How the diagnostic decides
There is no magic in it, and deliberately no AI model behind it – just the same ten questions we ask in the first twenty minutes of a training session, and the logic we use to answer them. It weighs:
- Your role and team size – a founder of three and a manager in a 200-person firm need different tiers and different data rules.
- Your function – marketing, sales, operations, finance, people, product, professional services, education or creative work each pull in a different second tool.
- What actually eats your week, and what you repeat every single day – the dull, repeatable jobs are where AI pays for itself first.
- What you have to produce – emails, reports, decks, spreadsheets, proposals, posts, notes, designs or video.
- Where your day happens – Microsoft 365, Google Workspace, or neither.
- What ‘great’ looks like in your role – speed, accuracy, insight, persuasion or originality. This is what picks the model, not just the tool.
- Who your work is for – internal colleagues, clients, a board, or the public. The higher the stakes, the stronger the model.
- Whether you handle confidential or regulated data – which changes what is safe to use at all.
Out the other end comes a core stack of three or four tools with a role and a realistic cost against each, a cheat sheet for which model to use for which task, and three things to do this week. You can copy it, share it, or save it as a one-page PDF.
Free, paid or business tier?
Free tiers in 2026 are genuinely good. Start there, always. Then upgrade when one of these three things is true – not before:
- The work matters. The strongest reasoning models sit behind the paid tier, and the difference shows up exactly where it counts: the board paper, the proposal, the analysis someone will push back on.
- You are hitting limits every week. If you are rationing your own usage, roughly £20 a month is the cheapest hour you will ever buy back.
- You handle confidential data. This one is not about capability, it is about safety – see below.
If you handle confidential data, read this first
Do not paste client, financial, health, legal or personal data into a free consumer AI tool. Free tiers can use what you type to train future models, and once it is in, it is not coming back out. Use a paid Team, Business or Enterprise plan with a data-processing agreement and model training switched off, or whichever tools your organisation has already approved. The tool above flags this for you and adjusts its recommendations to a data-safe tier if you say your work is sensitive.
Then master one tool before you add the next
This is the part people skip, so it is worth saying plainly. Take the dullest repeatable job in your week and do it with your daily driver every day for a week. Use the strong model for one thing that genuinely matters. Only when the first tool is a habit should you add the second. That order – depth first, breadth later – is the whole difference between a stack that saves you a day a week and a folder of bookmarks.
It is also, more or less, what we teach. Our AI training programmes start exactly here: your job, your week, your tools, in plain English.
Two worked examples
The layers make more sense against real shapes of business, so here are two, deliberately modest. Neither is a recommendation to buy exactly this; both show the reasoning.
A solo consultant or one-person business. One general-purpose assistant on a paid tier, chosen for long-document work because most of the week is reading and drafting. One research tool with citations, for anything that has to be defensible in front of a client. Whatever is already bundled with your office software, used properly rather than ignored. Total: two paid subscriptions and something you already own. The temptation at this scale is to add a specialist writing tool, a specialist meeting tool and a specialist social tool; the result is four bills and less capability than the two tools used well.
A team of eight to fifteen. The shape changes because consistency now matters more than individual preference. One general-purpose assistant on a business tier for everybody — the business tier mainly for the data terms and administrative control rather than the features. Your office suite's built-in assistant if you are a Microsoft or Google shop, since it is already sitting where people work. One research tool, shared. Then one specialist tool only if a specific team has a specific recurring job that the general tools genuinely do not cover.
Notice what is absent from both: a tool for each task. That is the default state of most businesses that have been buying AI reactively for two years, and it is almost always worse than a smaller stack used competently.
Why adding tools is the expensive mistake
The instinct when AI is not delivering is to conclude you have the wrong tool. Occasionally true. Usually the problem is that nobody has learned the tool you already pay for, and adding another guarantees the same outcome at higher cost.
Three costs compound each time you add one. Learning cost, which is real and mostly unpaid — every tool has its own quirks, and a team that half-knows five tools is less capable than one that knows two properly. Fragmentation cost, where work and context scatter, so the useful setup someone built in one tool is unavailable in the one they are using today. Governance cost, the quiet one: each tool is another set of terms nobody read, another place company data sits, another account to remove when someone leaves.
