AI for small business · 15 October 2026 · 7 min read

A sensible AI stack for a UK small business: how the pieces connect

Key takeaways

A stack is about connections, not shopping. One assistant on a business plan, joined to the systems that hold your facts, beats five tools that never talk to each other.
The assistant becomes properly useful once it can read your own systems (accounts, orders, CRM), and MCP is the open standard that connects them.
Check the data terms before customer information goes anywhere. Consumer plans and business plans are not the same thing.
The stack differs by business: tradies, online sellers and professional firms each start in a different place.
Cancel what you don’t use every quarter. Subscription creep is the most common way AI spend goes wrong.

Short answer: one good assistant, the office suite you already pay for, your existing systems of record, a connector that lets the assistant read them, and the boring security basics underneath. Buy in that order and stop when the next layer stops paying for itself.

This post is about how those pieces fit together. If you want the categories of tool worth buying, that is the best kinds of AI tools for UK small businesses. If you haven’t yet worked out where AI would help at all, start with how to use AI in a small UK business.

Why “stack” is the right word

The mistake we see most is treating AI as a list of apps to try. Each one does a party trick, none of them talks to the others, and six months later you have five subscriptions and the same Tuesday. A stack is the opposite: a few pieces, chosen so each makes the others more useful.

We are deliberately not naming prices here. They change monthly and an out-of-date number is worse than none. The AI Tools Report costs a stack against your own answers instead.

Layer 1: one general assistant

Claude, ChatGPT, Gemini and Copilot are all general assistants, and for everyday drafting and summarising they are more alike than their marketing suggests. Pick one and put the whole team on it, so prompts and habits can be shared.

The detail that matters more than the brand is the plan. If staff will paste in anything about customers, use a business or team plan and read what it says about data retention and training. The ICO’s guidance on AI and data protection is written for exactly this decision.

Layer 2: the suite you already pay for

If the business runs on Google Workspace or Microsoft 365, both now sell AI features inside their business plans or as add-ons, so check what you already pay for first. What matters for the stack is that email, documents and calendars are where most office hours go, and an assistant built into them saves you the copying and pasting.

Layer 3: your systems of record

This is the layer people forget is part of the stack: your accounting software (Xero, QuickBooks), your order management (Linnworks, Shopify), your CRM. These hold the facts. AI on its own knows a great deal about the world and nothing about your business, so the value comes from putting the two together.

Nothing new to buy here, usually. But it is worth tidying first: an assistant reading a messy catalogue gives you confident answers about a mess. We wrote about getting the existing mess in order.

Layer 4: a connector

The Model Context Protocol (MCP) is an open-source standard for connecting AI applications to other systems, and both Claude and ChatGPT support it. With a connector in place, “what’s stuck in open orders?” is a question you type, not a report you build.

Start read-only. Reading is where nearly all the value is, and a connector that can’t change anything is a much shorter conversation with whoever has to approve it. We make one for Linnworks in two forms: source code you host yourself, and a hosted version that is coming very soon.

Layer 5: the boring layer

A password manager, two-factor sign-in on every account that offers it, and backups. None of it is AI, and all of it matters more once an assistant can reach your systems. The NCSC’s small organisations guide covers it in an afternoon’s reading.

Three starting points

A tradie or a small contractor. Admin happens on a phone in the van, so start there. An assistant on your phone for writing quotes and messages, your accounting software, and something that captures receipts and mileage as they happen. That last part is what GraftG does over WhatsApp.

An online seller or wholesaler. The facts live in your order management system. Assistant, plus a connector into the order system, plus your accounting software. The questions that pay are about stock, stuck orders and returns. See where the hours are in e-commerce and wholesale.

A professional firm. Your hours go into documents and email, so the suite’s own AI features plus one assistant may be the whole stack. Add quote follow-up if proposals go quiet; see AI for small professional services firms.

The parts you can skip

Separate AI tools for jobs your assistant already does. A second assistant “for comparison”. Anything sold as an “AI employee” before you have worked out which job it would do. And any annual plan bought in the first month: monthly until it proves itself. We have a whole post on what to cancel.

Frequently asked questions

Which AI assistant should a UK small business use?

The one your team will actually use, on a business plan. For everyday drafting and summarising the main assistants are closer than their marketing suggests; the plan’s data terms matter more than the brand.

Do I need a business plan rather than a free one?

If anyone will paste in customer details, yes. Check what the plan says about retention and whether your data is used for training, and read the ICO’s guidance on AI and data protection.

What is MCP and do I need it?

The Model Context Protocol is an open standard that lets an AI assistant query your own systems, such as your order management or accounts, instead of you pasting data in. You need it once the useful questions are about your own data rather than the world in general.

How much should an AI stack cost?

It depends on headcount and on which layers you already pay for, and prices change too often to print. The AI Tools Report costs the stack against your own answers and shows the net return.

Sources

Questions, or want a tool we don't have yet? Email hello@grafto.co.uk — a real person replies.