n8n, Zapier or custom code: choosing how to automate

"Should we use Zapier, n8n, or just get something built" comes up the moment a business decides to automate anything, and it usually gets answered by whichever tool the person asking has already heard of. That is not a terrible way to decide, but it skips a real question with a real answer.

What each one actually is

Zapier connects apps you already use — your form, your CRM, your invoicing, your spreadsheet — through triggers and actions, with almost no setup. It has the largest library of ready-made connections of the three, and for a common pairing of well-known tools it is running within the hour.

n8n does roughly the same job with more control. It can branch on conditions, run a script step when nothing off-the-shelf fits, and — because it is open source — it can be self-hosted instead of paid for per task. The cost of that control is a steeper setup, and if you self-host it, a server that needs someone to own it.

Custom code is a program written for your process specifically, with no platform underneath it. It has no ceiling on what it can do and no built-in library of connections either — everything it talks to has to be built by hand.

The question that actually decides it

Not "which is more powerful." All three can do most simple jobs. The question is where your job stops being simple.

Do the tools you are connecting already have ready-made integrations? If your form, your CRM and your invoicing tool are all reasonably common, Zapier or n8n already has the connection built. Custom code would mean building and maintaining that connection yourself for no reason.

Does the logic branch in ways a workflow builder struggles with? Simple rules — this happens, so do that — fit any of the three. Genuinely conditional logic with several branches and exceptions still usually fits n8n, which is built for exactly this, before it justifies custom code.

Does the volume make per-task pricing expensive? Zapier charges by the number of tasks run each month, and a workflow that fires thousands of times can get costly. That is the honest case for n8n's self-hosted option, or occasionally for custom code — but only once the volume is real, not while you are still testing whether the automation is worth having at all.

Is there a genuine gap — a system with no API, or a piece of logic specific to how you operate? This is the actual case for building something, and it is rarer than it looks. Most "we need something custom" requests turn out to be a workflow that Zapier or n8n already covers, described in language that made it sound bespoke.

Start on the platform. Only pay for custom code once you can name the specific thing the platform cannot do — not because it sounds more serious.

The part that gets skipped

Self-hosting n8n is presented as the budget option, and the licence is free. What is not free is the server it runs on, the security patches it needs, and the fact that if it goes down at 6pm on a Friday, someone has to notice and fix it. For a business with nobody who owns infrastructure, n8n's own cloud hosting or Zapier's subscription is usually cheaper than it looks, because the alternative's maintenance cost is real and simply uncounted.

That is worth saying plainly: we would rather see a business pay a monthly fee for n8n Cloud or Zapier than self-host a server nobody is responsible for. The saved subscription is not a saving if a broken automation goes unnoticed for a week.

A sensible order

Write down the workflow the way it actually happens today, including the manual step you are trying to remove. Check whether Zapier or n8n already has a connector for every system involved — most common business tools do. If it does, build it there, and measure whether it is actually used before doing anything more elaborate.

Only once that is running, and either the volume or a genuine logic gap makes the platform the wrong shape, is it worth talking about something built specifically for the job. The same test applies as with any automation decision: can you describe exactly what should happen, including the exceptions? If yes, a platform will very likely already do it. If the honest answer keeps changing, no tool — bought or built — will hold still under it.