Every software provider you already pay for has added an AI feature in the last two years, and every one of them will tell you it’s essential. Some of that is genuinely useful. A lot of it is a subscription add-on looking for a reason to exist. This guide is about the difference — where AI actually saves a small business time or money right now, and where it’s still a solution looking for a problem.
The starting point isn’t “should I use AI,” it’s “what task is eating time I’d rather spend elsewhere, and is there a tool that genuinely takes that off my plate.” That’s the same question that decides whether you need a CRM, an MSP, or any other bit of business software — AI isn’t a special category that gets to skip it.
What “using AI” actually means day to day
For most SMEs this isn’t about building anything custom. It’s one of two things: a general-purpose AI assistant (ChatGPT, Claude, Gemini, Copilot) that you or your team type into directly, or an AI feature bolted onto software you already use — your accounting package suggesting a transaction category, your CRM drafting a follow-up email, your calling app writing up meeting notes automatically. The second kind usually costs nothing extra and is worth switching on. The first kind is a genuine buying decision, and that’s where most of the confusion sits.
Where it tends to earn its keep
Admin and paperwork. Drafting first versions of emails, letters, job descriptions, and standard replies. Summarising a long document or contract before you read it properly yourself. Turning messy meeting notes into a clean action list.
Customer-facing text. First drafts of product descriptions, social posts, and website copy — reviewed and adjusted by a human before it goes out, not published unread.
Bookkeeping and admin automation. Tools like Dext and Hubdoc use AI to read receipts and invoices and code them automatically, cutting down manual data entry, especially if you’re already on Xero or QuickBooks.
Meeting notes and transcription. Tools like Otter.ai record and summarise calls and meetings automatically, useful if you’re the only person who was in the room and need a record for someone who wasn’t.
Spreadsheet and data work. Asking a general-purpose assistant to spot patterns in a spreadsheet, draft a formula, or summarise a set of figures — a genuine time-saver if spreadsheets aren’t your strong suit.
Where it’s shakier
Customer service chatbots. Promising in theory, disappointing in practice unless properly set up on your own product information. A generic chatbot that hasn’t been trained on your actual prices, policies, and stock will confidently give a customer the wrong answer — which is worse for trust than not having one at all.
Anything you’d publish unread. AI output reads fluently and confidently even when it’s wrong. Treat it as a first draft from a keen junior, not a finished answer — especially for anything involving figures, legal wording, or a claim about your own business.
Hiring and credit decisions. Screening CVs or scoring applicants with AI is legally sensitive territory (see below) and not something to bolt on casually just because a recruitment platform offers it.
Choosing a general-purpose assistant
If your team already runs on Microsoft 365 or Google Workspace, the practical starting point is usually the AI already sitting inside it, rather than a separate subscription:
- Microsoft 365 Copilot – a paid add-on (roughly £16–25 per user, per month on top of your existing Microsoft 365 licence, with pricing under review as Microsoft restructures its Copilot bundles through 2026). Works inside Word, Excel, Outlook and Teams — the tools most UK offices already live in.
- Google Gemini – bundled into Google Workspace Business Standard and above at no extra charge, working inside Gmail, Docs, Sheets and Drive. If you’re already on a qualifying Workspace plan, this is the cheapest way in since there’s nothing new to buy.
- ChatGPT (from OpenAI) and Claude (from Anthropic) – general-purpose assistants used directly, independent of any office suite. Both offer a free tier, an individual paid tier (roughly £16–20/month), and a Team tier (roughly £20–25 per user/month) with shared workspaces and stronger data-handling controls. Worth considering if your team’s stack is mixed, or you want a tool that isn’t tied to one ecosystem.
There’s no single “best” one — the right choice tracks whichever software your business already runs on. The genuine mistake to avoid is stacking several of these at once “just in case.” Most SMEs only need one general-purpose assistant plus whatever’s already bundled into their existing software.
The cost that doesn’t show up on the invoice
The subscription price is rarely the real cost. Two things matter more:
Time to get useful results. A free or cheap AI tool that nobody on the team has learned to use well is money and time wasted either way. A short amount of time working out good prompts for your three or four most common tasks pays back far more than upgrading to a pricier tier.
Subscription creep. It’s easy to end up paying for a general assistant, an office-suite AI add-on, a bookkeeping AI tool, and a transcription tool simultaneously, with only one or two people using each. Before adding a new AI subscription, check whether a tool you already pay for already does the job.
Data protection: the part that’s easy to skip
AI tools process the text and data you put into them, and UK GDPR doesn’t stop applying just because the processing is done by an AI model. A few practical points that matter for any size of business:
- Be careful what you paste in. Don’t put customer personal data, financial details, or anything commercially sensitive into a free or personal-tier AI tool unless you’ve checked its data-handling terms — free tiers more often use input data to train the underlying model. Paid business/team tiers typically offer stronger guarantees on this, but “typically” isn’t “always,” so it’s worth checking each provider’s terms directly.
- Tell customers if they’re talking to a bot. If you use an AI chatbot or automated responder for customer interactions, UK guidance is clear that this should be disclosed, not left for the customer to guess.
- Human review still applies to significant decisions. Following changes to UK data protection law that took effect in February 2026, businesses have somewhat more flexibility to use automated processing — but a decision that has a legal or otherwise significant effect on someone (for example, rejecting a job application or a credit application) still requires genuine human involvement, not a rubber stamp on whatever the AI decided.
- A Data Protection Impact Assessment (DPIA) is generally expected before rolling out any AI tool that will process customer or employee personal data at scale — worth a quick check before a business-wide rollout, even if it turns out not to be needed.
None of this rules AI out. It just means treating an AI subscription the same way you’d treat any other tool that touches customer data — with a quick check of what it does with that data before you commit, not after.
Limited Company and Sole Trader Perspectives
Sole trader, one-person marketing consultancy. Uses Claude’s free tier for first drafts of client proposals and social copy, upgrading to the £16/month individual tier once volume justified it. As a sole trader with no separate legal entity, the subscription is simply an allowable business expense set against income; if not VAT-registered, there’s nothing to reclaim, so the monthly charge goes through as a straightforward cost.
Limited company, 8-person marketing agency. Standardised the whole team on Google Workspace’s bundled Gemini rather than adding a separate AI subscription on top, since the team was already paying for Business Standard. As a VAT-registered company, the incremental cost sits within the existing Workspace invoice, input VAT is reclaimed in the normal way, and the cost is deductible as an ordinary business expense — no apportionment needed, since it’s used wholly for business purposes.
Checklist before you commit
- Identify one or two specific tasks eating real time, rather than buying AI “in general”
- Check whether a tool you already pay for has a built-in AI feature that already does the job
- If choosing a general-purpose assistant, match it to your existing office software (Microsoft 365 → Copilot, Google Workspace → Gemini) unless you have a specific reason not to
- Treat every AI output as a first draft that needs checking, not a finished answer
- Check a tool’s data-handling terms before pasting in anything involving customer or financial data
- Disclose AI-driven chatbots or automated responses to customers
- Keep genuine human review in place for any decision with legal or significant effect on a person (hiring, credit, similar)
- Review what you’re actually paying for every few months — AI subscriptions are an easy place for costs to creep