In practice, for our clients that means:
- Starting the Microsoft 365 connector read-only, and approving write actions only for roles with a defined use (for example, a sales lead whose agent drafts follow-up emails).
- Using the organisation controls on Team and Enterprise, where an owner must enable each connector before members can use it, so nobody wires in an unvetted service on their own.
- Reviewing each third-party permissions screen line by line, and leaving delete and share permissions off unless the workflow needs them.
- On Enterprise, using custom roles to decide which groups get connectors, web search, Cowork and Claude Code, and audit logs to see what happened afterwards.
The UK's National Cyber Security Centre explains that large language models don't keep a firm boundary between instructions and data, which is what makes prompt injection possible. An email or web page can contain text that tries to steer an AI agent. An agent that can only read does less damage when that happens than one that can send email and delete files.
What are Claude skills?
A skill is a saved set of instructions that teaches Claude to do a specific job your way. Anthropic describes Agent Skills as folders of instructions, scripts and resources, built around a SKILL.md file, that Claude loads when they're relevant. Anthropic publishes a directory of ready-made skills, such as themes for slide decks and documents, and you can write your own.
Skills are a good home for house standards such as your proposal structure and brand voice. Write them once and every member of the team works to the same standard. Read any skill before installing it, because it's an instruction set Claude will follow.
Level six: what is Claude Cowork, and is it safe for business?
Cowork lets Claude carry out multi-step tasks on its own inside the desktop app, working on files in folders you choose. Anthropic's Cowork guide confirms it runs on Claude Desktop for macOS and Windows on all paid plans, and that it uses the same agentic architecture as Claude Code. Point it at a folder of scanned expense receipts and it can read each one, build a spreadsheet of date, supplier and amount, and rename every file to match.
Cowork asks how much freedom to give it. Manual approval means Claude asks before each action. Automatic approval lets it work on its own with safety checks, and a third mode skips approvals entirely. New users should start on manual, where Claude asks before it opens or changes any file.
In a business, the organisation should decide the approval mode and write it into policy, so individuals aren't choosing it task by task. Owners can switch Cowork on or off for the whole organisation, and Enterprise can enable it for specific teams. We'd keep manual approval for anything touching client data or finance until a workflow has run cleanly for a few weeks, and we wouldn't use the skip-everything mode on a machine that holds company data.
How we use Claude at Inflection Point
Level six is where Claude pays for itself in my own week. Email management saves me hours every week: Claude triages my inbox, drafts replies in my voice and surfaces what needs me, and I approve before anything goes out. We also use it across CRM and marketing, and we've built an end-to-end go-to-market (GTM) pipeline on Claude that takes us from research and content through to campaign follow-up. This post went through the same pipeline, from research through to verified citations and a draft for my review.
In every one of these workflows, a person approves anything that leaves the building. We set out the rules we apply to email agents in two AI agents worth building for your business.
Can Claude run tasks on a schedule?
Yes. Cowork's scheduled tasks run a task you describe once on a recurring schedule. A business might schedule a Monday round-up of sector news or a Friday pipeline review pulled from the CRM. Run each scheduled task once by hand first so any permissions it needs are approved while you're watching.
Beyond the six levels: Claude Code and Claude Design
Two further tools sit above the six levels. Claude Code is an agentic tool that writes code, edits files and runs commands, and despite the name it's usable by non-programmers, because you still describe what you want in plain English. Claude Design, which Anthropic launched as a research preview in April 2026, turns a description into designs and prototypes and can hand them to Claude Code to build.
A typical business use is a small internal tool. Ask Claude Code for a script that reads your asset register and produces print-ready asset labels, and once it's built it runs on the computer with no conversation and no usage limits. A marketing team might use Claude Design to mock up a landing page, then hand it to Claude Code to build. Anything that runs code on company machines falls under the same rules as any other software, so involve whoever manages your IT before staff start building.
Where does Claude process and store UK business data?
Claude's commercial data is stored in the US. Anthropic's server location page says that for commercial products, requests may be processed in the US, Europe, Asia or Australia by default, and data is stored in the US. Anthropic's own API offers an inference location setting of "global" or "us", with no UK or EU-only option.
That doesn't rule Claude out for UK firms. Under Anthropic's Data Processing Addendum, the customer is the controller and Anthropic the processor, and the DPA includes the standard contractual clauses and the UK International Data Transfer Addendum, which is the ICO's mechanism for lawful transfers out of the UK. Anthropic also holds ISO 27001:2022, ISO/IEC 42001:2023 and SOC 2 Type II. For most SMEs that's an acceptable basis, recorded in your data protection impact assessment.
