Clients keep asking us the same thing: "Should we build this automation in Zapier, Make or n8n?" The marketing pages all say "connect your apps", so here is the version a developer actually needs: where each one runs, what it counts, and the licence detail that trips people up. Everything below comes from each vendor's own docs; plan limits and prices change often, so check those yourself.
1. Where it runs
- Zapier: cloud only. Its data privacy overview says data is hosted on AWS in the United States, with no EU-only storage option.
- Make: cloud only, but each organisation picks a data zone (US1, US2, EU1 or EU2).
- n8n: n8n Cloud, or self-hosted on your own infrastructure (Docker, Docker Compose, npm, or AWS/Azure/GCP).
Self-hosting n8n gives you full control, and n8n's docs are blunt about the price of that: updates, SSL, backups and monitoring become your job.
2. "Is n8n open source?"
Not in the OSI sense. n8n calls itself fair-code and ships under its own Sustainable Use License, with some enterprise files under a separate licence. You can read the source and self-host it, but use is limited to your own internal business purposes or non-commercial/personal use. Automating your own company's work is fine. Embedding n8n in a product you sell is a conversation to have with n8n first.
3. What each tool counts
This is where estimates go wrong, because the three tools meter different things:
| Tool | Unit | What counts |
|---|---|---|
| Zapier | task | each successful action step; triggers, Filter/Paths, failed actions and some built-ins (Formatter, Delay) don't. AI steps and Zapier MCP calls can count extra. |
| Make | credit | for non-AI apps, 1 operation = 1 credit; an operation is one module run, so a module sending 5 emails = 5. A trigger costs 1 per check. AI features vary by tokens. |
| n8n | execution | one whole workflow run. On paid plans only production executions count, not manual test runs. |
So a 10-step workflow that fires 1,000 times is roughly 1,000 executions on n8n, but many more tasks or credits on the other two. Sketch your three busiest workflows and count them in each unit before you choose.
4. Integrations and escape hatches
Zapier says it connects to 9,000+ apps; n8n's GitHub lists 1,500+ integrations; Make splits apps into verified, community and custom. Catalogue size matters less than whether the exact trigger or action you need exists ("update a deal", not just "CRM supported"). When it doesn't, all three can hit an API with an HTTP or webhook step, and n8n also lets you drop into JavaScript or Python nodes.
5. AI agents
- n8n: AI workflows with OpenAI, Anthropic, Google and others, plus tools, memory and MCP servers. You can require a human to approve an agent's tool call before it runs, which is handy for anything that sends or deletes.
- Zapier: AI by Zapier steps, Zapier Agents and Zapier MCP.
- Make: Make AI Agents (model + system prompt + knowledge files + tools). Note that Make says its EU zones don't guarantee every AI request stays in the EU.
6. Data protection, briefly
If workflows touch personal data, the provider is usually your processor (n8n says it isn't one for self-hosted instances, because it never sees your data). For UK and EU clients, map which apps and AI models each workflow sends data to, and how each transfer outside the UK/EU is covered. Self-hosting in-region reduces transfers, but every model or app you call is still a recipient.
Rule of thumb
- Zapier: biggest catalogue, gentlest learning curve, no servers.
- Make: branch-heavy, data-shaping workflows on a visual canvas, with an EU data zone.
- n8n: self-hosting, custom code, or long multi-step workflows where per-run counting helps, if someone can look after it.
The full comparison, with UK GDPR notes and links to every source, is on the Dzimark blog: n8n vs Zapier vs Make for a small business.
