n8n vs Zapier vs Make: What the Same Three Workflows Actually Cost in a Year

Pricing pages for automation tools are written to be compared on the wrong axis. Zapier shows you tasks, Make shows you credits, n8n shows you executions, and each page quietly implies that its number is the generous one. The truth is that the three units measure completely different things, so the same workflow can be cheap on one platform and absurdly expensive on another without either vendor lying.

I run a one-person business on a mix of these tools, and I have migrated the same automations between them more than once. The most useful exercise I ever did was taking three real workflows, counting exactly how each platform would bill them, and adding it up for a year. This article walks through that exercise so you can do it with your own numbers.

One caveat before the math: plan prices and tier boundaries change often. I use approximate figures from the pricing pages at the time of writing, and I round generously. Treat the dollar amounts as orders of magnitude and check current pricing before you commit. The counting rules are the part that stays stable, and the counting rules are what decide your bill.

The three billing units, explained plainly

Zapier bills tasks. A task is one successful action step. The trigger does not count. Built-in utility steps like Filter, Formatter, and Paths do not count either. So a Zap with a trigger and four actions uses four tasks every time it runs. If a step loops over a list, each item’s actions count separately.

Make bills credits (historically called operations). Nearly every module that runs counts, including the trigger. The detail that catches people: a polling trigger that checks for new data uses a credit each time it checks, even when it finds nothing. Iterators and routers are modules too. Some AI modules consume more than one credit per run.

n8n bills executions on its cloud plans. An execution is one run of a workflow, from trigger to finish, no matter how many nodes it passes through. A workflow with thirty nodes and a loop over five hundred items is still one execution. In current versions, a polling trigger that finds nothing new generally does not start an execution. And if you self-host the Community Edition, there is no execution bill at all, just the server and your time.

Already you can see the shape of the answer. Zapier punishes many action steps and loops. Make punishes frequent polling and long scenarios. n8n barely cares how complex a workflow is, only how often it runs.

The three workflows

These are simplified versions of automations I actually run. They are deliberately ordinary.

Workflow A: lead intake. A form submission arrives by webhook. The automation creates a contact in the CRM, posts a message to a Slack channel, sends a welcome email, and appends a row to a tracking sheet. Four actions. About 400 submissions a month.

Workflow B: nightly sync. Every night, fetch about 300 rows from a spreadsheet, and for each row look up a matching record in a database and update it. Two actions per row. Runs 30 times a month.

Workflow C: invoice watcher. Check an inbox every five minutes for emails with invoice attachments. When one arrives, save the file to cloud storage and send a notification. About 40 matching emails a month.

Nothing exotic. Most small businesses have something like each of these.

Counting it out

Zapier

• A: 4 tasks × 400 runs = 1,600 tasks.
• B: 1 fetch task plus 2 tasks × 300 rows, so about 600 tasks per night × 30 nights = roughly 18,000 tasks.
• C: polling is free on Zapier, since triggers do not count. 2 tasks × 40 matches = 80 tasks.

Total: roughly 19,700 tasks a month. That lands you on a Professional plan tier around 20,000 tasks, which at the time of writing runs into the hundreds of dollars a month. Over a year, you are looking at something in the range of a few thousand dollars. Notice that workflow B, the boring nightly sync, is more than 90 percent of the bill.

Make

• A: webhook trigger plus 4 modules = 5 credits × 400 = 2,000 credits.
• B: trigger, fetch, and iterator, plus 2 modules × 300 rows, about 603 credits per night × 30 = roughly 18,100 credits.
• C: polling every five minutes is 12 checks an hour, 288 a day, about 8,640 a month, each costing a credit whether or not an email arrived. Add 2 × 40 for the actual matches: roughly 8,720 credits.

Total: roughly 28,800 credits a month. Make’s per-credit pricing is much lower than Zapier’s per-task pricing, so even with more units, the bill lands in the tens of dollars a month. Call it a few hundred dollars a year. But look at workflow C: an automation that does real work 40 times a month is spending nearly a third of the budget asking an inbox “anything new?”

n8n Cloud

• A: 400 executions.
• B: 30 executions. The 300-row loop happens inside each one.
• C: about 40 executions, since empty polls do not start one.

Total: roughly 470 executions a month. That fits comfortably in n8n Cloud’s entry plan, which at the time of writing is in the neighborhood of twenty-odd euros a month. A couple of hundred a year.

n8n self-hosted

Zero platform fees. A small VPS that runs n8n well for this load costs somewhere between five and fifteen dollars a month, so sixty to a hundred and eighty a year, plus your time. We will come back to the time.