If you need processing to stay in the UK, the route is through a cloud platform. Claude is available on Amazon Bedrock, Google Cloud and Microsoft Foundry, and Anthropic's regional compliance page shows Bedrock and Google Cloud available in Europe, with Foundry in Europe listed as coming soon. AWS lists Claude Opus 5 on Bedrock in its London region. That's a build-your-own route that suits regulated firms with development resource, and it's a different product from the Claude app your staff use.
How does Microsoft 365 Copilot compare on data residency?
Microsoft offers more regional processing options than Anthropic's own apps. Microsoft's Copilot privacy documentation says Copilot routes requests to the nearest regional data centres, with extra safeguards for EU users under the EU Data Boundary, and the UK isn't part of that boundary. In November 2025 Microsoft announced in-country Copilot processing for 15 countries, with the UK among the first, and an April 2026 update moved UK local processing to the end of 2026. At the time of writing, UK in-country processing for Copilot hasn't launched.
Anthropic's Claude models are also available inside Microsoft 365 Copilot, in tools such as Researcher and Copilot Studio, and Microsoft's own documentation says those models are excluded from its EU Data Boundary and in-country commitments. They're off by default for UK customers, with Anthropic acting as a Microsoft subprocessor. Choosing Copilot doesn't automatically keep Claude out of your data flows, so whoever runs your tenant should know which setting you've chosen.
Google applies a similar principle. Google's Workspace privacy hub says Gemini only retrieves content a user can already access, and existing protections such as data-region policies apply automatically, with an exception for Gemini Notebook data.
Claude or Microsoft 365 Copilot: which is right for a UK SME?
There's a time and place for both, and plenty of our clients will end up using both. Copilot is the natural choice for work that lives inside Word, Excel, Outlook and Teams, governed by the Microsoft 365 permissions you already manage. Claude is stronger for writing quality, long documents, building tools and running multi-step tasks in Cowork, and its Microsoft 365 connector lets it work over the same SharePoint and Outlook content.
Microsoft 365 Copilot Claude (Team plan) UK price, annual billing Copilot Business £16.10 user/month ex VAT (promotional £13.80 to 31 Dec 2026); enterprise Copilot £23.10 ex VAT. Needs a qualifying Microsoft 365 plan Standard seat £18 user/month inc VAT (£15 ex VAT). Runs alongside your existing Microsoft 365 or Google Workspace Where it works Inside Word, Excel, PowerPoint, Outlook and Teams Claude apps on web, desktop and mobile, reaching Microsoft 365 through a connector Access to your data Microsoft Graph; only what the user can already view Delegated Microsoft 365 permissions; only what the user can already view; read-only by default Admin and identity Entra ID, Purview, SharePoint Advanced Management and the Microsoft 365 admin centre SSO and SCIM on Team and Enterprise; audit logs and custom roles on Enterprise UK data residency Regional routing today; UK in-country processing expected by end of 2026 US storage for the Claude app; UK hosting only via cloud platforms such as Amazon Bedrock London Strongest at Everyday work inside documents, meetings and email Writing, analysis, building tools and autonomous tasks through Cowork and Claude Code
Copilot prices come from Microsoft's UK Copilot pricing page and enterprise page. Microsoft also bundles Copilot into Business Standard (£18.10) and Business Premium (£24.60) per user a month, ex VAT. Microsoft's deployment guidance includes fixing oversharing as a step before rollout, with SharePoint Advanced Management included in the Copilot licence.
Our rule of thumb: if most of your team's day happens inside Outlook, Teams and Office documents, start with Copilot and get the permissions right. If you have people who write, analyse, build or run repeatable processes, a smaller group of Claude Team seats alongside Copilot usually pays back quickly. We've compared ChatGPT, Claude, Gemini and Copilot plans for UK businesses in more detail, and our guide to making Microsoft 365 Copilot work for your business covers the Microsoft side.
What should you do this week?
Find out which AI tools your team already uses. Ask everyone, without blame, which AI tools they use for work and what kind of information goes into them, and note whether each account is personal or company-owned.
Then decide where company data is allowed to go, move the people who need Claude onto a business plan, and connect it to Microsoft 365 or Google Workspace with read-only access first. Agree the Cowork approval mode as a business and write it down. In our experience, most firms get from that first conversation to a governed rollout within a few weeks.
That's the work our AI consulting team does every week, backed by our own ISO 27001 and Cyber Essentials Plus certification, and our business process automation team builds the Cowork and Claude Code workflows that follow. If you'd like a second opinion on where your business sits on the ladder, Book a Discovery Call and we'll start with what your team is using today.
