n8n vs Zapier vs Make: which should a company choose for automation?

What do the three tools actually differ in?

From a distance all three do the same thing: connect apps, react to triggers, run a sequence of steps. The differences start underneath that.

Deployment model. Zapier and Make exist only as the vendor's cloud. There's no option to run either on your own server. n8n has both paths: n8n Cloud and a free self-hosted Community Edition, no execution limit, on infrastructure you control.

Room for code. Zapier has a Code step (JavaScript or Python) on higher plans, with limits on runtime and packages. Make has a Custom JS module on some integrations and a data-transformation tool, but not a full runtime attached to every step. n8n has a Code node available on every plan, including self-hosted for free, with full Node.js or Python and access to data from the whole workflow. That matters when a ready-made integration block doesn't do exactly what you need: in n8n you add ten lines of code, in Zapier and Make you're often looking for a workaround or an extra tool.

Billing unit. This is the difference that hits the bill hardest at real volume, so I give it its own section below.

Interface. Zapier is linear: a trigger and a list of steps underneath, the easiest to pick up. Make is visual and graphical, with routers and branches visible on the canvas, good for processes with real branching logic. n8n is also a node canvas, closer to Make than to Zapier, but with more options packed into each node and the ability to preview data step by step while you build.

Number of integrations. Zapier has the largest ready-made integration library on the market. Make has a solid and growing list. n8n has a somewhat smaller list of native integrations, but makes up for it with the HTTP Request and Code nodes, which connect to any API even without a dedicated node.

How does each one charge, and what does a 20-step process cost on each?

This is where the decision actually gets made, more than the integration list on the homepage.

Zapier counts tasks. A task is a successfully completed action step in a Zap: per Zapier's own wording, "a task is counted whenever Zapier successfully completes a unit of work," including programmatic calls like Zapier MCP tool calls. The trigger is free, as are built-in tools like Formatter, Paths and Filter. Failed actions don't count against the limit. The September 2026 price list (zapier.com/pricing, checked 2026-09-10): the Free plan has 100 tasks a month and only two-step Zaps (one trigger, one action), you can't build a 20-step process on it at all. Pro starts at $19.99 a month billed annually for 750 tasks. Team starts at $69 a month for 2,000 tasks. Self-serve plans scale up to 2,000,000 tasks a month. Overage beyond the plan limit costs 2.5x the base rate on monthly billing or 1.25x on annual, and usage halts once you cross 3x the plan limit until the next cycle.

Make counts operations. An operation is roughly one module run in a scenario: reading data, searching, creating, updating or deleting a record, transforming data, iterating or aggregating rows each typically count as one operation; error-handler and router modules cost nothing. The September 2026 price list (make.com/en/pricing, checked 2026-09-10): the Free plan has 1,000 operations a month, a maximum of two active scenarios, a 15-minute minimum interval between scheduled runs, and a 5-minute maximum execution time per scenario. The paid Core, Pro and Teams plans all start at 10,000 operations a month billed annually, at $9, $16 and $29 a month respectively for that same operations floor, they differ in features (log retention, seats, custom variables), and the operations volume scales up independently of the plan via a slider, up to millions of operations a month.

n8n counts workflow executions. One execution is one run from trigger to end, regardless of how many nodes sit in between. The September 2026 n8n price list: Cloud Starter is €20 a month for 2,500 executions, Pro €50 for 10,000, Business €667 for 40,000. The self-hosted Community Edition has no licence fee and no execution limit at all.

That difference in billing unit produces a real difference in practice. Take a 20-step process, say: receive a webhook, pull data from three different APIs, run it through several transformation steps, write it to a spreadsheet, send a notification, handle two conditional branches. These are illustrative estimates, not price-list quotes:

Frequency	n8n	Zapier	Make
Once a day (30 runs/mo)	30 executions, trivial on any plan	roughly 450–570 tasks (15–19 billable steps × 30), fits Pro (750)	roughly 600 operations (20 modules × 30), fits even Free (1,000)
Hourly (720 runs/mo)	720 executions, fits Starter (2,500)	roughly 10,800–13,680 tasks, needs a higher Team tier	roughly 14,400 operations, needs Pro/Teams with the slider raised
Every 5 minutes (~8,640 runs/mo)	8,640 executions, needs Business or self-hosted	roughly 130,000–164,000 tasks, beyond most self-serve plans	roughly 173,000 operations, Teams with the slider pushed high, expensive

The point isn't that Zapier and Make are bad; for a two- or three-step process the gap barely exists. It grows with step count and frequency. A workflow that runs every minute is 43,200 executions a month on its own, which is a good enough reason to price it out on all three platforms before you commit.

When is Zapier the right choice?

