Mutree SolutionsTalk to sales

Zapier vs Make vs n8n vs Custom Code: Where Each One Stops Making Sense

Ozzy NsakaBuilding agents8 minute read

Most comparisons of Zapier, Make and n8n are feature grids. They usually leave out the fourth option, writing the workflow as code, and they compare features when the decision tends to turn on three other things: where the workflow runs, what the vendor counts as usage, and who is told when a step fails.

This is how we weigh the four for a company with real operations work to automate. Every claim about a vendor comes from that vendor’s own documentation, listed under Sources at the end.

The short answer

  • Zapier when the people who will own the workflow are not technical, every app involved is a cloud app, and each run has only a few actions.
  • Make when the workflow branches, works through lists of records, and needs careful control over what happens when one module fails.
  • n8n when the data has to stay on infrastructure you run, or developers want to keep workflows in Git alongside a visual editor.
  • Custom code when the logic is the part of the business that sets you apart, needs tests on every change, or has outgrown what a canvas can show clearly.

How the four compare

ZapierMaken8nCustom code
Where it runsZapier’s fully managed cloudMake’s cloud; an Enterprise on-premise agent reaches systems inside your networkn8n Cloud, or self-hosted on your own servers or private cloudWherever you deploy it
What counts as usageTasks: each successful action stepOperations: each module run, once per bundleExecutions on n8n Cloud: one per production run of a workflowNothing per step: hosting and upkeep
Code inside a stepCode by Zapier: JavaScript or Python, sandboxedMake Code app: JavaScript or Python, in open beta on paid plansCode node: JavaScript or PythonAll of it is code
When a step failsError handler paths on each step, on Professional, Team and Enterprise plansError handlers: retry, resume, skip, commit or roll backAn error workflow that starts when a run failsWhatever you write and test
Change historyDrafts and version rollback, depending on planSaved versions kept for 60 daysGit-based source control on Business and Enterprise plansYour own repository and code review
Who keeps it runningZapier runs the platform; you look after the ZapsMake runs the platform; you look after the scenariosn8n on Cloud; you, when self-hostedYour engineers, or a firm you hire

Billing has a shape before it has a number

We do not quote vendor prices here. They change, and for this decision the shape matters more: each tool counts something different, and the thing it counts decides which workflows grow in cost as volume grows.

Zapier counts tasks. A task is an action step that completes successfully. Triggers do not count, and neither do filters, paths, several built-in utility steps or test runs. So usage tracks the number of actions in each run multiplied by the number of runs.

Make counts operations, one each time a module runs. A module runs once for every bundle it receives, so lists multiply the count. Make’s own example is a trigger that returns ten bundles followed by three modules: thirty-one operations for one scenario run. Iterating over long lists is where Make usage climbs.

n8n Cloud counts executions. One run of a workflow is one execution, however many steps it has, and only production runs count: manual runs, sub-workflow runs and error-workflow runs do not. Self-hosted, n8n runs as the free Community edition when no licence key is set, and the cost moves to your servers and the time of whoever looks after them.

Custom code is not metered by step at all. It is built once, then hosted and maintained. That shape suits steady, high-volume work whose logic rarely changes, and suits poorly anything that is redesigned every month.

When each one is the right answer

Zapier

Zapier fits a team without developers that wants to connect cloud apps it already uses: a web form into the CRM, a new order into a spreadsheet, a signed contract into a welcome email. It is fully managed, so there is nothing to host or patch. For the odd step that needs logic, Code by Zapier runs JavaScript or Python in a sandbox with time and memory limits.

It fits less well when each run fires many actions at high volume, since every successful action is a task, or when the systems to reach sit inside your own network. Zapier’s documented route to private systems is VPC peering for Enterprise customers on AWS, which links one cloud to another rather than reaching a server in your office.

Make

Make fits workflows with real structure: routers, iterators over lists of records, and different handling for different failures. Its error handlers let a scenario retry a module, skip the bad record, resume with a substitute value, or commit or roll back what has already run. That control is the main reason we would choose it over Zapier.

