Most businesses arrive at n8n vs Zapier the same way: the Zapier bill lands, someone searches for an alternative, and n8n appears with the word “free” attached. It is a fair question with a more interesting answer than the pricing pages suggest, because the three tools do not just differ on price — they differ on who does the maintenance, and that is the part that decides whether the saving is real.
The Zapier vs Make comparison covers those two in detail. This one adds the third option and answers the only question that matters: when is running your own automation server worth it?
The three, side by side
| Zapier | Make | n8n | |
|---|---|---|---|
| Hosting | Cloud only | Cloud only | Self-hosted or their cloud |
| Billing unit | Tasks (each action) | Operations (each module run) | Workflow executions — steps inside are free |
| Cost at volume | Highest | Middle | Lowest (flat server cost when self-hosted) |
| App library | Largest by far | Large | Good and growing, plus generic HTTP |
| Logic | Paths and loops | Routers, iterators, aggregators | Branching, loops, merges, plus code nodes (JS/Python) |
| Custom code | Limited | Some | First-class |
| AI / agent features | AI steps and assistants | AI modules | Agent nodes, your choice of model, local models possible |
| Data location | Their cloud | Their cloud | Your server, if you want |
| Learning curve | An afternoon | A few evenings | A few evenings, plus server skills to host |
| Who maintains it | Them | Them | You (self-hosted) |
The billing difference, in plain numbers
Take one flow: a form submission that creates a CRM contact, looks up a postcode, branches on service type, sends a text, notifies Slack and writes a row to a sheet. Six steps.
- Zapier counts roughly one task per action that runs — so about five or six per submission.
- Make counts every module run, including the router and filters — about seven operations, but each costs a fraction of a Zapier task.
- n8n counts one execution, whatever happens inside. Self-hosted, it counts nothing at all beyond the server.
At 200 submissions a month the difference is pocket change. At 5,000, or with flows twice as long, it is the difference between a meaningful monthly bill and a small server. That is the whole economic case, and it is genuine — but it only counts if the server is looked after.
What self-hosting actually involves
- A server: a small cloud instance is enough to start; more memory once flows get heavy or handle files.
- Docker and a reverse proxy with HTTPS, a domain, and the editor behind authentication — never exposed openly.
- A database (Postgres in production, not the default file store) and backups of it, tested. Your workflows and credentials live there.
- Upgrades. n8n moves fast; nodes and behaviour change. Update deliberately, on a staging instance, and read the release notes.
- Monitoring. Something that tells you the instance is down or executions are failing — otherwise the first sign is a customer who never got their confirmation.
- Secrets and access. Credentials for every connected system live in this box; treat it like production infrastructure, because it is.
Realistically that is a few hours to set up and an hour or two a month, assuming nothing breaks. If nobody in the business wants that job, n8n’s own cloud removes it and still bills per execution — usually landing between Make and self-hosting on cost.
What self-hosting actually costs
| Line | Typical | Note |
|---|---|---|
| Server | A small cloud VPS per month | 2 vCPU / 4 GB handles most small-business volumes; more for file-heavy flows |
| Database and backups | Included on the same box, plus off-site storage | Postgres in production; test the restore |
| Domain and certificate | Domain cost; certificate free via Let’s Encrypt | Needed for webhooks over HTTPS |
| Monitoring | Free tier is enough | Uptime plus a failed-execution alert |
| Your time | A few hours to set up, an hour or two a month | The real cost, and the one people forget |
Against a per-task bill that grows every month, that is a fixed floor — which is exactly why it wins at volume and loses at ten automations a week.
When self-hosted n8n pays
- Volume. Thousands of runs a month, or long multi-step flows where per-task billing compounds.
- AI at scale. Agent-style workflows that call models repeatedly — per-task fees on top of model costs get expensive fast, and n8n lets you point at your own models.
- Data you cannot send to a third-party cloud. Health, legal, financial or client data under strict terms; self-hosting keeps it on infrastructure you control.
- Agencies. One instance serving many clients, with templates you clone per account, is a far better margin than paying per client per month. This is the most common reason my agency partners move.
When to stay on Zapier or Make
- A handful of simple automations and nobody technical — Zapier, and the cost is trivial.
- You need an integration that only exists as a Zapier app.
- Middling volume with real logic — Make is the balanced choice and nothing to maintain.
- Your CRM already does it. For “form → CRM → text → task”, workflows and webhooks inside the CRM need no middle tool at all; that remains the cheapest automation there is.
Migrating without breaking things
- List every live automation with its trigger, steps and monthly run count. Expect to find a third are unused.
- Rebuild the highest-volume flows first — that is where the saving is.
- Run old and new in parallel for a week with test records, comparing outputs.
- Move webhook URLs last; keep the old flow paused, not deleted, for a month.
- Add failure alerts on day one. Self-hosted means nobody else is watching.
The honest summary: n8n is cheaper per unit of work and more powerful for AI and custom logic; Zapier and Make are cheaper per unit of your attention. Choose by which of those you have less of. And whichever you pick, the webhook doing the actual work is the same — the tool is a convenience layer around it.
Security notes for self-hosting
- Editor behind authentication and, ideally, a VPN or IP allowlist.
- HTTPS everywhere; never run webhooks over plain HTTP.
- Separate credentials per client or system, least privilege, rotated.
- Backups off the server, restore tested.
- Updates monthly, on staging first.
- Audit who can access the editor — a workflow editor is a code-execution environment.
Further reading: n8n self-hosting docs · n8n pricing · Zapier pricing · Make pricing.
FAQ
Is n8n free?
The self-hosted community edition is free to run under a fair-code licence — you pay for the server and your own time. n8n also sells a managed cloud plan and enterprise licences. “Free” therefore means “no per-task fee”, not “no cost”: budget for hosting, backups and maintenance.
Is n8n cheaper than Zapier at scale?
Usually by a wide margin, because n8n bills per workflow execution (or nothing when self-hosted) rather than per task or operation. A flow with twenty steps costs the same as a flow with two. The crossover point for most businesses is a few thousand runs a month, or the moment workflows get long.
Do I need a developer to run n8n?
To use it, no — the editor is visual and the templates are good. To host it, effectively yes: someone has to keep a server updated, backed up and secured, and fix it when an upgrade breaks a node. Use n8n Cloud if nobody in the business wants that job.
Which is best for AI workflows?
n8n, for building agent-style flows where you control models, tools and data, and want to run them at volume without per-task fees. Zapier and Make both have capable AI steps and are quicker to start with; the difference is control and cost as usage grows.
Want the automations designed, built and monitored — on whichever of the three actually fits your volume? See the automation and integration services; for agencies, a shared self-hosted instance with per-client templates is usually the setup that pays for itself fastest.
