n8n vs Make vs Zapier: the honest comparison after a dozen projects shipped

TL;DR — verdict in 30 seconds

Short on time for 2,000 words? Decision in four lines. For 90% of serious B2B projects: self-hosted n8n, full stop. For a marketing team automating their own life without tech help, under ten zaps: Zapier, good enough. For visual-heavy workflows with a business team that needs to prototype fast: Make, but watch the per-operation pricing.

The rest of the article explains why, on which criteria, with concrete 2026 cost calculations and two real production cases.

Zapier — when it’s OK and when it cracks

Zapier remains the best-known tool, the easiest to get started with, and — to be clear — the right pick in some situations. The problem is we usually see it installed in the wrong ones. When an SMB executive tells us “we already have Zapier, we just want to extend it,” our first reflex is to look at the monthly invoice. It often shows usage already at the edge of what per-task pricing allows.

Zapier's model: you pay per task. One task = one step of a zap that executes. If your zap runs five steps (fetch a lead, enrich it, create it in the CRM, send a welcome email, ping Slack), it consumes five tasks. Multiply by 1,000 leads/month and you're at 5,000 tasks. On Professional you're already at the cap. Double the volume, you move to Team, then Company at hundreds of euros per month depending on complexity. Note: Zapier adjusts its pricing every 6 to 12 months — the figures here should be reconfirmed on the date you read this.

At 5,000 tasks/month, Zapier roughly costs €75 to €100/month. Self-hosted n8n on a Scaleway VPS at ~€8/month handles the same volume without breaking a sweat. At 50,000 tasks/month, Zapier shifts to Company at several hundred euros — still €8 to €20 on the n8n VPS. The gap widens fast, and at a certain volume it becomes absurd.

Real production case: the Toshify project — automated qualification of drivers on WhatsApp for Cabify Argentina. The workflow has to handle a wide mix of exotic cases inside a single readable flow: driver profiles with prior history on other platforms, qualification in Argentine Spanish (voseo) vs Iberian, conditional handoffs based on CRM profile, asynchronous document checks. On Zapier, this would have meant chaining seven or eight zaps with manual state passing between them — exploding monthly bill, impossible maintenance at six months. On n8n, it's a single versioned workflow. See the full case →

When Zapier remains the right pick. A five-person marketing team that wants to connect HubSpot, Slack, Google Sheets, and Calendly without IT support. Under ten active zaps, each two to four simple steps, moderate volume, no particularly sensitive data. In that context, the effort to install and maintain n8n isn't justified — it's technical debt for nothing. Zapier does the job, and that's perfectly fine.

Make — the right tool we misuse

Make has one asset that neither Zapier nor n8n really has: a pleasant visual editor for modeling complex workflows with branches, error handling, retries. When you want to graphically represent “if X then Y else Z, and if Y fails retry three times,” Make is more readable than n8n and far more expressive than Zapier.

The trap: Make charges per operation, not per scenario run. One operation = one API call or one elementary module. A scenario that loops over 100 items to process them consumes 100+ operations per execution. On naturally iterative workflows (list processing, bulk updates, catalog syncs), the bill explodes fast — and the pricing is less readable than Zapier's.

The other limit: no self-hosting. Your data flows through Make servers (they offer an EU residency option, which is GDPR-acceptable), but you can't fully isolate processing inside your own infrastructure. For clients with a strict security policy or absolute data-localization requirements, this is still a blocker.

When Make remains the right pick. A business team (marketing, ops) that needs to prototype quickly, with moderate volume and no hard GDPR/security constraints. Make is also excellent for workflows that essentially consume external APIs (HubSpot, Notion, Airtable) rather than heavy transformation logic. In that case, the visual editor wins out over n8n's flexibility.

n8n — why we install it on 90% of our projects

On 90% of serious B2B projects, n8n wins over the other two for four structural technical reasons. Not operator preference — four concrete points that all line up the same way at decision time.

Self-hostable and fair-code. n8n is not fully open source — its license is called “fair-code” (Sustainable Use License) — but the code is readable, modifiable, and self-hosting is explicitly allowed at no cost. Concretely: it runs on a European VPS (Scaleway, Hetzner, OVH) at €8-20/month. Your credentials, workflows, logs: everything stays with you. For a client subject to a security audit by their own customer, that's often the criterion that closes the discussion. It's the stack we deployed at Toshify: every driver conversation flows through their own n8n instance, never through third-party SaaS servers.

