n8n vs Make vs Zapier: Which One Should You Actually Use for Client Work in 2026

Most comparisons of these three platforms are written for people automating their own business. Client work is a different problem. You are not choosing a tool for one company with one budget and one tolerance for downtime — you are choosing something you will deploy across ten or twenty accounts, hand over to non-technical staff, and still be maintaining eighteen months from now when the person who signed the contract has left.

That changes the answer. The tool that wins on a feature comparison chart is frequently the wrong one to build a delivery business on.

The only difference that matters: how each platform counts

Everything else in this comparison is secondary to the billing unit. Get this wrong and you will quote a client a monthly platform cost that turns out to be four times higher in production.

Zapier bills per task. A task is one successful action. If your workflow has eight action steps and runs a thousand times a month, you have used eight thousand tasks. Triggers do not count, and neither do built-in logic steps like Filters, Paths, Formatter, and Delay — a detail that genuinely does soften the blow, because it means you can add branching without inflating the bill.

Make bills per module call. Every module that fires consumes a credit, including triggers and iterators. Make renamed this unit from "operations" to "credits" in late 2025, so most tutorials and Reddit threads you find still use the old word for the same thing. A five-module scenario running a thousand times costs roughly five thousand credits.

n8n bills per execution. One complete workflow run is one execution, regardless of whether it contains three nodes or forty. Sub-workflow runs do not count separately.

Now put a real workload through all three. Ten-node workflow, one thousand runs a month:

Platform

Consumption

Unit

n8n

1,000

executions

Zapier

~9,000

tasks

Make

~10,000

credits

That is the same automation costing wildly different amounts depending on where you built it. And the gap widens with every node you add, because only n8n’s meter is indifferent to workflow complexity.

The practical rule: the more steps in your typical workflow, the more n8n’s pricing model favours you, and the more punishing Zapier becomes. Short two-or-three-step automations barely feel the difference. Long enrichment-and-routing chains feel it enormously.

What each one actually costs

Prices below are approximate and both Make and n8n adjust them periodically — verify against the vendor pages before you quote a client.

Zapier has a free tier at 100 tasks a month, limited to two-step Zaps, which is not enough to build anything real. Professional starts around $20/month on annual billing for 750 tasks, scaling up a slider to millions. Team runs roughly $69/month annually for 2,000 tasks and adds shared connections and folders.

The number that catches people out: 750 tasks is nothing. A five-step Zap running 150 times a month consumes the entire Professional base tier. One workflow. Build a second and you are into overage.

Make is free up to 1,000 credits with two active scenarios. Core lands somewhere around $9–12/month annually for 10,000 credits, Pro around $16–21, Teams around $29–38. Roughly three to five times cheaper than Zapier for comparable volume, though less than the headline suggests once you account for the fact that Make counts more things than Zapier does.

n8n Cloud starts at €20/month annually for 2,500 executions, Pro at €50 for 10,000, Business at €667 for 40,000. There is no permanent free tier anymore. The self-hosted Community Edition is free software with unlimited executions — you pay only for the server, which is somewhere between $5 and $7 a month on a basic VPS if you are running light, considerably more if you are running production workloads for paying clients.

One trap specific to n8n Cloud: when you hit the execution cap, workflows stop. No overage billing, no grace period, no warning. Everything halts until the billing cycle resets. If you are running client automations on a capped plan, that is an outage you are responsible for.

The licensing issue nobody mentions

This is the part that matters most for agencies and that almost every comparison article skips.

n8n is not open source in the standard sense. It uses the Sustainable Use License — a "fair-code" licence adapted from Elastic’s — which grants free rights to use, modify and redistribute, with the significant catch that use is limited to your own internal business purposes.

The line n8n draws is clearer than most people assume, and it is worth stating precisely because the vague version scares agencies off unnecessarily.

What is allowed: commercial consulting and support. Building workflows and automations for paying clients is expressly permitted. You do not need a commercial licence to be an agency that implements n8n.

What is not allowed: white-labelling n8n and selling access to it, or hosting n8n and charging people money to use it. n8n names both explicitly. The test is whether n8n itself has become the product you are selling.

So: building a client’s lead-routing workflow on their own self-hosted instance, fine. Standing up one instance, putting twenty clients on it, and billing them monthly for access, not fine without a commercial licence. Plenty of agencies have drifted from the first model into the second without noticing, because it happens gradually. If that is the direction your delivery model is heading, read the licence before you scale rather than after.

Zapier and Make have no equivalent restriction, because the client holds their own account and you are not redistributing anything.

Which one for which client

Use Zapier when the client will maintain it themselves. This is the honest case for Zapier and it is stronger than the pricing makes it look. If you are building something a marketing coordinator has to modify after handover, Zapier’s interface, documentation, and app library are meaningfully ahead. The premium you pay is a training and support cost you are not paying instead. For a client running a handful of short workflows, the difference between $20 and $9 a month is irrelevant next to two hours of your time explaining a Make router.

Use Make when workflows are complex and the budget is tight. Branching, iterators, data transformation, visual debugging of a scenario with thirty modules — Make handles this better than Zapier and more accessibly than n8n. It is the middle option in the truest sense: cheaper than Zapier, gentler than n8n, and capable enough for most SMB work.

Use n8n when volume is high, logic is complex, or the workload is AI-heavy. Self-hosted n8n removes execution cost entirely, which matters enormously for agent workflows where you are already paying per LLM call. It also gives you code nodes, proper version control, and an escape hatch for anything the visual tools cannot do. The cost is that you now own a server, and when it breaks at 2am on a client’s lead-routing workflow, that is your problem.

The mistake that costs agencies the most

It is not picking the wrong platform. It is estimating usage from workflow count instead of execution volume.

Every platform lets you build unlimited workflows. Dormant workflows cost nothing. One badly designed polling workflow can consume an entire plan by itself — a Gmail poll checking every five minutes burns through n8n Cloud Starter’s 2,500 executions in under two weeks, and it will do comparable damage on Make.

Before you quote, do this: take the workflow, multiply expected trigger volume by counted steps, add 30% for reruns and testing, then check that number against the next tier’s break-even. Then build the thing and watch real usage for a week before you commit anyone to annual billing.

Two design habits that cut consumption sharply regardless of platform:

Replace polling with webhooks wherever the source app supports them. Polling burns runs whether or not there is anything to process; a webhook fires only when something actually happened.

Loop inside one workflow rather than triggering a workflow per item. Teams migrating from Zapier tend to carry the one-item-per-trigger habit across and then wonder why their execution count exploded.

What we actually do

We run all three, chosen per client rather than by preference. The split in practice is roughly: Zapier where a non-technical team owns the automation after handover, Make for mid-complexity work where cost matters, self-hosted n8n for high-volume and AI agent workloads where execution pricing would otherwise dominate the bill.

The platform is genuinely the smaller decision. Whether the workflow is designed to fail loudly, whether someone is monitoring it, and whether the client can tell you within a day that it stopped working — those determine whether an automation is still running a year later. Most of the broken automations we inherit were not broken by the wrong tool. They were built with no error handling and nobody watching.

If you are weighing this up for your own business and would rather not run the migration math yourself, that is the kind of thing our workflow and marketing automation services cover — including auditing what you are currently paying for and whether it needs to be on that platform at all.

Pricing figures verified against vendor pages and third-party breakdowns as of September 2026. All three platforms change pricing regularly; check current rates before committing to annual billing.