Use Zapier for under 50 monthly tasks and zero engineering capacity. Use Make for visual builders, complex multi-branch logic, and teams that live in spreadsheets. Use n8n when you need self-hosting, AI nodes, complex branching, or have over 5,000 ops/month — its cost curve flattens to near-zero at scale. Lock-in risk: Zapier (high) · Make (medium) · n8n (none — you own the code).
Every other week, an SMB owner pings us asking some version of the same question: "We're paying Zapier $300 a month and it keeps growing. Should we switch to n8n? Or Make? Or just live with it?" This piece is the answer we send. It is honest about where each tool wins and where each tool will quietly bleed you. Numbers come from 60+ real workflows NexFlow has shipped or audited in the last 18 months — we have run all three platforms in production, often side by side for the same client during a migration.
If you read nothing else, the cost table further down and the lock-in section near the end are what actually decide most SMB choices. The rest is detail you can come back to.
The honest cost comparison at three real volumes
The three platforms price on different units, which makes apples-to-apples hard. Zapier prices per task (one node firing once). Make prices per operation (similar but counts data-fetch ops separately). n8n self-hosted has no per-operation cost — you pay for the server. The table below uses Zapier's task model as the baseline because it is the unit our customers think in.
| Monthly volume | Zapier | Make | n8n self-host |
|---|---|---|---|
| 1,000 tasks / month | US$29 (Starter) | US$9 (Core) | ~US$5 (VPS) |
| 10,000 tasks / month | US$119 (Pro) | US$29 (Pro) | ~US$15 |
| 100,000 tasks / month | US$799+ (Company) | US$199 (Teams) | ~US$50 |
| 1,000,000 tasks / month | Custom (enterprise) | US$1,499+ | ~US$200 |
Two things matter about that table. First, the gap between Zapier and n8n is small at the bottom and very large at the top — for a business doing 100k tasks/month, Zapier costs 16× what self-hosted n8n costs. Second, Make sits in a sweet spot under 10k tasks/month if you want a visual builder without engineering capacity. Above 10k, n8n is decisively cheaper.
The n8n "self-host" cost line includes only the VPS. It does not include the engineer who maintains it, sets up backups, handles SSL renewal, and updates n8n on the security patches that ship roughly fortnightly. For a business without an ops engineer, add ~US$200–400/month in managed-hosting fees (Railway, Render, or a managed n8n provider) and the curve still beats Zapier at scale, just not by 16×.
Where each tool actually wins
Zapier wins on: time-to-first-automation
Nothing beats Zapier for a non-technical user who needs to wire two SaaS tools together this afternoon. The Zap editor is the product-design equivalent of a wide flat road — no ambiguity, no concepts, two dropdowns and you're done. We have built Zapier prototypes for clients in 12 minutes that took us four hours to reproduce in n8n.
Zapier also has the largest app library by a margin — over 7,000 integrations as of 2026. For obscure SaaS tools (the kind that don't have a public REST API), Zapier is often the only option, period. Their app partnership team does the integration work the long tail will never do for itself.
Make wins on: visual complexity
Make's editor (formerly Integromat) renders workflows as a connected graph rather than Zapier's linear list. For a workflow with three branches, a router, an aggregator, and parallel execution legs, Make is the easiest of the three to understand at a glance two months later. Maintainability is an underrated metric.
Make's operations are also slightly cheaper per unit than Zapier's, and its free tier (1,000 ops/month) is meaningful — we have clients running real low-volume workflows on Make Free indefinitely.
n8n wins on: everything else, if you can run it
n8n is the only one of the three that is open-source, self-hostable, and ships a JavaScript code node out of the box. That sounds like a feature list. It is actually a sea change in what's possible:
- You can run it next to your data. Healthcare clients with patient data that cannot leave their VPC self-host n8n on the same network as their EMR. Zapier and Make cannot do this without an enterprise contract.
- AI workflows feel native. The OpenAI, Anthropic, Ollama, and LangChain nodes are first-class. You can ship a retrieval-augmented generation pipeline in n8n in an afternoon. Zapier's AI nodes work for simple use cases but route every request through Zapier's infrastructure.
- The code node solves the long tail. Every complex workflow eventually hits the "no node for that" wall. In n8n, you drop in a Code node and write JavaScript or Python. In Zapier and Make, you wait for the integration team or build a custom app, which is months of work.
- The pricing model is sane. No per-task cost. The constraint is your server, not your invoice.
Lock-in risk — the part nobody calculates
Workflow automation is sticky software. Once a business has thirty workflows running on Zapier, switching platforms means rebuilding all thirty, testing them all, and migrating data. The platforms know this and price accordingly — Zapier's price increases over the last five years have outpaced inflation by 3-4×, and customers grumble but stay.
| Lock-in dimension | Zapier | Make | n8n |
|---|---|---|---|
| Export workflows as JSON | No | Yes (Make-only re-import) | Yes (standard schema) |
| Self-host option | No | No | Yes |
| Open-source engine | No | No | Yes (fair-code license) |
| Bring your own database | No | No | Yes (Postgres, MySQL) |
| Custom code execution | Paid add-on | Limited (Tools module) | First-class (JS, Python) |
n8n's open-source posture is the biggest single argument for choosing it. Even if n8n the company shut down tomorrow, the engine keeps running on every server it was deployed to. Zapier and Make do not offer that.
