What each one actually is
Zapier is the largest automation platform in the category and the easiest to start with. It is vendor-hosted only, it has the biggest connector catalogue anyone has built, and its unit of billing is the task — roughly one action step operating on one item. A five-step workflow that runs on an order consumes five tasks.
n8n is a node-based automation tool that you can run on n8n’s cloud or self-host on your own infrastructure. It is source-available under a fair-code licence rather than open source in the OSI sense, which matters for compliance sign-off and is worth reading before you assume either way. Its unit of billing is the execution — one workflow run, regardless of how many nodes fire inside it. The same five-step workflow consumes one execution.
That single difference is the comparison. Everything else is preference; this is arithmetic.
n8n vs Zapier, line by line
| Axis | n8n | Zapier |
|---|---|---|
| Hosting | Self-hosted on your own infrastructure, or n8n’s cloud | Vendor cloud only |
| Billing unit | Per workflow execution, however many steps run inside it | Per task — roughly one action step on one item |
| Where customer data goes | Stays inside your own infrastructure when self-hosted | Passes through the vendor’s cloud |
| Connector coverage | Large, plus an HTTP node that reaches anything with an API | The largest catalogue in the category, by a distance |
| Logic | Branching, merging, loops, batching, and a code node | Paths, filters, formatter, and code steps |
| Bulk and scheduled work | Comfortable processing thousands of rows in one run | Each row is a task, which is where the bill comes from |
| Debugging | Full execution log with the data at every node, and error workflows | Task history and replay; less visibility into intermediate data |
| Who operates it | Somebody owns the server: upgrades, backups, uptime | Nobody — that is the product |
| Learning curve | Steeper. It is a node editor, not a wizard | The gentlest in the category |
| List price | — per execution / per month | — per task / per month |
Both list prices are marked — until we re-verify them. The ratio between the two billing models is the point of this page, and the ratio does not change when the prices do. Pricing to confirm before publish
Pricing reality: the per-task trap
Here is the shape of it without either vendor’s numbers.
A single Shopify order arrives. Your workflow enriches it with the customer’s subscription status, checks stock against the 3PL, writes a row to a sheet, tags the order, and posts to a Slack channel if the customer is a first-time subscriber. That is five tasks in Zapier and one execution in n8n, every time an order lands.
Now make it a subscription brand. Every renewal is an order, so the trigger volume is not your acquisition rate — it is your active subscriber count multiplied by billing frequency, and it grows precisely as the business succeeds. Automation that gets more expensive the better retention gets is the wrong incentive to build a company on.
Three costs that are not on either pricing page:
Trigger latency has a price. Polling triggers on lower plan tiers check on an interval rather than firing instantly. On a Slack notification that is irrelevant. On a failed-payment retry or an inventory hold it is money, because the window in which an intervention still works is measured in hours. Check the trigger type on every workflow that touches revenue, not just the plan tier.
Which steps count is the whole cost model. Filters and paths have historically behaved differently from actions in Zapier’s task accounting, and the details have changed over time. Confirm the current terms before you design a workflow around them, because “add a filter to save tasks” is a plan that either works completely or not at all.
Self-hosting is not free, it is capitalised. A self-hosted n8n needs a server, backups, an upgrade cadence, and a person who notices when it stops. For heavier workloads it wants queue mode and a worker. That cost is real; it is also fixed, which is exactly why it wins as volume rises.
The honest version: below a few thousand runs a month, this is a preference. Above that, the per-task model stops being a subscription and starts being a variable cost of goods.
Migration difficulty
There is no automated conversion between the two. Every workflow is rebuilt by hand, node by node. That is worth saying plainly because it changes how you should decide: pick deliberately at the point where volume makes the answer obvious, rather than drifting into an expensive platform one Zap at a time.
The practical route we use:
- Read the task-usage report. It ranks your workflows by cost and hands you the migration order.
- Leave everything else alone. There is no prize for moving a workflow that runs eleven times a month.
- Build every new order-volume workflow in n8n from today, so the problem stops growing while you work.
- Port the expensive ones in usage order, one at a time, running both in parallel for a cycle before switching the trigger off.
- Reissue credentials rather than copying them. A migration is the right moment to rotate API keys, and copying them across is how a key ends up in two places nobody is tracking.
Budget for the fact that the rebuild is also a rewrite. Most workflows that have been in production for a year contain at least one step that exists because of a problem that no longer exists.
Where Make fits
Make is the third name in this conversation and it sits between the two: vendor-hosted like Zapier, with a visual canvas closer to n8n’s, billed per operation. For cost purposes, per-operation behaves like per-task — it scales with steps, not with runs. If your reason for looking is the billing model, Make does not solve it. If your reason is that you want a better canvas without running a server, it is a reasonable middle.
Who each one is genuinely better for
n8n is the better answer when
- The workflows fire on order or subscriber volume, which for a subscription brand is most of the ones that matter.
- You need branching, loops or batch processing — reconciling a few thousand SKUs nightly, or exploding a bundle into components before an order reaches the warehouse.
- Customer data should not leave your infrastructure, for a compliance reason or a policy one.
- You want the execution log. Being able to see the exact payload at the node that failed, three weeks later, is worth more than it sounds.
- You have an engineer, or an agency, who will own the instance.
Zapier is the better answer when
- The workflows are low-volume and internal: a form submission to a spreadsheet, a Slack alert, a calendar entry.
- The trigger app is long-tail and only Zapier has a connector for it.
- Nobody on the team is technical and nobody is going to become technical.
- An outage would be an inconvenience rather than a revenue event.
Our take
We run self-hosted n8n for client automation, so read this knowing that. The reason is the one above: anything that fires once per order belongs somewhere a step does not cost money, and on a subscription brand almost everything fires once per order.
The recommendation is not “replace Zapier”. It is a boundary. Order-volume, revenue-touching and bulk work goes in n8n, where the execution log makes it debuggable and the billing model makes it affordable. Low-volume internal glue stays in Zapier, where the connector already exists and nobody has to maintain anything. Most brands we work with end up running both on purpose, and the ones who are unhappy are the ones who ended up running both by accident.
If you are not sure which side of the line your workflows fall on, ops automation is the service that draws it, and the audit prices the hours currently going into the manual version.
Disclosure: Pointerflow takes no affiliate revenue, referral fee or partner commission from n8n, Zapier, Make or any other tool named on this site.
Next: Recharge vs Skio for the subscription layer, or how to connect Shopify to ShipBob for the workflow this is usually built to fix.