n8n vs Make vs Zapier in 2026: which one your business should actually build on

Most comparisons of these three tools rank them on features nobody chooses on. The feature lists converged years ago; all three connect to the systems a normal business uses, all three can call a language model, all three have a visual builder. What has not converged is how they charge you, and that is the decision that shows up on an invoice eighteen months later.

Last updated: 27 August 2026.

How we compared them

We build client workflows on all three and hold no partnership, reseller agreement, or affiliate arrangement with any of them, so there is no commission riding on this ranking. The criteria are the ones that actually change the outcome of a project:

• Pricing model, not sticker price — how the meter runs as workflows get longer;
• Behaviour under branching and looping, where simple tools break down;
• Escape hatch — what happens when the visual builder cannot express your logic;
• Maintenance burden, including who can fix it when it breaks at 9am;
• Realistic ceiling — the volume at which you outgrow it.

Pricing tiers below were current at the time of writing and all three vendors change them; treat the ranges as orders of magnitude and check the current page before committing.

The comparison that matters

1. n8n

How it bills: per workflow execution. One run of a workflow is one billable unit whether it contains three steps or fifty. Price: Community edition is free to self-host under the Sustainable Use License — internal business use is permitted, reselling it as a service is not. Cloud starts in the low tens of euros per month. Best for: multi-step workflows at volume, anything needing a code step, teams that want to self-host for data residency or cost reasons, agent-style workflows with tool calls. Not for: businesses with nobody willing to open a workflow graph. It is the most capable of the three and the least forgiving.

The per-execution model is the real argument for n8n, and it compounds in a specific way: it removes the incentive to keep workflows short. On per-step billing you find yourself merging steps and skipping the validation and logging that make a workflow debuggable, because each one costs money. On n8n those steps are free, so the workflow you build is the one you actually wanted.

The code escape hatch is the second argument. Every visual builder eventually meets logic it cannot express, and the question is what happens next. On n8n you write a function node and move on.

The honest cost of self-hosting: a small server, plus someone’s attention for upgrades, backups, and the morning something stops. Free is a licence fee, not a total cost.

2. Make

How it bills: per operation — every module execution inside a scenario counts. Price: free tier around 1,000 operations a month; paid tiers from roughly $9–19 a month for meaningful volume. Best for: workflows that branch and iterate, heavier data transformation, teams that want real capability without leaving a visual canvas. Not for: high-volume workflows with many steps each — the operation counter is what eventually pushes companies to n8n.

Make sits where most small businesses should actually start once their automations stop being linear. Its handling of arrays, iterators, and conditional routing is genuinely good, and the visual model stays readable at a complexity where Zapier’s linear list has long stopped making sense.

The catch is the same as Zapier’s, one order of magnitude cheaper: you are still paying by the step, so a well-instrumented workflow costs more to run than a sloppy one. That is a bad incentive, but at small-business volume the absolute numbers are low enough that it rarely matters.

3. Zapier

How it bills: per task — each action step in a Zap. Price: free tier around 100 tasks a month; paid from roughly $30 a month. Best for: simple, linear, low-volume automations; the largest app catalogue of the three; handing a workflow to a non-technical colleague. Not for: anything with real branching, high volume, or long-running logic.

Zapier gets criticised by people building complex systems, which misses what it is for. If your automation is “form submitted → row in a spreadsheet → notify a channel”, Zapier will have it working in fifteen minutes and a colleague with no technical background will be able to read and edit it. That is worth real money in a company with no technical staff.

Its app catalogue is still the widest, and for a long-tail SaaS tool the others have not connected to, Zapier is sometimes the only option that does not involve writing an API integration yourself.

The failure mode is predictable: the Zap grows, filters and paths get added, volume rises, and the bill becomes conspicuous. That is the point to migrate — not before.

Which one to pick

A short decision tree. Read down to the line that matches you.

• Two or three steps, low volume, nobody technical → Zapier. Do not overthink it; the subscription is cheaper than the hours you would spend evaluating alternatives.
• The workflow branches, loops, or reshapes data, and you want a visual builder → Make. This is where most small businesses genuinely belong.
• Many steps per run, or thousands of runs a month → n8n. Run the multiplication above; the answer is usually obvious.
• Data cannot leave your infrastructure → n8n, self-hosted. It is the only one of the three with a serious self-hosting story.
• You need custom code inside the workflow → n8n. The others will force a workaround through an external function service.
• One scheduled job between two systems you control → none of them. A small script on a cron schedule is cheaper and less fragile.
• You do not know your volume yet → start on Make’s free tier and instrument it. Guessing volume is how companies end up on the wrong pricing model.

A one-week test before you commit

Platform decisions get made on blog posts and regretted on invoices. This costs nothing and takes a week.

• Day 1 — Write the process down step by step, including what happens when something is missing or malformed. Count the steps. This number is your billing multiplier on two of the three platforms.
• Day 2 — Get the real monthly volume from the source system. Search the inbox, count the records. Do not estimate.
• Day 3 — Multiply steps × volume. If the product is above roughly 10,000, price all three properly before choosing; below that, pick on ease of use.
• Day 4 — Check every system in the chain against each platform’s connector list. A missing connector is a hard constraint that outranks every preference above.
• Day 5 — Build the happy path on a free tier. Any of the three will get you that far in an afternoon.
• Day 6 — Break it on purpose. Feed it a malformed record, a missing field, a duplicate. How the platform surfaces that failure is what you will live with daily.
• Day 7 — Show it to whoever will own it. If they cannot describe what it does from looking at it, you have picked the wrong tool for your team regardless of what the pricing says.

If day 6 is where your prototype falls apart, that is not a platform problem. Exception handling is the majority of the work in any real automation, and it is the reason quoted build prices vary by 5x for the same one-sentence description — covered in how much AI automation costs.