A practical buyer’s guide for UK SMEs comparing AI tools by workflow fit, data handling, administration and measurable business value.
Use consistent evidence for every option.
| Criterion | Question | Evidence |
|---|---|---|
| Workflow fit | Does the product solve a defined process problem? | Evidence from a controlled pilot |
| Governance | Can administrators control users, data and integrations? | Supplier documentation and tenant testing |
| Commercial model | Are licences, usage limits and support predictable? | Current quotation and contract terms |
| Exit readiness | Can data be exported and access removed cleanly? | Documented offboarding test |
Start with the bottleneck, not the tool
The most expensive mistake small businesses make with AI is buying a general-purpose assistant for everyone and hoping something good happens. Adoption stalls, licences go unused, and the conclusion becomes "AI didn't work for us".
The businesses that get value start from a specific, repeated, annoying task — quoting, chasing invoices, writing job adverts, answering the same twelve customer questions — and fix that one thing.
The five categories that actually pay for themselves
| Category | What it replaces | Representative tools | Realistic monthly cost |
|---|---|---|---|
| General assistant | Drafting, summarising, research, first-pass analysis | Microsoft 365 Copilot, ChatGPT Business, Claude Team, Gemini in Workspace | £15–25 per user |
| Meeting notes | Manual note-taking and follow-up admin | Fathom, Fireflies, Otter, built-in Teams/Meet | £0–16 per user |
| Customer support | First-line response to repeat questions | Intercom Fin, Zendesk AI, Tidio | Usage-based, often £0.70–1 per resolution |
| Sales and CRM assist | Manual CRM updating, follow-up drafting | HubSpot Breeze, Salesforce Agentforce, Apollo | Bundled or £20+ per user |
| Document and finance | Manual data entry from invoices and receipts | Dext, AutoEntry, Xero features | £15–40 per month total |
Which assistant, given what you already run
This is the decision most small businesses agonise over unnecessarily. In practice your existing email and document platform should decide it.
- Already on Microsoft 365? Microsoft 365 Copilot. It sits inside Outlook, Teams, Word and Excel and can reason over your own files and mail. Around £18–21 per user per month on the business plans.
- Already on Google Workspace? Gemini is now bundled into Workspace plans rather than sold as an add-on, so you may already be paying for it. Check before buying anything else.
- Neither, or you want the strongest standalone reasoning? ChatGPT Business or Claude Team, both around £15–20 per user per month.
The integration matters more than benchmark scores. A slightly weaker model that can see your inbox and files beats a stronger one you have to copy and paste into.
What it actually costs for a ten-person business
| Item | Monthly |
|---|---|
| Assistant licences, 10 users at ~£20 | £200 |
| Meeting notes, 4 users who run client calls | £0–65 |
| Document capture for bookkeeping | £20–40 |
| Typical total | £220–305 |
Against that, the realistic saving is a few hours per person per week on drafting and admin. For most UK SMEs that maths works — but only if people actually change how they work, which is a management problem rather than a licensing one.
What not to buy yet
- An AI tool for a process you have not defined. If two people do the task differently and neither can explain it, automating it will produce two kinds of wrong answer faster.
- Separate point tools that duplicate your assistant. A standalone "AI writer" adds little if you already pay for Copilot or Gemini.
- Anything requiring a twelve-month commitment before a pilot. Monthly terms exist; use them.
- Custom model training. Almost never the right first move commercially.
Doing it safely without a compliance department
Three rules cover most of the risk for a small business:
- Buy the business tier, block the free one. Consumer accounts typically train on your input; business tiers contractually do not. This single decision removes most of the data protection exposure.
- Write one page of rules. What may go in, what may not (customer personal data, anything confidential, credentials), and who to ask when unsure.
- Keep a human check on anything that reaches a customer or affects a person — quotes, contracts, hiring decisions, complaint responses.
The ICO's guidance is genuinely readable and worth thirty minutes if you handle customer data. For a fuller treatment see our guide to GDPR-friendly AI tools for European businesses.
A sensible first ninety days
- Weeks 1–2: pick one workflow. Measure how long it takes today.
- Weeks 3–4: licence one tool for two or three willing people, not the whole company.
- Weeks 5–8: let them use it properly, and collect what worked and what produced nonsense.
- Weeks 9–12: decide — expand, change tool, or stop. Then repeat with the next workflow.
Running one workflow properly beats rolling out five badly. Most failed SME AI projects are failed change projects, not failed technology.
Frequently asked questions
Is the free version good enough?
For genuinely non-sensitive work, sometimes. For anything touching customer data, no — free tiers generally lack a data processing agreement and typically use your input for training.
Will it replace staff?
For most small businesses it shifts where time goes rather than reducing headcount — less admin, more customer-facing work. Plan for redeployment rather than reduction unless you have a specific, measured case.
How do we know if it is working?
Pick the measure before you start: hours on the task, response time, quote turnaround, invoices chased. "People say they like it" is not evidence, and neither is licence count.