AI tools fall into a handful of categories rather than one list: general assistants (ChatGPT, Claude, Gemini, Copilot), document and search tools, meeting and transcription tools, workflow automation (n8n, Make, Zapier), coding assistants, image and video generators, and AI features already built into software you own. Most businesses need two or three, not twenty.
Asking what AI tools are there gets you a list of five hundred, which is the least useful possible answer. The tools worth knowing fall into a handful of categories that do genuinely different jobs, and inside each category the options are close enough that the choice rarely decides your outcome.
The useful move is to identify which two categories touch your actual bottleneck, then pick a competent tool in each. Most SMEs need two or three tools, not twenty, and the ones that struggle usually have the opposite problem: nine subscriptions and no workflow.
ChatGPT, Claude, Gemini and Microsoft Copilot. They write, summarise, analyse documents and answer questions, and for ordinary office work the differences between them are smaller than the comparisons suggest. Choose on where your work already lives — Copilot if your company runs on Microsoft 365, otherwise whichever your team prefers — and on the commercial terms, because that is what decides which documents your staff may put into it.
n8n, Make and Zapier connect the systems you already use and run a process without a person triggering it. This is the category that produces measurable savings fastest, because it removes retyping rather than adding a new thing to learn. They differ mostly in hosting and billing model rather than features now, which is the comparison made in n8n vs Make vs Zapier.
Transcription and meeting-notes tools, document Q&A, and search across your own files. These are the easiest wins and the easiest to over-buy: a general assistant with document upload already covers a surprising amount of it. Buy the specialist tool when volume or compliance requires it — recorded consultations, regulated records — not because a demo was impressive.
The most overlooked category. Your accounting package, CRM, helpdesk and ERP have shipped AI features you are probably already paying for — invoice recognition, reply suggestions, forecasting. Auditing what you own before buying anything new routinely removes half a shortlist, and it avoids the worst outcome: a new tool that duplicates a feature you already licensed, holding a second copy of your data.
Two or three, chosen deliberately. Tool count is not a maturity signal; used-in-a-workflow count is.
For evaluating, yes. For running on, usually not — and the reason is rarely capability.
Start from the process, not the catalogue. The right first tool is decided by which task is repetitive, high-volume and low-judgement.
Want this applied in your business? See how we take it to production:
We build this AI in production — at fixed prices, with one named expert. Start with a free consultation.
Book a free consultation →