n8n vs Zapier for small businesses: what it actually costs to own your automation

The per-task trap

Zapier bills by the task, and Zapier defines a task as a successful action step. Triggers do not count. Filters, paths and Formatter steps do not count either. That sounds generous until you look at what it means in practice: a workflow that receives an invoice, extracts the fields, writes a row to your accounts sheet, and sends a confirmation email is spending three tasks every single time it fires.

So "tasks per month" is not the same as "how often my automation runs", and the gap between those two numbers is where the bill hides. Take a modest business doing fifty tasks a day. That is not fifty automations firing — with a three-action workflow it is closer to seventeen actual runs, which is a small company quietly getting its admin done. Fifty tasks a day is 1,500 tasks a month.

Here is what that costs on Zapier's Professional plan as of August 2026, checked against Zapier's published plan rates. Annual billing is roughly a third cheaper than monthly, and the prices are in US dollars regardless of where you are:

Our fifty-tasks-a-day business lands on the 1,500 tier: $39 a month if it commits to a year, $58.50 if it does not. That is $468 or $702 a year to move its own data between its own systems. The free plan, for completeness, is 100 tasks a month — enough to prove a concept and nothing more.

Now the same workload on n8n. n8n is source-available under its Sustainable Use Licence and can be self-hosted. What you are buying is a small server: a DigitalOcean basic Droplet with 2GB of RAM is $12 a month and the 4GB box is $24, while Hetzner Cloud has comparable shared-vCPU instances at roughly €4 to €7. Any of those will chew through a small business's automation without noticing.

The number that matters is not that £5 beats $39. It is that the server cost does not move when your volume does. Fifty tasks a day and five hundred tasks a day cost the same. On Zapier, 5,000 tasks a month is $89 and 10,000 is $129, still on annual billing. The more useful your automation becomes, the more you pay for it — which is a strange incentive to build into the tool that is supposed to make you efficient.

What self-hosting actually involves

Here is the part most comparisons skip. Self-hosting is not free. You are not paying Zapier, but you are taking on a server, and a server is a thing somebody has to own.

In practice that means a Linux box running n8n in Docker, behind HTTPS, with backups of the workflows and the database. n8n ships updates often, and while you do not have to take every one, letting it drift for a year is how you end up on a version with a known problem. Something needs to watch that the service is actually up, because a workflow that has silently stopped is worse than no workflow at all — you will carry on assuming the invoices went out.

None of that is difficult if it is your job. All of it is friction if it is not. The honest framing is that self-hosting converts a predictable monthly bill into a small, irregular maintenance obligation, and you should be clear-eyed about who absorbs it. If that is you and you enjoy this sort of thing, the maths is overwhelming. If it is you and you do not, you have swapped a $39 invoice for a background worry, which is a worse trade than it looks on a spreadsheet.

The third option is that someone else absorbs it, which is the arrangement I usually end up in: I do the build and the setup, hand over documentation, and the client either runs it themselves or keeps me on a small retainer to watch it. The setup is the expensive part and it happens once. What follows is cheap.

When Zapier is the right answer

I would rather talk you out of a server than sell you one you did not want, so here are the cases where Zapier genuinely wins and I will say so.

Your volume is tiny. If you are under 100 tasks a month, the free plan covers it and building infrastructure for that is silly. Even at 750 tasks, $19.99 a month annually is not worth restructuring your life over.

Non-technical colleagues need to build their own. This is the strongest argument for Zapier and it is not close. Someone in operations who can wire up their own Zap without asking anyone is a real capability, and n8n's editor — while good — expects more of you. If four people across your business each want to automate their own corner of it, pay Zapier.

You depend on an obscure integration. Zapier's connector library is enormous. If the one piece of software your business runs on is only supported there, that single fact decides it, and no amount of cost modelling changes the answer.

Nobody will own the server. If there is genuinely no one — not you, not a contractor, not a retainer — then a managed service is the correct choice. An unmaintained server is a liability, not a saving.

Make, formerly Integromat, sits in much the same territory. I build on it happily when a client is already there. The argument in this article is about pricing models and ownership, not about anyone's software being bad.

The part that is not about money

Stop paying Zapier and your automations stop. Not degrade — stop. Two years of accumulated logic, the edge cases you discovered the hard way, the small corrections that make the thing actually work: all of it lives in an account you rent. You can export a Zap for reference, but you cannot take the running system with you.

Self-hosted n8n workflows are JSON on a machine you control. You can export them, read them, move them to a different server, or hand them to a different developer. If we stop working together, nothing switches off and nothing needs migrating.

That is the same position I take on everything else here: you own the code, the design and the accounts, and you are never locked into my infrastructure because there is no such thing. Workflow automation is built the same way for the same reason.

What this looks like in practice

These are the shapes I run daily in my own businesses, with the task arithmetic attached, because the abstract version of this argument is not convincing.

Automated invoicing. On a schedule, pull the records that are due, generate the invoice, send it, mark it raised. Four actions per invoice. Forty invoices a month is 160 tasks before anyone has chased anything.

Payment matching. Every incoming payment gets read, matched against outstanding invoices, and either closed off or flagged to a human. Three actions per payment, and payments do not arrive tidily one per invoice. Another 150 or so.

Scheduled posting. Content queued once and published across channels on a timetable. Two posts a day across two channels is 120 publishing actions a month, plus whatever formatting and logging sits around them.

AI steps inside the workflow. An OpenAI node reading a messy email and returning clean structured fields for the next step to file. Worth noting that Zapier's own AI steps bill at their own rates rather than as plain tasks, so this is the line item most likely to surprise you. If the AI work is the point rather than a step, that is a different kind of build.

Add up just those and an entirely ordinary small business is somewhere between 1,500 and 2,000 tasks a month without doing anything clever. That is the $39 to $49 tier, annually billed, for automation nobody would describe as ambitious. The first genuinely useful thing you add pushes you into the next bracket.

The verdict, by size

One person, under 100 tasks a month. Zapier free. Do not overthink it and do not let anyone sell you a server.

One to three people, under 750 tasks, nobody technical. Zapier Professional. Roughly $240 a year on annual billing is a fair price for never thinking about infrastructure.

Any size, 1,500 tasks a month and climbing. Self-hosted n8n. At this point you are paying $468 to $702 a year for something a €5 server does, and the gap widens every time the automation earns its keep. The build cost is recovered inside the first year and the running cost stops tracking your success.

A team where several non-technical people build their own. Zapier, or n8n with someone maintaining it for you. What you must not do is self-host and leave four people who cannot debug it depending on it.

Data you cannot send to a third party. Self-hosted, and it is not really a decision. If the information cannot leave your infrastructure, it cannot leave your infrastructure.

The short version: below about a thousand tasks a month, Zapier is a reasonable purchase and the convenience is worth the money. Above it, you are renting something you could own, and the rent goes up precisely as the thing becomes more valuable to you. Work out your real task count first — actions, not runs — and the decision usually makes itself.