When the process is short, the apps are popular and have ready integrations in Zapier, and nobody at the company wants to (or needs to) think about a server. Onboarding is the fastest of the three; a non-technical person can build a simple Zap with no introduction at all. The integration library is the largest on the market, which matters for niche SaaS tools you won't find elsewhere. If a client asks me about connecting two popular apps in a simple scenario, low volume, no sensitive data, I tell them plainly: Zapier is enough, you don't need anyone for this.

When is Make the right choice?

When the process has real branching, somewhere from a few to a dozen or so steps, moderate volume, and the company wants a visual, no-code editor without running a server. Make's canvas shows branching and iteration logic more clearly than Zapier's linear step list. The price-to-operations ratio in the middle segment is reasonable as long as volume doesn't explode. This is a tool I recommend more often than someone who knows me mainly through n8n might expect. If a process fits inside Make without hitting the operations ceiling and without needing data kept on your own server, there's no point paying for more than that.

When is n8n the right choice?

Four situations I see most often with clients:

• Data has to stay inside the company. Documents, personal data, invoices, HR records, customer data. Self-hosted n8n keeps them on infrastructure you control, without sending them to an automation vendor's cloud.
• Volume is growing. At tens of thousands of runs a month, or on schedules that fire every few minutes, the bill for tasks or operations grows faster than the bill for n8n executions, and the self-hosted Community Edition has no limit at all.
• The process has a lot of steps or complex logic. 20, 30, 50 steps in one workflow is n8n's natural territory, because price doesn't grow with node count.
• You need code where the ready-made blocks stop. The Code node with a full runtime is available on every plan, including self-hosted for free. This is the difference clients tend to appreciate only once they hit an integration that can't be built from ready-made blocks.

The price for that: you need someone who can stand up and maintain a server, or pay someone external to do it. That's a real cost, and I don't hide it from clients.

What about AI agents building the automations?

This is where the gap between n8n and the rest of the market is largest today, not smaller. n8n-mcp is a Model Context Protocol server used by more than 143,000 users, letting AI agents (Claude, GPT and others) build, validate, deploy and fix n8n workflows programmatically, with full access to the node catalog and templates. From n8n-mcp's own telemetry: in the week of August 25–31, 2026, agents used it to build 98,122 new workflows, with a median of 7 nodes per workflow, and 74% of them contained a Code node. Agents reach for code in nearly three out of four workflows, exactly where ready-made blocks run out.

Zapier and Make have nothing at that level of integration with agents building automations programmatically, with validation and the ability to iterate. Both platforms have their own AI features (Zapier has Copilot to suggest steps, Make has AI in the editor), but those are assistants for building inside the interface, not a protocol that lets an external agent manage the whole workflow lifecycle: creation, validation, deployment and fixing errors. If a company's plan involves AI agents building and maintaining part of its automations on their own, n8n has the mature tooling for that today, not the competition.

What does migrating between them involve?

It's never one click. Logic has to be rebuilt by hand, because none of these tools imports another's format directly.

From Zapier or Make to n8n. Every step has to be rebuilt: trigger, field mapping, conditions, data formatting. n8n workflows export and import as JSON, which makes versioning and copying between environments easy, but that doesn't help with a migration from another platform; you're building from scratch there either way. The upside is you can often simplify the process at the same time, since a Code node frequently replaces several separate data-transformation steps.

From n8n to Zapier or Make. I see this less often, usually when a company stops wanting to run a server. You need to check whether the logic in a Code node can be rebuilt in ready-made blocks or in a Code step/Custom JS module; sometimes it can't without simplifying the process.

Between Zapier and Make. Also manual, step by step, though both share a similar linear/visual model, so mapping the logic across is usually simpler than moving to n8n.

The typical moment clients decide to migrate to n8n is the first process touching documents or personal data that shouldn't leave the company, or the first unpleasant bill for tasks or operations once volume grew faster than anyone planned.

How do I decide in ten minutes?

1. Does the process handle documents, personal data or customer data that can't go to an automation vendor's cloud? Yes: self-hosted n8n. No: keep going.
2. Does the process have more than a dozen or so steps, or run on a frequent schedule (more than once an hour)? Yes: price it out on Zapier, Make and n8n using the table above before you choose. No: keep going.
3. Do you need custom logic that ready-made blocks can't express? Yes: n8n, or Make with a Custom JS module, depending on how much code you actually need. No: keep going.
4. Does nobody at the company want to administer a server, and is the process simple and infrequent? Yes: Zapier to start, Make if the process branches.
5. Does the plan involve AI agents building and maintaining workflows on their own? Yes: n8n with n8n-mcp, the only one of the three with that level of maturity.

For most of the clients I work with, the first two questions settle it. If both answers are no, Zapier or Make are an honest choice, and I say so directly. Not every process needs n8n, and paying for capabilities you won't use is a bad way to start with automation.