n8n vs Zapier vs Make: which one fits your workflow?

n8n vs Zapier vs Make is usually presented as a feature contest. That misses the decision a small business actually faces: what does the workflow do, how often does it run, and who owns it when something fails?

This guide starts with the billing meter, then compares maintenance, code, and hosting. It also adds an option most comparisons skip: building the automation directly with a coding agent such as Codex or Claude Code.

If you have not read the plain-English version of what these tools do, start with what is workflow automation and come back.

n8n vs Zapier vs Make: what is the actual difference?

All three connect apps, receive a trigger, move data, and run actions. All three now include AI features. All three can also run custom code: n8n has code nodes, Zapier has Code by Zapier, and Make has Make Code.

The difference is the path they make easiest. Zapier has the widest ready-made app catalogue — its official site lists more than 9,000 integrations — and is usually the quickest handoff for a short trigger-and-action flow. Make gives you a visual canvas that is good at showing routers, filters, and several branches. Its product page lists 3,000+ pre-built apps plus custom API and code options.

n8n asks for more technical comfort but gives you a managed Cloud plan or a self-hosted Community Edition. That choice changes where the automation runs, who controls its infrastructure, and who handles updates, backups, credentials, and uptime.

How does each platform count usage?

The headline plan prices are not a clean comparison because each platform counts a different unit. API fees and AI model charges can also sit outside the platform bill.

Zapier counts tasks

Zapier defines a task as a successful action. Its current task guide says triggers, Filters, Paths, and steps that fail or halt do not use standard tasks. A Code by Zapier step normally uses one task per run, with more possible for extended runtime.

A workflow with several successful actions can therefore use several tasks each time it runs. That is easy to budget when the flow is short and occasional. It matters more when a workflow runs all day.

Make bills in credits

Make replaced “operations” with credits as its billing unit. For most non-AI apps, one operation uses one credit. Some built-in AI features can also use credits based on tokens, the model, or other usage factors.

An operation still means one module run. Make’s own example says a module that sends five emails creates five operations. A four-module diagram can therefore consume far more than four credits when it loops over a list of records.

n8n Cloud counts full workflow executions

n8n’s paid plans count one full workflow run as an execution, no matter how many nodes are inside it. Self-hosted n8n Community Edition has no n8n platform fee per execution. It is not cost-free: you still pay for the server, outside APIs, AI models, monitoring, and the time needed to keep it running.

The trade is useful when a long workflow runs often and someone technical can own it. It is a poor bargain when a small hosted flow would save more maintenance time than the server saves in fees.

How do n8n, Zapier, and Make compare?

How to read the table: billing, hosting, code, and app counts come from the platforms’ current documentation. Learning curve, ownership risk, and best fit are our editorial assessment based on those facts and current operator discussions.

Is n8n obsolete now that coding agents can build automations?

No. Codex and Claude Code add a fourth route; they do not erase the other three. A coding agent can help write the scheduler, API calls, filters, tests, database logic, and Telegram messages for a custom automation. Someone still has to decide the rules, secure the credentials, deploy it, watch the logs, and repair it.

Custom code is worth considering when the workflow has unusual state, several validation rules, or logic that becomes harder to understand as a canvas full of nodes. It also fits a technical owner who wants the code in version control and can test changes before release.

A visual platform is often better when the task uses common business apps and a non-developer needs to inspect or change it later. Zapier, Make, and n8n also give you credential handling, run histories, retries, and connectors that a custom build must provide for itself.

What did a custom flight-monitor agent teach us?

I built a flight-monitoring agent in Codex for my own family travel search. It runs at 6:00 in the morning and 6:00 in the evening, checks a flight-search API against the requirements I gave it, and sends the result to a Telegram bot on my phone.

The message includes the route, dates, airlines, flight numbers, journey time, displayed total, and a link where I can check the current price and book. The useful part is not the notification. It is what happens when the search result is incomplete.

That is the reason I preferred the coding-agent route for this build. I could describe the requirements, add a schedule, and make the warning logic explicit. It does not prove that custom code is always better. It proves that n8n, Zapier, and Make are not the only choices.

It also follows the rule I use across my AI work: direction, guardrails, and human review. I do not treat the first generated result as the final answer. The agent narrows the work; I verify the decision before anything important happens.

A travel agency could use the same pattern for saved client searches, provided it has permission to use the flight data source and keeps a person in control of fare, baggage, and booking checks. The automation narrows the search; it should not pretend an unverified fare is ready to buy.

What does the n8n licence actually allow?

n8n is fair-code rather than fully open source. Its Sustainable Use License permits internal business use and non-commercial or personal use.

In plain terms: running n8n to automate your own operations is fine. Packaging it up and letting your customers connect their accounts and build workflows inside it is a different thing, and you should read the licence before you plan a business around it.

This rarely changes a local service business’s internal workflow. It matters if you plan to make n8n itself available to customers as part of a competing hosted product.

Which tool fits your workflow?

You do not need to pick a logo before you describe the job. Start with frequency, judgement, maintenance, and the number of steps. The four-question chooser below gives you a starting point, not a buying decision.

When the tool is not the real question

Two situations come up often enough to name.

The job has no judgement in it. If a workflow is pure rules — move this field there, send that reminder — you do not need an AI step at all, and adding one makes it slower to debug and more expensive to run. The plain guide to how AI automation works sets out where the thinking step genuinely earns its place.

The process underneath is broken. Automation makes an existing process faster, including a bad one. If quotes are inconsistent by hand, they will be inconsistent automatically. Decide what a good output looks like first; the platform argument can wait.

Both questions are worth more than a feature count, and neither is settled by a pricing page.

Frequently asked questions

Written by

Harman Chahal

Harman is the founder of IshaanEx Digital, an Abbotsford agency specialising in local SEO, custom-coded websites, AI search optimisation, and AI automation for Lower Mainland service businesses. He builds and reviews the workflows himself and works with clients in English and Punjabi.

About Harman and IshaanEx Digital

Based in Abbotsford, working across the Fraser Valley and Lower Mainland. Our published pricing and the workflows we build sit on the Abbotsford AI agency page, and the wider service is on AI automation services.