Flat pricing regardless of volume. Whether you process 1,000 or 1,000,000 operations a month, the cost of self-hosted n8n barely changes (just the VPS tier if you really saturate). For projects that consume volume — lead qualification, continuous database syncs, event scoring — it's an economic game-changer. The math above: at 50,000 tasks/month, n8n costs an order of magnitude less than Zapier.

JavaScript Code Node when no-code isn't enough. No no-code tool covers 100% of needs. For the remaining 5-10%, n8n has a Code Node where you write JavaScript directly. You're never blocked by a tool limit — when something's missing, you code it in ten lines instead of upgrading tiers or switching tools.

Real production example: at Heex, a business weighting matrix had to be applied on top of the raw score returned by Gemini — closing weighted more than discovery, non-zero budget qualification mandatory, penalty if the sales rep didn't ask a key question. No no-code node exposes that level of logic. In half an hour inside a Code Node, it was written, tested, and readable by the client during review. On Make or Zapier, that's a five-step chain with broken recalculations on every prompt edit.

The real downsides. The learning curve is steeper than Zapier or Make. The UI is less polished than Make's. Debugging can be painful when a workflow crashes in prod at 2 AM. And self-hosting implies having someone who knows how to manage Docker on a VPS — so not a purely business team. Without an in-house technical team or a partner who operates the stack, n8n isn't the right pick: you need someone behind it. That's exactly what our AI Automation service covers on the Mynos side.

The criterion everyone forgets: exit

In most tool decisions we see in audits, no one asked: what happens if I want to leave this tool in 18 months? It's the most important long-term criterion, and the one that most sharply separates the three.

With Zapier, you're locked in. Zaps are an internal proprietary representation. There's no structured export usable elsewhere. When you leave, you start from scratch on another tool. If Zapier changes its pricing, closes a major integration, or evolves in a direction that doesn't suit you — all of which has already happened — you pay for the migration in full redevelopment time.

With Make, it's better: you can export a scenario as JSON, but the format is Make-specific. No other tool can import it as-is. Useful for backup, not really for portability.

With self-hosted n8n, your workflows are standardized JSON, your Postgres database holds encrypted credentials, and you can migrate everything to another n8n instance in minutes. If tomorrow the n8n.io cloud service is acquired, changes model, or disappears, you depend on nothing — you run on your VPS. Business continuity is entirely in your hands.

This criterion never shows up during the “I need to automate, fast” phase. It invariably shows up 18 to 24 months later, the day you want to switch for one reason or another. Anticipate it now.

Our 5-question decision tree

When a client asks which to pick, we run five questions in this order. Depending on the answers, the choice is almost automatic.

1. What monthly task/operation volume are you expecting? Under 1,000/month, anything works — go for simplicity (Zapier). Above 5,000/month, self-hosted n8n is unbeatable economically.

2. Is your data subject to GDPR with strict localization requirements? If yes, self-hosted n8n in an EU region. Make with the EU residency option is acceptable. Zapier (US-based by default) should be avoided if strict data localization is required.

3. Do you have an in-house tech team or a partner who can operate a VPS? If not, forget self-hosted n8n — go for n8n.cloud (the SaaS version) or Make. Zapier stays valid for simple usage.

4. Will your workflow need to evolve frequently with complex business logic? If yes, n8n (Code Node). Make as second choice. Zapier will trap you fast.

5. What monthly budget can you accept just for the tool? Under €50/month, Zapier basic or self-hosted n8n on a small VPS. Under €200/month, wide choice possible. Above that, it signals volume or complexity — and n8n becomes the right pick again.

The reference n8n stack in 2026

For a self-hosted n8n instance in B2B production, here's the architecture that scales — deployable in-house with an engineer familiar with Docker.

European VPS: Scaleway DEV1-S (~€8/month) for small volumes, or DEV1-L (~€20/month) starting at 50k tasks/month. Hetzner and OVH are excellent alternatives — the criterion is EU localization and the perf/price ratio.

Docker containerization: n8n runs in a container, with Postgres separate for workflow and credential persistence. Docker Compose orchestrates both. Daily DB backup to an S3-compatible bucket (Scaleway Object Storage or OVH Storage).

Reverse proxy: Caddy or Traefik in front of n8n for automatic TLS (Let's Encrypt). Incoming webhooks are HMAC-SHA256 signed when the sender supports it, otherwise filtered by IP allowlist.

Monitoring: Uptime Kuma on the same VPS to watch the n8n instance's health. Slack notifications on incident. n8n logs exported via Loki on more mature setups.

Updates: Watchtower for minor automatic patches, manual upgrade for major versions (n8n can break workflows when nodes change their API in major releases).