FA©26
30 April 2026·3 min readAutomationn8nTooling

n8n, Make, Zapier, GoHighLevel — how I actually choose

I get asked to settle this argument a lot, usually in the form of “which one is best”. None of them is best. They fail differently, they cost differently at volume, and the wrong one only reveals itself six months in — which is exactly when it's most expensive to move.

The four questions

  1. Who has to open this later — me, or the client?

  2. Does it run in seconds, or does it wait on something for days?

  3. What happens the first time a step fails at 3am?

  4. What does this cost when the volume is ten times what it is today?

Every real decision I've made falls out of those four. The feature comparison tables never come into it.

Where each one earns its place

Zapier is for connections that need to exist by tomorrow morning. Enormous catalogue, near-zero setup, and it's the right answer far more often than engineers like to admit — a one-hop integration that works today beats an elegant one that ships next sprint.

Make is for workflows somebody has to understand by looking at them. When the branching gets genuinely complicated, a scenario drawn like a circuit diagram explains itself in a way a list of steps never will. I reach for it when the audience for the automation is a person, not a server.

n8n is the daily driver. Self-hostable, so the data stays where it should and the cost curve doesn't bend upward with volume. Real code inside nodes when a transformation gets awkward. It's what I use when an automation stops being a convenience and starts being infrastructure.

GoHighLevel is a different category and gets compared unfairly. It isn't a workflow engine — it's the CRM the client actually logs into. Pipelines, follow-ups, AI agents and voice assistants live next to the customer record, and that adjacency is the whole point. Nobody wants a beautiful workflow that reaches into a CRM they can't see.

The thing that actually decides it

Error handling. Every one of these tools can move data between two APIs on a good day. What separates them in production is what happens on a bad one: whether a failure is visible, whether it retries, whether a half-finished run leaves a client's data in a state somebody has to clean up by hand.

Before I ship an automation I make it fail on purpose — kill the credential, feed it a malformed record, let a rate limit hit — and I watch what it does. If I can't see the failure without going looking for it, it isn't finished.

When to stop automating

There's a point where a workflow has too many branches, too much state, and too much business logic buried in nodes nobody wants to touch. That's not an automation any more; it's an application that happens to be written in a workflow tool.

When the automation starts needing its own database, it wanted to be an app all along.

Recognizing that moment early is most of the job. Both of my own products crossed that line — and the honest thing to do at the line is to stop adding nodes and start writing software.