AI workflows in 2026 — the differentiator that wasn't there in 2023
The category has moved fast since 2023, and 2026 is the year AI nodes stopped being a checkbox and started being the reason people pick a platform. The three platforms differ sharply here:
- n8n ships native nodes for OpenAI (including the Realtime API and Assistants), Anthropic Claude (with prompt caching exposed as a parameter), Cohere, Hugging Face inference endpoints, Ollama for self-hosted models, and a generic LangChain node. Vector stores are first-class: Pinecone, Qdrant, Weaviate, Supabase pgvector. You can build a production RAG pipeline without leaving the editor.
- Make shipped a credible OpenAI module in 2024 and an Anthropic module in 2025. Vector stores landed in late 2025. The experience is solid but feels two beats behind n8n.
- Zapier's AI offering as of 2026 is centred on "Zapier Agents" — a higher-level abstraction that hides the model choice and prompt engineering. It is the right product for a marketing manager who wants to "use AI to summarise leads." It is the wrong product for a NexFlow client who wants to control the prompt, choose the model, and audit the output.
The rule of thumb we apply: if your workflow's value depends on you owning the prompt, the model, and the data path, you want n8n. If you just want "AI somewhere in the flow," Zapier or Make is fine.
The migration question — "we're on Zapier, should we move?"
The question we get most often. The honest answer in four bullets:
- Yes, if you're paying more than US$200/month, hitting Zapier's task limits often, or running workflows complex enough that you've started leaving comments in zap step names.
- Yes, if any of your workflows touch AI, PII, or regulated data. The cost of an audit-failed workflow on Zapier dwarfs the cost of migration.
- No, if you have under ten active zaps, no engineering capacity, and the bill is under US$150/month. The migration cost will eat your savings for a year.
- Maybe, if you're in the middle. Audit your top five most-expensive zaps, see if they'd be 3+ nodes in n8n, and decide on that.
- Zapier wins on simplicity for under 50 tasks/month and loses on cost above 5,000 ops/month.
- Make is the visual-builder sweet spot for ops and RevOps teams, especially with complex branching.
- n8n's edge is self-hosting plus AI nodes plus the absence of per-task pricing — it's the only one of the three you can run on your own infrastructure.
- Lock-in risk is the cost nobody plans for. Zapier (high), Make (medium), n8n (effectively none).
- Migration off Zapier pays back in under 8 weeks for businesses above US$200/month in Zapier spend.
- If your workflow depends on owning the prompt, model, and data path — pick n8n.
Frequently asked questions
Is n8n really free?
Self-hosted n8n is free under a fair-code license — free to use, free to modify, free to deploy commercially, with one restriction: you cannot sell n8n itself as a service. NexFlow uses n8n in client work daily and ships every workflow under that license. n8n Cloud (the hosted version) is paid; it starts at US$20/month for a small business plan.
Does Make have an AI agent feature comparable to Zapier Agents?
Make rolled out "Make AI Agents" in early 2026 — comparable in capability to Zapier Agents and arguably better integrated with the rest of the Make graph. It's worth a look if you've already standardised on Make.
Can I run n8n on Cloudflare Workers / Vercel / a serverless platform?
Not really. n8n is a long-running Node.js process with persistent storage; it does not fit cleanly into a serverless function model. The natural deployment shapes are: a small VPS (DigitalOcean, Hetzner, Linode), a managed container (Railway, Render, Fly.io), or Kubernetes for enterprise. For a single-team SMB, a US$15/month Hetzner box and Caddy in front of it is the path of least resistance.
What about Pipedream, Activepieces, Windmill?
Three honourable mentions in 2026. Pipedream is closest to "Zapier with code" — strong for developers, weaker for non-technical users. Activepieces is an open-source Zapier clone, more polished than n8n in some ways, smaller integration library. Windmill is the developer-power-user choice: write workflows as scripts, get a UI for free. None of these has the SMB-friendly AI node story n8n does in 2026, but Activepieces is worth watching closely.
Stuck on a platform decision?
Book a 15-minute map with NexFlow — US$50, credited to a build. We'll look at your top five workflows and tell you in 15 minutes whether to migrate, stay, or rebuild. No deck, no pitch.
Sources & method
- Cost figures sourced from zapier.com/pricing, make.com/pricing, and self-host estimates based on n8n's official deployment docs (docs.n8n.io/hosting), as of May 2026.
- Migration timing and ROI averages drawn from NexFlow's internal records of 32 Zapier-to-n8n migrations completed between Jan 2024 and Apr 2026. Median migration: 18 days from kickoff to production cutover.
- Feature comparison validated against each platform's 2026 changelog through 1 May 2026.
- n8n license is the Sustainable Use License (fair-code), not strict open-source (OSI). The distinction rarely matters for SMB use; we flag it for completeness.