The sensible test before adding anything: name the specific recurring task the new tool would do, confirm you have genuinely tried it in a tool you already pay for, and name the person who will learn it properly and show everyone else. If any of the three is missing, you are buying a subscription rather than solving a problem.
Mastery beats coverage, consistently and by a wide margin. The businesses getting the most from AI are rarely the ones with the most tools; they are the ones where a few people got genuinely good at two.
Frequently asked questions
What is an AI stack?
An AI stack is the small set of AI tools you actually use, with a clear job for each. A typical one has four or five layers rather than fifteen tools: a daily driver assistant, a research engine that shows its sources, a meeting scribe, and a visuals or media tool if you publish anything.
How many AI tools do I actually need?
Three or four for most people, and almost never more than five. One assistant you use every day does the bulk of the work. Add the second only once the first is a habit.
Which AI tool should I use for work?
It depends on where your day already happens, what great work looks like in your role, and whether you handle confidential data. Microsoft 365 users get most from Copilot; Google Workspace users from Gemini; if you are tool-agnostic, Claude suits careful writing and long documents while ChatGPT is the broadest all-rounder. The tool above asks and picks.
Is ChatGPT, Claude, Gemini or Microsoft Copilot best?
There is no single winner, which is why the argument never ends. They are close on capability and differ on fit: Copilot for proximity to your Office files, Gemini inside Workspace and on very large documents, Claude for long-form writing and nuance, ChatGPT as the broadest generalist. Pick on fit, then learn that one properly.
Do I need to pay, or is the free version enough?
Free is the right place to start and fine for everyday drafting and summarising. Pay when you need the strongest reasoning models for work that matters, when you hit limits weekly, or when you handle confidential data and need a plan where your inputs are not used for training.
Is it safe to use AI with client or confidential data?
Not on a free consumer tier. Use a paid Team, Business or Enterprise plan with a data-processing agreement and training switched off, or your organisation’s approved tools. Never paste real personal data into a tool that has not been approved for it.
Which AI model should I use for which task?
Use the fast default model for quick drafts, replies and tidying up. Switch to the strongest reasoning model for anything that matters. Use deep research mode when you need sourced, cited evidence rather than a confident-sounding answer. The tool above builds this cheat sheet for whichever assistant it recommends.
How long does it take, and is it free?
Under two minutes, ten questions, completely free, no sign-up to see your results. Copy your stack, share it, or save it as a one-page PDF at the end.
How many AI tools does a small business actually need?
Usually two or three. A solo operator does well with one general-purpose assistant on a paid tier and one research tool with citations, plus whatever is bundled with their office software. A team of eight to fifteen typically wants one general assistant on a business tier for everyone, the built-in assistant in their office suite, one shared research tool, and a specialist tool only where a team has a specific recurring job the others genuinely cannot do.
Should I add another AI tool if we are not getting value?
Almost certainly not. The usual cause is that nobody has learned the tool you already pay for, and adding another reproduces the problem at higher cost. Before buying, name the specific recurring task it would handle, confirm you have genuinely tried it in a tool you already have, and name the person who will learn it properly and teach the others. If any of those is missing, you are buying a subscription rather than solving a problem.
Is the business tier worth paying for?
For a team, usually yes — but for the terms rather than the features. Business and enterprise tiers typically offer clearer commitments on whether your inputs train the provider's models, better administrative control over accounts and access, and the documentation you will want if a customer or auditor asks. If you handle client or personal data, that is the tier worth paying for regardless of whether you need the extra capability.
The Oxford AI School helps business owners, teams and individuals across Oxfordshire learn to use AI tools in a way that’s practical, fun, easy and relevant. Built by Harry Lang, marketer of 25 years and reformed tool hoarder.
Got your stack? The next step is using it properly. Explore our AI training for business, see programmes and pricing, or book a free 10-minute intro call and we will sanity-check your stack with you.
