Zapier vs Make vs n8n: how they differ, and when to use custom code instead

What is the difference between Zapier, Make and n8n?

A lead fills in a form on your site, and four things should follow: a row in a sheet, a contact in the CRM, a message to the sales channel and a welcome email. Zapier, Make and n8n all build that kind of chain, from a trigger and a run of steps, and any of the three can build this one. Where they part ways:

• What they bill for: Zapier counts tasks, one for each successful action step (October 2026). Make counts credits, one for each module run on standard apps, and n8n counts executions, one for each run of a whole workflow, however many steps it has (both read November 2026).
• Where they run: Zapier and Make are hosted services. Zapier describes itself as a cloud-based platform on fully managed infrastructure (September 2026), and Make stores and processes each organization’s data in a data center in the US or the EU (read November 2026). n8n runs on n8n’s cloud or on a server of your own, and its community edition, the one you host yourself, is free.
• How you build: all three use a visual editor (read November 2026). Zapier adds Code by Zapier steps for what the editor can’t do, and n8n’s Code node runs JavaScript or Python inside a workflow.
• How many apps they list: Zapier says it connects more than 9,000 apps, Make lists more than 3,500 and n8n more than 2,000 integrations (read November 2026).

For any one workflow, check that each app it uses is listed, with the triggers and actions the workflow needs.

How do Zapier, Make and n8n compare side by side?

Set side by side from each vendor’s own pages, read in November 2026, the billing unit is the row that changes the most from one tool to the next.

How does each one count usage?

The same workflow can count three very different ways. Take one that handles 1,000 new leads a month, each through a trigger and nine action steps:

• Zapier: each lead is one Zap run. The trigger is free and each of the nine actions is a task when it succeeds, so the month comes to 9,000 tasks. Filter, Paths, Formatter and Delay steps don’t count (October 2026), so if three of the nine are Formatter steps, it’s 6,000.
• Make: each action module runs once per lead, and on standard apps one operation is one credit, so the actions come to 9,000 credits. The trigger adds one operation each time it checks for new data: checking every 15 minutes makes 2,880 checks in a 30-day month, for 11,880 credits in all (read November 2026).
• n8n: if each lead starts its own run, the month comes to 1,000 executions, because n8n counts each run of the whole workflow once, however many steps it has. On the free community edition, which you host yourself, the cost to plan for is the server and the time to keep it running.

So the more steps a workflow has, the further apart the three counts get, while a short chain that runs a few hundred times a month fits any of the three. Each vendor charges its own amount per unit and per plan, and plans change, so put your own workflows’ runs and steps against each vendor’s current plans before you choose.

Three more rules matter once the workflows are running:

• Only successful actions count on Zapier: an action that errors or halts, or doesn’t run because a filter, a path or an earlier error stopped the Zap, uses no task (October 2026).
• What happens at the limit: on a Zapier plan without pay-per-task billing, new Zap runs are held once the task limit is reached, until the billing cycle resets. With pay-per-task billing, extra tasks are charged up to three times the plan’s limit, and then runs are held (October 2026).
• Code and AI steps: a Code by Zapier step uses one task per run on the standard runtime (October 2026). Some of Make’s AI modules use more than one credit per operation, and with your own AI provider connection you pay that provider for tokens on top (read November 2026).

What is n8n?

n8n is a workflow automation tool you can use on n8n’s cloud or run on your own server. Three things set it apart from the other two:

• Its license: most of n8n’s code is published on GitHub under its Sustainable Use License, which allows use and changes for your own internal business purposes, or for non-commercial or personal use, and passing the software on to others only free of charge and for non-commercial purposes (read November 2026). n8n’s own license FAQ gives examples: using the community version to help run your business is allowed, and hosting n8n as a service for your clients to build workflows isn’t (September 2026).
• The community edition: the free edition, which runs without a license key, has almost the complete feature set. It leaves out, among other things, environments, external secrets, log streaming, multi-main mode, projects, single sign-on through SAML and LDAP, sharing of workflows and credentials, and version control with Git, which come with the paid Business and Enterprise plans (October 2026).
• Code inside workflows: the Code node runs JavaScript or Python as a step, so logic the visual nodes can’t express stays in the same workflow (October 2026).

