n8n vs Make vs Zapier — when we pick which

We get this question on every scoping call: 'Should we use Zapier? Or Make? Should we self-host n8n?' The honest answer: it depends on three things — ownership, complexity, and the volume curve. Here's the decision tree we use internally.

Zapier bills per task. At a few hundred tasks per month it's pocket change. At 100,000 tasks per month, it's $300+. At a million, you're paying enterprise pricing for what amounts to glue code.

Make.com is 5–10× cheaper per operation, which makes it a strong middle-ground for marketing ops where iteration count matters more than developer time.

n8n self-hosted is essentially free at the runtime layer — a $20 droplet handles serious volume — but you trade money for the operational overhead of running your own service.

Rule of thumb: if monthly task count crosses 50k and you have any engineering capacity, n8n self-hosted will pay for itself inside a quarter.

Zapier owns your runtime. If they change pricing or deprecate a connector, you have no recourse. Workflows live inside their UI; export is limited.

Make is similar but gives you cleaner exports and a more transparent pricing curve.

n8n is open source. Workflows are JSON. You can put them in git, code-review changes, deploy via CI, run multiple environments. You can also write custom JavaScript nodes when a connector is missing or brittle.

If your business runs on these workflows, ownership matters. We default to n8n for anything that touches sensitive data, runs at high volume, or has compliance implications.

Zapier handles complexity well enough for 80% of cases — Paths, Filters, Sub-Zaps, Lookup tables, Schedule triggers. When you hit a wall, the wall is hard.

Make's visual scenario builder beats Zapier on iterators, aggregators, and explicit error branches. If your workflow has nested loops or needs to roll up partial failures, Make is the cleanest no-code option.

n8n wins on truly complex logic: state machines, conditional retry strategies, custom error queues, recursive workflows. And when it can't, you write a JavaScript node and move on.

Zapier is the fastest path from idea to running automation. The 6,000+ connector library is unmatched. For a one-week scope to wire 5 SaaS apps together, nothing beats it.

Make takes 1.5–2× as long to build but the result is more maintainable.

n8n self-hosted has a real setup cost — infra, auth, backups, observability. We typically scope a 1-week buffer just for the platform itself. After that, building workflows is fast.

Less than 10k tasks/month, breadth matters more than depth, you want it live yesterday → Zapier.

Marketing-ops complexity, multi-step scenarios, mid-volume → Make.

Sensitive data, high volume, custom logic, you want git-versioned workflows → n8n self-hosted.

We'll often run two of these in parallel — Zapier handling speed-critical glue work while n8n handles the volume-critical core. The tools aren't enemies; they have different sweet spots.

All three platforms now have AI nodes. OpenAI, Anthropic, and a few open-source options are first-class steps. Which means the choice between platforms is increasingly less about 'where the AI lives' and more about 'who owns the runtime around it.' That ownership question doesn't go away — it gets sharper.