Automation tools look interchangeable until the first invoice. The plans are not expensive in the abstract; they are expensive in the specific workflow shape the billing model punishes. A workflow with twelve branches and one that runs one linear action per trigger can cost wildly different amounts for the same number of records.
This page compares the three most common starting points — n8n, Zapier, and Make — by how they meter work, because that is the part that decides.
The short answer
Zapier charges per task, so it is cheap for simple, low-branch workflows and expensive for anything with loops. Make charges per operation, which makes branching comparatively cheap and makes long chains visible. n8n is open-source and self-hostable, so the marginal cost is infrastructure rather than tasks — the trade is engineering time and maintenance.
What each meter counts
| Tool | Meter | A branch to a second app costs | Predictability |
|---|---|---|---|
| Zapier | Tasks (one per action step) | 1 task | High |
| Make | Operations (per module run) | 1+ operations | Medium |
| n8n | Self-hosted executions / cloud executions | No per-task charge | Depends on host |
The row about branching is the one that surprises people. On a task-based plan, a workflow that checks a condition and then takes one of two actions still consumes a task for the action; on an operation-based plan the router itself is metered too. Neither is wrong; they just make different workflows cheap.
Where Zapier earns its price
The catalog is the product: thousands of connectors, including applications that nobody else supports. For a team automating a SaaS stack with no engineering time, the premium buys the long tail and a support experience.
The failure mode is using it for data pipelines. A workflow that iterates over a list of records can consume hundreds of tasks in a single run, and the task counter does not care that the work was trivial.
Where Make wins
Make's visual model exposes the flow: you can see the operations a scenario will consume before you run it, and the router is a first-class node. That makes iterating and looping possible in a way that task-based pricing discourages.
The cost is a learning curve. The interface rewards people who think in data structures, and it punishes anyone who wants a simple two-step automation.
Where n8n pays off
Self-hosting moves the bill from tasks to a server, which is a structural advantage once volume grows. You also get access to the workflow JSON, version control, and the ability to write a code node without a plan limitation.
The price is real: a server to maintain, backups, upgrades, and a queue for executions that run long. A team without someone responsible for that server should not choose this path, however attractive the invoice looks.
The three questions that decide
First, how much does a single run cost in each meter? Count the action steps for your busiest workflow, including error branches. Second, how often does it loop? Loops are where task pricing becomes unpredictable. Third, who maintains it when it breaks at 2am?
The answers usually settle it: a marketing team with a 300-task automation should stay on Zapier; a product team moving thousands of records should self-host; a mixed team with visual flows and budget pressure usually lands on Make.
The migration cost
Switching tools later is more expensive than the pricing difference for most teams. Workflows are not portable: triggers, authentication, and error handling are all tool-specific. Choosing on a $20/month difference and migrating in a year is usually a net loss.
The practical approach is to price the workflow you have now, at the volume you expect in six months, and accept that the answer will change. The transferable asset is the process description, not the workflow.
A worked example
Consider a lead-routing automation: form submission, enrich the company, decide a segment, post to Slack, create a CRM record. That is five steps with one branch.
On task pricing, each run consumes roughly one task per action, so the branch does not double the cost — the unchanged path does. On operation pricing, the router is metered, so branching is visible but not punitive. Self-hosted, the cost is one execution regardless of branches, which is why complex logic tends to migrate there first.
None of those numbers are surprising in isolation. What surprises teams is the tail: the workflow that runs a hundred times a day with a loop that fires three times per record. That is the shape to model before choosing a plan.
Frequently asked questions
Is self-hosting always cheaper? At low volume, no — a server plus maintenance can exceed a small SaaS plan. It becomes cheaper as executions grow and complex logic becomes common.
Can I mix tools? Yes, and many teams do: a SaaS tool for the long-tail connectors, a self-hosted one for the high-volume pipeline.
Which is easiest for non-technical users? Zapier, then Make. n8n assumes comfort with a config file and a terminal. See automation tools for the rest of the field.
Last checked: 9 November 2026. Prices and meter definitions change; check the current plan page before committing.
Related: self-hosting automation, AI workflow builders, and automation tools.