n8n vs Zapier vs Make: what actually breaks at scale

Every comparison of these three tools measures the same things: number of integrations, price per task, whether there is a free tier. Those are the differences you notice in week one. They are not the differences you live with.

We have inherited automation estates built on all three. What separates them shows up later — usually the first time something fails quietly and nobody notices for a fortnight.

The honest summary

Zapier is the easiest to start and the hardest to operate at volume. Make gives you far more control over data shape at the cost of a steeper interface. n8n gives you the most control and the option to self-host, and asks you to take on more responsibility in exchange.

None of that is a recommendation yet, because the right answer depends on a question most comparisons never ask.

What actually breaks

Error handling is where the difference lives

In Zapier, a failed step stops that run. You get an email. If nobody reads the email, the failure is invisible — and because each Zap is its own island, there is no single place that shows you the health of everything.

Make surfaces errors better and lets you build explicit error routes per module, which is a genuine step up. You can define what happens on failure rather than just being told about it.

n8n gives you per-node error outputs, retries, and the ability to wire a failure into an actual process — post to Slack, write to a table, trigger a compensating workflow. It also gives you full execution history you can inspect and replay.

This is the single biggest practical difference between them, and no comparison table lists it.

Cost curves diverge sharply

Zapier charges per task, where a task is roughly one step-execution. This is fine at low volume and becomes punishing quickly: a five-step Zap running on a thousand records is five thousand tasks. Businesses routinely discover this the month after they automate something successful.

Make's operations model is cheaper for the same work, and its iterator/aggregator handling means you can process a batch inside one scenario rather than triggering per record.

n8n's cloud plans charge per workflow execution rather than per step, so a workflow that loops over five hundred records is one execution. Self-hosted, the marginal cost is your own infrastructure.

If you are moving meaningful volume, this is not a rounding difference. It changes which automations are economically worth building.

Data shape

Zapier's model assumes one item flowing through steps. Anything involving arrays, nested objects or batching gets awkward fast, and people end up bolting on Code steps to reshape data — at which point the no-code benefit has evaporated but the per-task pricing has not.

Make handles arrays natively and is genuinely good at data transformation.

n8n treats items as a list by default, which is the right primitive for the work most businesses actually automate: process these two hundred rows, not this one row.

Version control and review

Zapier has no meaningful diff or review process. Changes are live edits.

Make has version history.

n8n workflows are JSON. They can live in Git, be diffed, be reviewed, and be promoted between environments. If your automations touch billing, contracts or customer data, this stops being a nice-to-have.

Self-hosting

Only n8n offers it. This matters for two reasons that come up repeatedly: data residency requirements that forbid customer data transiting a third-party SaaS, and cost at high volume.

It also means you now own uptime, upgrades and backups. That is a real cost, not a free win.

The question that actually decides it

Not "which is most powerful," but: who maintains this in eighteen months?

If the answer is a marketing generalist who automates a handful of things and needs it to be obvious — Zapier. The premium is real but so is the accessibility, and an automation nobody can maintain is worth less than a slightly expensive one that anyone can.

If the answer is an ops person who is comfortable with data structures and wants control without running infrastructure — Make.

If the answer is someone technical, or the workflows touch regulated data, or volume is high enough that per-task pricing hurts — n8n.

The migration question

We get asked to move estates from Zapier to n8n fairly often, almost always triggered by a bill rather than a capability gap.

Two things worth knowing before you start. First, it is rarely a one-to-one port: Zaps built around single-item processing usually want redesigning around batches, and that redesign is most of the work. Second, migrate the expensive workflows first. The long tail of low-volume Zaps often costs less to leave where it is than to move.

What we would say to a business choosing today

Pick for the maintenance story, not the feature list. Every one of these tools can do the thing you are currently trying to do. The difference shows up on the morning a workflow has been silently failing for two weeks and someone has to work out why — and at that moment, error visibility and execution history matter more than integration count.