n8n vs Zapier vs Make: hosting, data location, pricing and AI steps

Choose by where your data may go first, by who will maintain the workflows second, and by price last. Run an n8n vs Zapier vs Make comparison in that order and the shortlist shrinks fast. If personal or regulated data must stay on servers you control, n8n is the only one of the three whose own pages describe a self-hosted option. If a small team wants a hosted tool with no servers to run and can accept data stored in the United States, Zapier fits. Make sits between them: a hosted visual builder with a choice of US or EU data zones.

All product facts below come from each vendor's own pages as published in October 2026. Plans and prices change, so check them again before you sign.

n8n vs Zapier vs Make at a glance

If your organisation runs on Microsoft 365, add Power Automate to the shortlist.

Where does your data go? Hosting and data location

Data location is the question to settle first, because it can rule a tool out before any demo. An automation tool sees everything that passes through it: names, phone numbers, invoice lines, patient or customer notes. It also keeps run history for a while, so the data rests there too.

n8n

The n8n pricing page states that n8n Cloud stores data in Frankfurt, Germany, and that self-hosted data sits wherever you host it. The n8n documentation says that without a licence key, n8n runs as the free Community edition. Its Business plan is currently available only for self-hosted n8n. A self-hosted install can run on a server inside your own country, which no hosted option in this comparison offers.

Zapier

Zapier's data privacy page says it hosts data on AWS servers in the United States. Asked whether data can be stored only within the EU, it answers that Zapier does not support this option. Customers on Company or Enterprise plans can set a custom retention period of 7 to 30 days for data held in their account. The pricing page now lists Free, Professional, Team and Enterprise plans, so ask Zapier which current plans include this.

Make

Make sets a data zone for each organisation: US1, US2, EU1 or EU2. Its help page on AI providers and data zones adds a caveat: an EU zone does not mean every AI request stays in the EU. Some features, such as OpenAI models in Make's AI Provider, are processed in the US whatever zone you choose, and Make serves some Claude models from the UK, which is outside the EU. If you connect your own account with providers such as OpenAI, Anthropic Claude or Google Gemini, Make does not guarantee EU processing; routing follows your settings with that provider.

What the law may require

A cloud region is not a country. Frankfurt is outside Jordan and outside Saudi Arabia. Jordan's Personal Data Protection Law No. (24) of 2023 (Article 15) bars transfers to a person outside the Kingdom whose level of protection is lower than the law's, except in listed cases. In Saudi Arabia, SDAIA publishes the Regulation on Personal Data Transfer Outside the Kingdom (Version 2.0, August 2024) under the Personal Data Protection Law. Sector regulators, such as central banks and health authorities, can add hosting rules of their own. This is general information, not legal advice; your obligations depend on your sector and the data each workflow carries.

How each tool charges: pricing models as published

The three tools count usage in different units, and the unit decides which workflows get expensive.

• n8n counts executions. Its pricing page says one execution is a single run of your entire workflow, regardless of steps or data volume. Long workflows cost the same as short ones.
• Zapier counts tasks. A successful action step uses tasks; triggers, failed actions and built-in tools such as Filter, Paths and Formatter do not. The pricing page gives one MCP tool call as an example of 2 tasks. AI steps and code steps draw from the same task allowance, and the AI model tier or code run time decides how many tasks each one uses.
• Make counts credits. Its pricing page says most actions consume 1 credit, and reading, searching, transforming and iterating rows all count. Router and error-handler modules do not. Make's AI Provider charges credits based on tokens as well as operations.

Here is an illustrative sum using those published rules. Take a workflow with a trigger and 10 action steps that runs 1,000 times a month. In n8n that is 1,000 executions. In Zapier it is about 10,000 tasks if each step uses one task, because triggers do not count. It is fewer if some steps are built-in tools, and more if some are AI or code steps that use several tasks each. In Make it is about 11,000 credits, because the module that reads the trigger data also counts, and more if a step loops over rows. The plan price per unit differs, so this tells you which unit to watch, not which tool is cheaper.

Two more cost lines sit outside the subscription. Self-hosted n8n needs a server, backups, monitoring and someone to apply updates. AI steps can also bring a separate model bill when you connect your own provider account.

AI steps and agents in each tool

All three now let a language model read input, decide and call other tools. It also raises the risk, because an agent can act on records without a person in the loop.

• n8n lists an AI Agent node, an MCP Server Trigger and an MCP Client, and human approval for tool calls. The pricing page says that on self-hosted n8n the Assistant runs on your own model key (BYOK).
• Zapier offers AI by Zapier steps on the Professional plan and above, and Zapier Agents billed in activities separately from tasks. Zapier MCP calls draw from the task allowance.
• Make says its current Make AI Agents (New) choose among MCP, module and scenario tools for each task.

Whatever you pick, decide which actions need a person's confirmation before you switch an agent on. Our guide to AI agent governance sets out the guardrails.

Which tool fits which organisation?

Our position: we recommend self-hosted n8n only when a named person will patch, back up and monitor it. An unmaintained automation server holding customer data is a worse risk than a hosted tool with a clear data map.

Illustrative examples

A private clinic group in Riyadh wants appointment reminders and lab-result notifications. The messages carry health data, and its IT team already runs servers in a Saudi data centre. Self-hosted n8n on that infrastructure fits, with a confirmation step before any message goes out, once counsel confirms what the health regulator and the PDPL require for those messages.

A 15-person marketing agency in Amman links web forms, a CRM and a shared inbox. It has no server staff and handles little sensitive data. Zapier or Make fits; the agency should still check what personal data each workflow passes and record where it is stored.

A five-question checklist before you sign

1. Which data classes will pass through the tool, and may they leave the country?
2. Where will the run history be stored, and for how long?
3. Who will build, maintain and fix the workflows when the builder leaves?
4. Which unit do you pay for, and what will your three busiest workflows use each month?
5. Which AI steps can act without a person's confirmation, and where are their model calls processed?

If an answer is "we don't know", ask the vendor in writing and keep the reply with your contract.

How to start

Pick one workflow that matters, map its data, and test it in the tool your answers point to. Our guide on which workflows to automate first helps you choose it. Our Enablement Lab delivers Workflow Automation & Integration across Zapier, Make, n8n or Power Automate, and our workflow design page shows how we map a workflow before automating it. If you are unsure what data your current tools already expose, start with the free AI Exposure Scan, a 45-minute conversation that ends in a written next step.

Related lab: Enablement Lab ↗