Go and look at your business credit card statement for last month. Count the AI subscriptions. Then ask who used each one in the last 30 days.

Most small businesses I see are paying for the same capability three or four times over, and some of those tools do jobs you could set up yourself in an afternoon. This article shows you how the stacking happens, what you’re really paying for, and a simple audit to decide what to keep, what to cancel and what to build.

The subscription stacking problem

Nobody sets out to spend €200 a month on overlapping AI tools. It accumulates.

Here’s a typical pattern. Take a hypothetical 6-person marketing agency. The owner pays for ChatGPT Plus. One designer prefers Claude, so there’s a Claude Pro subscription too. Someone signed up for an AI meeting note-taker after a webinar. There’s an AI writing add-on in the browser, a “chat with your PDFs” tool somebody needed for one tender, an AI image generator, and the business recently added Microsoft 365 Copilot for two people because the accountant mentioned it.

Each tool costs somewhere between €10 and €30 a month per person. None of them looks expensive on its own. Add them up and you’re well past €200 a month, or €2,400 a year, before anyone has asked whether the tools overlap.

They overlap a lot. Most of them summarise text, draft text and answer questions about documents. That’s what the general-purpose chat assistants already do.

What you are actually paying for

A large share of AI point tools are built on top of the same few underlying models from OpenAI, Anthropic and Google. The tool adds an interface, some pre-written instructions and sometimes an integration with another app.

Sometimes that’s worth paying for. A well-built meeting assistant that joins your calls, transcribes them and files the notes in the right place is doing real work. But a tool that takes your text, wraps it in a prompt and sends it to the same model you can already reach through your ChatGPT or Claude subscription is charging you rent for a prompt.

The test is simple. Strip away the branding and ask what the tool does that your main AI assistant can’t. If the honest answer is “it has a nicer button”, you’re paying for the button.

The build-it-yourself alternative

The main AI assistants now let you build small, reusable tools without writing code. Here’s what that looks like in practice:

  • Custom assistants. ChatGPT has custom GPTs, Gemini has Gems, and Claude has Projects. Each lets you save a set of instructions and reference documents so the assistant behaves the same way every time. A “proposal writer” that knows your services, your pricing structure and your tone of voice takes about an hour to set up.
  • Saved prompts and templates. A lot of single-purpose tools are one good prompt. Write it once, test it on 5 real examples, and save it somewhere the team can find it.
  • Small apps built by the AI itself. Claude and ChatGPT can both build simple working tools from a plain-English description: a quote calculator, a checklist, a form that tidies up data. You describe it, test it and refine it in conversation.
  • Automation in tools you already own. If you’re on Microsoft 365, Power Automate is likely already in your licence for many simple flows. On Google Workspace, Apps Script can do a surprising amount, and the AI will happily write the script for you.

Here’s the kind of thing I mean. Dedicated meeting-summary tools charge a monthly fee per user to turn transcripts into action lists. Teams and Google Meet can both produce transcripts on many business plans already. A Claude Project or custom GPT with a good set of instructions (“Pull out every decision, every action, who owns it and the deadline. Flag anything ambiguous.”) does the same job inside a subscription you’re already paying for. Setting it up takes less time than the tool’s onboarding emails.

If you want to learn to build these properly, our Google AI ecosystem course covers no-code tools in Google’s world, and Getting More From AI covers custom assistants in ChatGPT, Claude and Gemini.

When to build vs when to buy

Building isn’t always the right answer. Somebody has to maintain what gets built, and “Ciarán made it before he left” is a known failure mode.

Buy when:

  • The tool handles sensitive data and you need proper security, contracts and an audit trail.
  • It integrates deeply with other systems, such as your CRM, accounts package or phone system.
  • Lots of people will use it and need support when it breaks.
  • It does something your main assistant genuinely can’t, like live transcription or specialist image editing.

Build when:

  • The job is narrow and well-defined, such as drafting one type of document.
  • The data stays inside a business-grade tool you’ve already approved.
  • Getting it slightly wrong is cheap, because a person reviews the output anyway.
  • One or two people will use it and can fix it themselves.

When in doubt, try building first. If your custom assistant handles 80% of the job after an hour’s work, you’ve saved the subscription. If it can’t, you’ve learned exactly what the paid tool needs to do better, which makes the buying decision much easier.

A simple tool audit

Set aside an hour. Here’s the process.

  1. Find every AI charge. Go through 3 months of card statements and expense claims. Include personal cards people expense. Ask the team what they use on free accounts too, because those carry data risk even when they cost nothing.
  2. Record usage. For each tool: who used it in the last 30 days, and for what? Tools nobody has opened in a month go straight on the cancel list.
  3. Map the overlap. Group the tools by job: writing, summarising, meetings, images, documents, research. Where 3 tools do the same job, you need one.
  4. Pick your main assistant. Choose one general-purpose tool for the business and put everyone on its business plan. At the time of writing, standard seats in ChatGPT Business and Claude Team both list at about $25 per seat per month billed monthly, or $20 billed annually. If you already pay for Microsoft 365 or Google Workspace, check what AI is included before you buy anything new. Microsoft lists Microsoft 365 Copilot Business as an add-on for businesses with up to 300 users.
  5. Rebuild the simple jobs. For each tool you’re cancelling, write the custom assistant or saved prompt that replaces it, and test it on real work before the subscription ends.
  6. Decide what earns its place. Whatever’s left is a tool that does something specific and valuable. Keep it, and add it to your AI tool register so you know who owns it.

A business plan for one main assistant often costs less than the scattered personal subscriptions it replaces, and it comes with business terms on how your data is handled. Consolidating saves money and reduces risk at the same time, which doesn’t happen often.

If you’d like a second pair of eyes on the result, our AI implementation guidance work includes exactly this kind of tool review.

Prices current as of 28 October 2026. AI vendors change plans and prices often, and US list prices exclude VAT, so check the vendor’s own pricing page in euro before you buy.