n8n vs Make vs Zapier: choosing an automation platform in 2026

Which should you choose: n8n, Make or Zapier?

Choose Zapier if the people who will own the automations are not technical and the workflows are short. Choose Make if one operations person builds branching scenarios that reshape a lot of records and cost per step matters. Choose n8n if you have a developer or IT person, run long or high-volume workflows, or must keep data on your own servers.

We build on all three. Our default is the platform you already pay for, set up on your own account, because a migration rarely pays for itself on price alone. The rest of this article explains the cases where it does.

How does each platform charge? (prices as of October 2026)

Zapier charges per successful action step, Make charges per credit (roughly one per module action), and n8n charges per full workflow run. A 10-step workflow therefore costs 1 execution in n8n, about 10 credits in Make and up to 9 tasks in Zapier, depending on which steps are free.

Enterprise tiers on all three are quoted by sales.

Zapier: tasks

The Zapier pricing page (opens in a new tab) counts a task each time Zapier successfully completes an action. Triggers do not count, failed actions do not count, and built-in tools such as Formatter, Filter, Paths, Delay and Looping do not count either. Prices scale by task volume, from 750 tasks to 2 million on Professional and Team. Annual billing is 33% cheaper than monthly. If you switch on pay-per-task, runs continue past your limit at 1.25 times the base rate on annual plans and 2.5 times on monthly plans, up to 3 times your subscription before Zapier pauses.

Two AI items sit outside the simple picture. Zapier Agents are billed in activities, a separate allowance (400 a month on Free, 1,500 on Professional). Zapier MCP calls draw from the task pool, at 2 tasks per tool call according to the same page.

Make: credits

Make moved its billing unit from operations to credits in 2025. Per the Make pricing FAQ (opens in a new tab), most module actions (read, search, create, update, delete, iterate, aggregate) use 1 credit, while the Router and error handler modules use none. Features that run on Make's own AI Provider can use more than 1 credit, priced on tokens. The Make credits help page (opens in a new tab) says that if you connect your own AI provider on a paid plan, credits follow operations and you pay the provider for tokens.

Two details matter at volume. Extra credits cost 25% more than credits included in the plan, and Core tops out at 300,000 credits a month, both per Make's November 2025 plan update (opens in a new tab). Make's pricing page sits behind bot protection that our checker could not pass, so the $9 Core figure comes from Make's own comparison page (opens in a new tab). Confirm the figure for your credit volume on the live slider.

n8n: executions

The n8n pricing page (opens in a new tab) defines an execution as "a single run of your entire workflow", regardless of step count or data volume. Prices are listed in euros, and annual billing saves 17%. Business is self-hosted only for now. Companies under 20 employees can apply for a startup plan at 50% off Business. The free Community Edition has no execution cap from n8n, but you run the server.

Example: one workflow priced three ways

This is an illustrative example, not a quote. Take a lead-routing workflow that runs 5,000 times a month: a webhook trigger, a filter, a formatting step, an OpenAI classification step using your own API key, a CRM search, a CRM update, a Slack message and a Google Sheets row.

• Zapier: the trigger, filter and formatter are free, leaving 5 billable actions per run, or 25,000 tasks a month. That is above the 20,000-task tier, so the 50,000-task Professional tier at $289 a month billed annually, or the smaller tier plus pay-per-task.
• Make: the filter and formatting live between modules, so 6 modules run, about 30,000 credits a month. Price that volume on Make's slider.
• n8n: 5,000 executions fit the Pro plan at €50 a month billed annually, or Community Edition on a small server you maintain.

In all three cases the OpenAI usage is billed separately by OpenAI. Per-step pricing punishes long workflows at volume. For 20 short workflows that each run a few hundred times a month, the gap mostly disappears.

A production version of this workflow, with enrichment, scoring, routing and a drafted reply for the rep, is the inbound lead qualification agent blueprint. It runs on n8n or Make and lists the build cost and running costs.

Which is easiest for a non-technical team?

Zapier is the easiest to learn and the easiest to hand to a non-technical owner. Make is learnable by a motivated operations person. n8n assumes comfort with JSON, expressions and APIs.

Zapier's editor is a vertical list of steps that reads like a form. Copilot drafts a Zap from a plain description, with unlimited messages on paid plans. Branching uses Paths, which costs no tasks. When something breaks, the person who built it can usually find and fix it.

Make draws scenarios on a canvas, with routers for branches and iterators and aggregators for lists. Once someone understands how bundles of data move between modules, it is the most expressive no-code builder of the three. Arrays are where new builders get stuck: one invoice with 12 line items behaves differently from 12 invoices.

n8n also uses a canvas, but every node shows its JSON input and output, and most real builds use expressions. The Code node runs JavaScript or Python, and the HTTP Request node calls any API that has documentation. That makes n8n the most capable of the three for technical builders. A non-technical owner will struggle to maintain what a developer built.

What AI and agent features does each have?