Hosting it yourself has a cost the price list doesn’t show. Someone has to install updates, watch the server, back up the data and fix it when a run fails at night.

What are the best workflow automation tools?

There isn’t a single best workflow automation tool. The right one depends on the apps you use, how often your workflows run and who will maintain them. Zapier, Make and n8n are the three this guide compares, and Microsoft Power Automate, part of Microsoft’s Power Platform, is another.

Here is one example of a workflow automation tool at work. When a customer pays an invoice, the tool marks the job paid in the CRM, posts a note to the account manager and adds a row to the month’s revenue sheet, and nobody copies anything by hand.

To choose between the tools, check five things:

• Your apps are listed, in depth: each one with the triggers and actions your workflow needs.
• Your run counts, on each billing unit: model a month of runs the way the example above does.
• Who builds and fixes it: a visual editor for the people who use the workflow, or code for a developer.
• Where the data goes: a hosted service, or your own server when the data shouldn’t leave it.
• How it fails: error alerts, retries and a run history someone will actually read.

When do you outgrow Zapier?

Outgrowing Zapier shows up in five ways, and more than one can appear at once:

• Task counts climb with every step: long Zaps at high volume multiply tasks, because every successful action counts.
• Runs get held: once a plan reaches its task limit, new runs wait, either straight away or after pay-per-task billing reaches three times the limit (October 2026).
• Timing matters: polling triggers check for new data every 1 to 15 minutes depending on the plan, and every 15 minutes on the Free plan (May 2026), which is too slow for a workflow that has to act at once.
• The logic outgrows the editor: branches, loops and code steps pile up until only one person can follow the Zap.
• The Zap has become the system: it holds the state of an order or a job, and the business stops when it does.

Outgrowing Zapier can mean moving a workflow to Make or n8n, when their billing unit fits it better, or moving it out of workflow tools altogether.

When is custom code better than a workflow automation tool?

A workflow tool connects systems. Custom code is better when the work needs something the tools don’t hold:

• Its own data and screens: a workflow that tracks state, keeps history and needs a screen people work in has become a system.
• Volume and speed: thousands of runs a day, or replies within seconds, where per-step billing and polling intervals both get in the way.
• Logic that needs tests: pricing, scheduling or checks that must be reviewed, versioned and tested before each change.
• Systems the tools can’t reach: an internal database, an older system with no connector, or an API a directory covers only in part.
• A step that reads and decides: reading a free-form email or document and choosing what happens next is work for a language model, with a person checking it, and that’s a different kind of build from a rules-based workflow.

The reverse holds too. A workflow that is rules only, runs a few thousand times a month and connects apps the tools list belongs on a tool, and writing it as custom code adds upkeep without adding much. The same question for apps has its own guide: no code vs low code vs custom code.

What doesn’t help when automations keep breaking?

When workflows fail every week, four quick fixes make the next failure harder to find:

• Retries on every step: they hide the cause, and a step that fails for a reason will fail again.
• A second tool on top: two tools mean two places to look when something stops.
• Copies of the same workflow: copies drift apart, and a fix lands in only one of them.
• One person holding it all: if only one person knows how the workflows run, every day they’re away is a risk.

What helps is plain: a list of every workflow with an owner, alerts that reach that owner, and a run history someone checks each week.

What should you set up first?

Before you pick a tool or move a workflow, get the facts in one place:

1. List every workflow you run: its trigger, its steps, the apps it touches and who looks after it.
2. Count a month of runs and steps for each: from each tool’s own run history.
3. Price each one on all three billing units: tasks, credits and executions, against the vendors’ current plans.
4. Check where the data goes: which workflows carry client or payment data, and whether a hosted service is acceptable for them.
5. Move or build one workflow first: with alerts and an owner, before you touch the rest.

Which work to automate in the first place is its own question, answered in what to automate first. What having workflows built tends to cost, with the market’s figures, is on the custom software cost page.

If your workflows are rules and the tools fit them, keep them on the tools: our Zapier, Make and n8n work builds rules-based workflows in all three, and when a plan you already pay for can do the job, we’ll say so. A workflow that needs its own screens and data is custom software, and a step that reads and decides belongs with AI automation. It all begins with a 15-minute call.