In short: Five connected tools beat fifteen separate subscriptions. What you need: a language model, transcription, one place for data, a chain that triggers things, and a dashboard for the numbers. Everything else is a special case.
The rule before the list
The most common state I see in businesses: ten subscriptions, each solving a fifth of a problem, none connected to the others. The data sits in five windows and nobody knows which number is right.
That's why I decide by connection, not by feature: a tool only gets added if it has an API and if it's clear which other tool it replaces or feeds.
The stack by task
| Task | What actually stays | What matters |
|---|---|---|
| Writing, summarizing, thinking | a large language model (for me Claude, plus ChatGPT) | business access with a contract, not the free tier |
| Software and analysis | Claude Code or something comparable in the terminal | access to files, not snippets in a chat |
| Calls and dictation | transcription, local if possible | content never leaves the building |
| Data and storage | one database instead of five spreadsheets | permissions per role, otherwise everyone sees everything |
| Chains and triggers | a builder like n8n | traceable, recoverable, no subscription sprawl |
| Customer contact | a CRM where everything really lands | phone, email and tasks in one place |
| Images, video, voice | one provider each, archive results locally | clear rights and origin before anything gets published |
| Numbers | a dashboard on one source | one source of truth, not three spreadsheets |
That's all a business needs to get started. Anything beyond it is either industry-specific software or the wish to try something new – both legitimate, but not the same as work.
GDPR: four points that matter in practice
- Business access, not free. Only there do you get a data processing agreement, and only there do the providers commit to not using your inputs for training. Terms change – if in doubt, re-read them, don't rely on memory.
- Records of processing. The tool nobody registered is the most common real violation.
- Minimize data. Replace names, shorten numbers, transcribe call content locally. More effective than any contract clause.
- Clarify location. Where does processing happen, where does the copy live? That applies not just to the model but also to transcription and storage.
Local models are now good enough for transcription and sorting, and they solve the data question at the root. For demanding writing and code, the big providers are still further ahead – there, only lean input plus a contract helps. I'm not a lawyer; for a binding assessment you need one.
What I run and why
- Claude as the main tool for copy, analysis and software – including terminal access to real files.
- Local transcription on my own machine, so call content stays in-house.
- Close as the CRM, because phone, email and tasks live in one place there.
- One database as the source of truth instead of spreadsheets, with permissions per role.
- My own services on a dedicated machine on the network for everything that has to run constantly – cheaper and easier to control than five subscriptions.
- A dashboard that pulls the daily numbers from the reports, so decisions rest on one source.
The cost trap
Twelve subscriptions at €30 each come to €360 a month for tools of which three get used. My test, once a quarter: every tool has to be able to name which task it does and what it replaced. If it can't answer both, it gets cancelled. The difference isn't the price per subscription – it's that unconnected tools create double work instead of saving it.
Frequently asked questions
Which AI tools are GDPR-compliant?
Compliance isn't a property of the tool but of how you use it: business access with a data processing agreement, an entry in your records of processing, as little personal data as possible and a clear processing location. Local models for transcription and sorting solve the issue at the root.
Do I need ChatGPT and Claude at the same time?
Many businesses run both: one as a subscription for breadth, one as API access for what gets built into their own software. If it has to be just one, the main task decides – software and long documents point to Claude, images and voice to ChatGPT.
Are local AI models worth it?
For transcription, sorting and recognition, yes – they're good enough, cost nothing per call and keep data in-house. For demanding writing and code, the big providers are clearly ahead.
How many AI tools does a small business need?
Five: a language model, transcription, one place for data, a chain for triggers, a dashboard for numbers. Connected. Everything beyond that is a special case or curiosity.
How do I spot a subscription I can cancel?
It has to answer two sentences: which task it does and what it replaced. If it can't say both, you're paying for possibility, not for work.
Mario De Luca I build sales systems and AI-powered software for businesses and founded the Clone Looping System (CLS). Every figure in these guides comes from my own projects, not from studies. Anything I haven't measured myself is labeled as an estimate. More about me · Journal
Who's writing this
I'm Mario De Luca, an AI systems builder from the DACH region and founder of NextCore and the Clone Looping System. My story, my projects and the principles behind these guides are on my about page. In the journal I write about what I've built, tested or messed up myself.
About Mario De Luca →