The rough yearly picture

If you stopped reading here, the conclusion would be “Zapier is ten times more expensive, use n8n.” That conclusion is half right, and the wrong half is where people get burned.

Redesign changes the bill more than switching vendors

Here is the same exercise after an hour of redesign, staying on each platform.

Workflow B, rewritten. Most databases and many SaaS APIs accept bulk upserts. Instead of looping over 300 rows with two steps each, fetch the rows and send them in one request using a Code step or a single HTTP call. On Zapier, that is about 2 tasks a night, 60 a month instead of 18,000. On Make, it is a handful of modules, maybe 4 credits a night.

Workflow C, rewritten. Instead of polling the inbox, set up a mail rule that forwards invoice emails to a webhook address, or use the mail provider’s push notification if one exists. On Make, that removes the 8,640 empty checks entirely.

After the redesign:

• Zapier: 1,600 + 60 + 80 = roughly 1,740 tasks a month. That fits in a low Professional tier. The yearly bill drops from thousands to a few hundred dollars.
• Make: 2,000 + about 120 + about 120 = roughly 2,240 credits a month. That fits in the cheapest paid tier, well under a couple of hundred dollars a year.
• n8n: essentially unchanged, because it was never billing the loop or the empty polls.

The gap between Zapier and n8n just shrank from roughly tenfold to less than double. The biggest cost lever was not the vendor. It was not looping per row on a platform that bills per step, and not polling on a platform that bills per check.

The costs that are not on the pricing page

Your time, on every platform. Building workflows takes roughly similar time in all three for simple cases. Zapier is fastest to get something working, Make is fastest for complex branching once you know it, and n8n rewards comfort with JSON and a bit of JavaScript. If you value your time at any real hourly rate, a few extra hours of learning curve can erase a year of subscription savings.

Self-hosting time. A self-hosted n8n instance is not free; it is prepaid in attention. Updates, database growth, the occasional breaking change in a node, TLS certificates, monitoring whether the container is actually up, and backups. In my experience that is an hour or two a month in a calm period and a lost evening a few times a year. You also own the data, including execution history that can quietly grow the database, which is why backing up n8n without losing runs is a real task, not a checkbox.

Retries and errors. Failed steps that retry can consume units on metered platforms. A flaky API that forces three retries per run triples those steps’ cost on Zapier or Make. On n8n Cloud, a retried execution is another execution. Watch error rates as a cost metric, not just a reliability one.

AI steps. All three now offer AI actions. On metered platforms, those may cost more than a normal step, and you still pay the model provider if you bring your own API key. An “AI summary” step on a high-volume workflow can dominate a bill that was otherwise tiny.

Plan limits beyond units. Minimum polling intervals, number of active workflows, concurrent executions, premium app access, and team seats vary by tier. Sometimes you upgrade not because you ran out of tasks but because you need a feature, and the per-unit math stops mattering.

Licensing. n8n’s self-hosted edition uses a fair-code license that allows internal business use but restricts offering n8n itself as a service to others. For most solo operators and small teams automating their own work, that is fine. If you are building automations for clients on your own instance, read the license before you price your offer.

How to run this math on your own workflows

1. List each workflow with its trigger type (webhook, schedule, or polling), the number of action steps, whether it loops, and how many times it runs per month.
2. Count Zapier tasks as action steps times runs, with loops multiplied by item count. Ignore triggers, filters, and formatters.
3. Count Make credits as all modules times runs, plus one credit per polling check whether or not it finds data. Loops multiply the modules inside them.
4. Count n8n executions as runs. That is it.
5. Find the biggest line item on each platform and ask whether a redesign would remove it: bulk API calls instead of loops, webhooks instead of polling, batching instead of per-event runs.
6. Add your time, honestly, including maintenance if you self-host.

A spreadsheet with these columns takes twenty minutes and tells you more than any comparison chart.

So which one is cheapest?

For a small number of simple, low-volume workflows, all three are cheap, and Zapier’s speed of setup often makes it the best value. You are paying a premium per unit, but you are buying hours back.

For workflows with loops, lots of steps, or high volume, n8n is structurally cheaper because it does not bill per step. n8n Cloud gives you that without the server; self-hosting makes sense if you already run infrastructure and do not mind owning another service.

Make sits in between: much cheaper per unit than Zapier, excellent for complex visual logic, and expensive only when you let polling triggers and long iterators run unchecked.

The real answer, though, is that the cheapest platform is the one whose billing unit matches the shape of your work. If your automations are wide and shallow, per-task billing is fine. If they are deep and loopy, pay per execution. And whichever you choose, the hour you spend redesigning your most expensive workflow will probably save more than the migration you were planning.