n8n vs Zapier vs Make, and when to write code instead

We sell automation work, so this n8n vs Zapier comparison, which covers Make too, takes every price and feature from the vendors' own pages, read in October 2026. It compares what you pay for, where your data is kept, what happens when a step fails and how AI is counted, then a fourth option, custom code, which is most of our business automation work, and when it beats all three.

The short answer

• Zapier for the quickest start, 9,000+ apps by its own count, and workflows a non-technical person can own. Costs rise with steps and volume.
• Make for workflows that branch and loop, or an EU data centre. In Zapier vs Make, Make gives finer control of errors and a choice of where data is stored.
• n8n if someone on your team is technical, workflows have many steps, or data must stay on servers you control. In n8n vs Make, per-run pricing and self-hosting are the difference.
• Custom code when the process is central to the business, runs at volume, or is part of a product you sell.

Not sure what the process is yet? Map it before choosing a tool; our guide to process mapping shows how.

How each one charges

Zapier: tasks

Zapier counts a task each time a Zap successfully completes an action. Triggers, checks for new data, failed actions and built-in tools such as Filter, Formatter, Paths, Delay and Looping don't count. As of October 2026, its pricing page, in US dollars with a currency selector that includes pounds, lists a free plan with 100 tasks a month and two-step Zaps, Professional from $19.99 a month billed yearly for 750 tasks ($29.99 paid monthly), and Team from $69 a month billed yearly for 2,000 tasks. At the limit, Zaps carry on at a higher per-task rate, or pause if you turn pay-per-task billing off.

Make: credits

Make bills in credits, which replaced operations. Most modules use one credit per operation, and a trigger uses one every time it runs, even when it finds nothing new, according to Make's guide to how features use credits. Routers, filters and error handlers are free. As of October 2026, Make's pricing page, in US dollars, gives the free plan 1,000 credits a month; Core, the cheapest paid plan, is $11 a month billed annually ($16 paid monthly) for 10,000 credits, rising with the credit tier.

n8n: executions

n8n counts one execution for each run of a whole workflow, however many steps it has. As of October 2026, its pricing page shows UK visitors Starter at £20 a month billed annually for 2,500 executions, Pro at £50 for 10,000, and Business, self-hosted only, at £569 for 40,000. All plans include unlimited users and workflows.

A worked example

Take one workflow: a web order arrives, then an invoice is created, the customer is added to the CRM, a message goes to Slack and a row is added to a sheet. At 2,000 orders a month:

• Zapier: the trigger is free and four actions per order make 8,000 tasks, which needs the 10,000-task Professional tier, listed at $129 a month billed yearly.
• Make: four credits per order for the actions, plus one per trigger run. An instant trigger runs once per order, making 10,000 credits: Core's first tier, at $11 a month billed annually. A trigger that polls every 15 minutes runs about 2,900 times a month instead, order or not, so the total passes 10,000; the next tier, 20,000 credits, is $20 a month billed annually.
• n8n Cloud: 2,000 executions, inside Starter's 2,500 at £20 a month billed annually.

Per-step pricing grows with every step and every order; per-run pricing ignores the steps. Self-hosting n8n swaps the subscription for a server and someone's time.

Where your data lives

For a UK business this often decides it: the ICO calls a transfer of personal information to a separate organisation outside the UK a restricted transfer, with its own rules.

• Zapier hosts data on AWS servers in the United States. Its data privacy page says it does not offer EU-only storage, and that it is certified under the UK Extension to the EU-US Data Privacy Framework.
• Make lets you choose a US or EU data centre when you create an organisation, fixed from then on. Its page on AI providers and data zones warns that on an EU zone some AI features still route to the US or the UK, and your own AI provider connection removes Make's EU guarantee.
• n8n Cloud stores data in the EU, on servers in Frankfurt, according to its pricing page. Self-hosted, the data is wherever you run it.

Self-hosting n8n, and its licence

n8n's guide to choosing says the free, self-hosted Community edition has almost the complete feature set, but needs technical expertise, and hosting, updates and scaling become your job.

n8n doesn't call itself open source. Its Sustainable Use License, which n8n calls fair-code, allows use for your own internal business purposes. The licence FAQ allows building automations for clients on your instance if they can't create or edit them, and using n8n behind the scenes in your own product, even where users connect their own accounts. Letting clients or your product's users build or configure their own workflows needs a commercial agreement with n8n.

When a step fails

APIs go down and sign-ins expire. What matters is whether the tool tries again, and whether a person finds out.

• Zapier autoreplay, on paid plans, retries a failed step up to five times, the last about 10 hours 35 minutes after the first error, and holds back error emails until then. A custom error handler, on any step except the trigger and Paths steps, runs other steps instead when it fails, and turns autoreplay off for that Zap.
• Make has five error handlers: Skip, Retry, Resume, Commit and Rollback. With incomplete executions switched on in a scenario's settings (they are off by default), Make retries rate-limit, connection and timeout errors automatically, eight times over almost eight hours. The Retry handler covers other errors, three times 15 minutes apart by default.
• n8n lets you switch on Retry On Fail for a node, and choose whether an error stops the workflow, carries on, or goes down a separate error output. Each workflow can name an error workflow that runs when an execution fails, to send an alert, for example.

None of them can tell whether a step is safe to repeat. If a retry raises a second invoice, that is a design problem; we cover it in our guide to business process automation.

AI features

• Zapier: its task usage rates put an AI by Zapier step at one task on standard models, three on advanced and five on premium. Zapier Agents is a separate add-on billed in activities: 400 a month free, or 1,500 on Agents Pro at $33.33 a month billed annually. One tool call through Zapier MCP uses two tasks.
• Make: per its credits guide, Make AI Agents and the AI Toolkit can use Make's own AI provider, paid in credits based on tokens, or, on paid plans, your own OpenAI or Anthropic account, paying the provider for tokens.
• n8n: its AI nodes are built on LangChain, with an AI Agent node and chat model nodes for Anthropic, OpenAI and Ollama, among others. You can require a person's approval before an agent uses a chosen tool.

An AI step sends data to whoever runs the model. If data mustn't leave your servers, the model must run there too.

Zapier, Make, n8n and custom code side by side

When to stop and write code

A workflow tool stops being sensible when:

• The process is the business. Anything that moves money deserves tests that run before every change, a log of what happened and why, and logic someone can review.
• Volume and steps keep growing. Bills counted per task or per credit grow with both.
• It is part of a product you sell. Customers connect their own accounts, each system with its own rules.
• Nobody dares change it. When a small change means an afternoon of clicking through test runs, the logic has outgrown the canvas.

Add to CRM, a Chrome extension we built and run that adds people from LinkedIn, Gmail and Outlook to any of 29 CRMs, is the kind of job where code wins. Each CRM has its own sign-in, field types, limits and quirks. About 2,000 CRM tests run without live credentials in about 90 seconds, and gate tests fail the build if any integration matches people its own way instead of through the shared rule. A bulk add stops after five identical failures instead of repeating one fault down a whole list, and text is never mapped to a picklist, so no junk options appear in someone's CRM. Rules like those are hard to keep true across 29 workflows on a canvas.

Code still needs upkeep: APIs change whoever wrote the integration. But the fix can be tested before it goes live.

If you want a hand

For a few simple automations between popular apps, pick one of the three tools and build them yourself. If the process is central, runs at volume, or has outgrown its workflows, start a project and tell us how it works today.