Make runs scenarios in its own cloud. When a scenario must reach a local database or application, the on-premise agent, an Enterprise feature, is installed inside your network and polls Make for work, so you do not open firewall ports. Keep an eye on the operation count of any scenario that loops over large lists.

n8n

n8n fits when data has to stay on infrastructure you control. It can be self-hosted on your own servers or in a private cloud. Code nodes take JavaScript or Python, an error workflow catches failed runs, and on Business and Enterprise plans workflows can be kept in Git with separate environments for testing and live use.

Two things to know before choosing it. Self-hosting means you provide and manage the infrastructure, and n8n’s documentation is plain that this needs technical expertise. And n8n is released under its own Sustainable Use License, which allows use and modification for your own internal business purposes. Read it before you build n8n into something you sell.

Custom code

Code fits when the workflow is a business rule that sets you apart, when it needs automated tests that run on every change, or when a canvas of dozens of boxes has become harder to follow than the code it stands in for. It also fits when volume is high and steady, since nothing is counted by step.

It is the slowest of the four to get started and needs engineers to keep it healthy. Our own products, Scout360 and Orbit Fleets, are built this way because they are products. Most internal workflows are not, and should not be.

When not to build custom

Do not write code for a workflow that moves records between two cloud apps and changes once a year. A Zap or a scenario does that job, and your team can read it.

Do not choose code when nobody will look after it. If no engineer, in-house or contracted, will answer when it breaks, a no-code tool that your operations lead can open and fix is the safer choice.

Do not rewrite to escape a usage bill until you have checked the shape of the bill. The fix is often a filter earlier in the Zap, fewer modules inside a loop, or one heavy step moved into a code step.

And do not write code while the process itself is still settling. Prove the workflow in a no-code tool first. Move it into code once it has stopped changing and the volume justifies the move.

Who owns what

Ownership differs on each route, and it is worth settling in writing before anything is built.

  • Zapier, Make and n8n Cloud. The workflows live in an account on the vendor’s platform. When we build them, we build them in your account, so they belong to you and stay with you when we leave. The platform itself remains the vendor’s.
  • n8n self-hosted. The workflows and the servers are yours. The software is used under n8n’s licence.
  • Custom code. When we build custom software for you, you own the code, along with the rules, prompts and data that make it work.

AI agents follow the same rule. An agent we build inside your accounts belongs to you, and so does any agent code we write for you.

Mixing routes

The best answer is often two routes rather than one: a handful of Zaps or scenarios for simple handoffs, plus one small custom service for the step that carries the rule only your business has. We have written separately about where an AI agent should live, and about the wider choice of custom software versus off-the-shelf.

For the kind of work these tools end up doing, see the operations problems we are asked to fix, or how we approach AI automation for Houston companies.

Questions people ask

Can Zapier or Make run on my own servers?

Zapier describes itself as fully managed cloud infrastructure. Make runs scenarios in its cloud, and its Enterprise on-premise agent lets them reach systems inside your network. If the workflow itself must run on your servers, the routes are self-hosted n8n or custom code.

Is self-hosted n8n free?

Without a licence key, self-hosted n8n runs as the free Community edition, under the Sustainable Use License, for your own internal business use. You still provide the servers and the time to run them, and features such as Git-based source control belong to the Business and Enterprise plans.

What happens when a step fails?

Each of the three tools lets you plan for it, as the table above shows. In code, failure handling is whatever you write and test. Whichever route you take, decide who gets told and what record is kept.

When should a workflow move into custom code?

When it carries a rule that sets you apart, when usage counting has started to shape how the workflow is designed, or when the team spends more time working around the canvas than working in it. Until then, leave it where it is.

Where to start

Write the workflow down in plain words: what starts it, which systems it touches, how often it runs and who should hear when it fails. That one page usually makes the choice obvious.

Talk to sales. Bring that page and we will tell you which of the four fits, including when the answer is a tool you already have.

Sources

Vendor documentation, checked on 21 November 2026.

Read next

Get new field notes by email

An email when we publish something new. You confirm by email first, and every email has a one-click unsubscribe.

We email you a link to confirm. Every email has a one-click unsubscribe. Privacy