All three can call a language model as a step, run agents and talk to Model Context Protocol servers as of October 2026. For a single model step inside a workflow, any of them works. For agents with retrieval and custom tools, n8n gives the most control.

n8n Pro also includes metric-based evaluations for AI workflows, which matter once a model step makes decisions you need to measure. On any platform, we recommend connecting your own OpenAI or Anthropic account so model spend shows up on its own invoice. We bill model usage at cost.

The bigger question is whether you need an agent at all. Our guide to AI agents vs workflow automation vs chatbots has a 5-question checklist.

Can you self-host, and where does your data live?

Only n8n can be self-hosted. Make and Zapier are cloud services, so your data passes through their infrastructure, and the questions become which region, which agreement and which sub-processors.

The n8n Community Edition uses the Sustainable Use License (opens in a new tab), which allows use "only for your own internal business purposes" or non-commercial use. Running your own automations is fine. Reselling n8n as a hosted service is not. The paid Business plan adds SSO, LDAP, Git version control and multiple environments on your own servers, and Enterprise adds log streaming, an external secret store and longer data retention.

Self-hosting moves work onto your team: the server, database backups, upgrades, security patches and uptime monitoring. Budget a few hours a month for it, or put it in a care plan. It pays when data must stay inside your infrastructure or when execution volume makes cloud pricing expensive.

Make runs on AWS and organizations live in a named region, visible in your account address (for example eu1 or us1). Zapier is cloud-only. If you handle health, financial or other regulated records, get the vendor's data processing agreement, region options and any required contract (such as a HIPAA business associate agreement in US healthcare) confirmed in writing before a single record flows through.

How do they handle errors and monitoring?

Make and n8n give you the most control over what happens when a step fails. Zapier gives you the simplest recovery: replay the failed run.

• Zapier keeps a Zap history of every run. Manual replay is on every plan (on Free, for errored runs only), and Autoreplay (opens in a new tab) on Professional and above retries errored runs automatically. It does not replay runs that halted safely.
• Make lets you attach error handler routes to any module, with Ignore, Resume, Commit, Rollback and Break directives. Break can store an incomplete execution to retry later, and error handlers cost no credits.
• n8n lets each node retry on failure or route failures down an error output, and an Error Trigger workflow runs whenever another workflow fails. The execution log shows the input and output of every node.

Whatever the platform, we set up the same 4 things: every failure posts to one Slack or Teams channel with a link to the run; scheduled workflows send a daily heartbeat so silence is noticed; error counts get a weekly review; and the runbook names one owner per workflow. Our automation care plans ($1,500 to $6,000 a month) run that routine for you.

How hard is it to move between platforms?

Plan for a rebuild. There is no importer between these platforms that we would trust in production, and each one models data, loops and errors differently.

A migration that does not lose data follows the same 6 steps:

1. Inventory every workflow: trigger, apps, step count, monthly runs, owner.
2. Switch off what nobody uses, and do not migrate it.
3. Recreate app connections on accounts the company owns, not on a contractor's login.
4. Rebuild the highest-volume workflows first, because that is where the saving is.
5. Run old and new side by side for 1 to 2 weeks and compare outputs record by record.
6. Move webhook URLs, switch off the old version and keep the old account read-only for 30 days.

Our workflow integrations work prices this at $1,500 to $6,000 per integration, or $8,000 to $20,000 for a bundle. A move pays when a few high-volume workflows dominate a per-step bill, or when compliance requires self-hosting. It rarely pays for a handful of low-volume Zaps.

What do we pick, by company type?

Our picks follow the people who will own the system after we leave, then volume, then data rules.

When is none of the three the right tool?

Sometimes the best automation platform is the one you already own, or plain code.

• Logic that lives inside one system. HubSpot workflows, Salesforce Flow and help-desk rules are cheaper and easier to govern than an external platform for work that never leaves that system.
• Very high volume or low latency. Event streams with millions of records a day or sub-second responses belong in code with a queue, not in a visual builder.
• Transactional ERP integrations with strict audit needs. Posting journals or inventory moves usually calls for the ERP's own integration tools, an enterprise integration platform or tested code.
• Production agents that need custom evaluation and permissions. These are usually better built in code with a model provider's SDK. See custom AI agents.
• No owner. Any of the three decays without someone reading the alerts. Decide who owns it before you pick a logo.

If you want help choosing, a business process automation engagement starts with a written inventory and a platform recommendation, and the n8n glossary entry covers the basics.

How we checked these facts

We read each vendor's own pages on October 28, 2026. Prices change often, so if a vendor page disagrees with this article, the vendor page wins.

• Zapier pricing (opens in a new tab): plans, task rules, pay-per-task, Agents and MCP.
• Zapier help: replay failed Zap runs (opens in a new tab): manual replay and Autoreplay by plan.
• Make pricing FAQ (opens in a new tab) and Make credits help (opens in a new tab): credit rules and AI usage.
• Make vs Zapier (opens in a new tab) and Make vs n8n (opens in a new tab), Make's own comparison pages: the Core entry price. Treat their comparisons of rivals as marketing.
• Make plan update, November 2025 (opens in a new tab): extra-credit pricing and the Core cap.
• n8n pricing (opens in a new tab): plans, executions and the startup discount.
• n8n Sustainable Use License (opens in a new tab): Community Edition terms.