n8n vs Make.com vs Zapier in 2026: the honest comparison for a 4–10 person ops team

If you run operations for a four-to-ten person team, the question we keep getting is which automation platform earns its keep in 2026: n8n, Make.com, or Zapier. The answer is not “it depends.” It is: the choice is binary between n8n self-hosted and Make.com Pro, and almost never Zapier at this size. Which of the two wins depends on a single condition.

The condition is whose time you are spending on the platform. If your operator is the owner-on-Saturday-night, n8n self-hosted wins on cash, gives you data-residency control, and gives you agent-step extensibility nobody else offers. If your operator is salaried and their hours are billable, Make.com Pro wins on cash-and-time combined and gets you to a working scenario faster than either alternative.

This is the longer version of why.

The volume profile we’re costing

Every comparison piece collapses if it doesn’t pin the workload. The shape we’re costing is what we typically see in a four-to-ten person agency or SMB ops team:

• Around fifty active automations, mostly notification/sync/triage shapes.
• About 5,000 monthly “operations” worth of work on Make’s unit. On Zapier’s unit that’s roughly 1,500 tasks, because Zapier counts each successful step in a multi-step Zap separately, and a typical fifty-automation portfolio averages three to four steps per Zap.
• Five of those automations include at least one agentic step calling Claude or GPT, typically classification, summarisation, or a small reasoning loop before a downstream action.

This profile is what changes the comparison. At ten times this volume the answer shifts toward enterprise tiers; at one-tenth, the free tiers will hold and the comparison is moot.

The 12-month total cost picture

All three vendors price the same workload differently. Pulling from the published pricing pages on 26 Apr 2026:

Two things worth surfacing from this table.

First, the n8n self-hosted line is doing the work in this comparison, and it depends on someone on the team being willing to keep a small VPS running. That is real labour, not free. Budget thirty minutes a month for backups and updates. At a €60/hour cost-of-time rate that is another €360/yr, which means n8n self-hosted’s true cost-of-money-plus-time is closer to €420–480.

That number does not beat Make.com Pro at €175/yr. The cash-and-time comparison flips on whose time you’re spending. If your team has a salaried operator whose hours come out of a billable budget, Make.com Pro wins outright on combined cost. If your operator is the owner doing infrastructure on Saturday nights, those hours don’t show up on a P&L and n8n self-hosted wins on cash. The honest question is not “which platform is cheaper” but “whose time is the platform actually consuming, and what is that time worth.”

Second, Zapier’s 750-task base tier is below the realistic workload. The pricing page is honest that Professional has variable task tiers; the real comparison is the next tier up, which puts Zapier on top of Make.com for the same nominal volume.

Where each one wins

The TCO numbers are necessary, not sufficient. Three other dimensions determine fit.

Time-to-first-automation. Make.com is the fastest of the three to a working scenario. Its visual builder is the genre-leader, and the ramp from “open the app” to “first running automation” is roughly an hour for an operator who has used Excel macros. n8n Cloud is roughly two hours for the same task because the node-graph metaphor is denser; n8n self-hosted is roughly four hours when you factor in the VPS provisioning and HTTPS termination. Zapier sits between Make.com and n8n Cloud: fast for two-step Zaps, slower than Make.com once the workflow forks.

Agentic-step extensibility. This is where the comparison opens up. n8n ships native AI nodes for Claude, GPT, and a generic LangChain node, and the self-hosted version lets you add MCP servers without vendor approval. Make.com supports agentic steps through its AI Agents (beta) and the Anthropic/OpenAI modules; the gating is volume-based, not feature-based, but you do not control the runtime. Zapier ships Zapier Agents and Zapier MCP across all paid tiers, but the agent runtime is Zapier-hosted with no model substitution and no local tool registration.

If your roadmap includes “we will run our own Claude project as part of an automation in the next six months,” n8n self-hosted is the only option that gets you there without a platform migration.

Data residency and audit. For a UK or EU SMB processing customer data through these workflows, n8n self-hosted on a Frankfurt or Helsinki VPS is the only configuration where the platform vendor cannot read the data in transit. Make.com and Zapier both publish DPAs and sub-processor lists, which covers the GDPR Art. 28 requirement on paper, but the data does pass through their cloud. For most SMB use cases this is acceptable; for the fraction handling regulated data (legal, healthcare, financial), it is not.

The 4-question filter

If you do not want to read the rest of the comparison work, the question we ask before recommending one of these to a four-to-ten person team is:

1. Whose time pays for the platform? If the owner does the infrastructure work as part of their day, n8n self-hosted. If a salaried operator’s hours are billable elsewhere, skip to (2).
2. Are you running more than 5,000 operations a month? If yes, Make.com Pro on the operations tier. If no, Make.com Core at $9/mo will hold the workload.
3. Are any of the integrations you depend on Zapier-exclusive? This is the only condition under which Zapier is the right answer in 2026, and the list of Zapier-exclusive integrations has shrunk year-on-year as Make.com and n8n closed the gap on long-tail SaaS.
4. Will you be running agentic steps you want to control? If yes, n8n self-hosted is the only option that lets you swap models, register tools, and inspect runtime, even if the answer to (1) was “we will hire someone for that.”

What changes this verdict

Cadence on this piece is 30 days, faster than the parent register’s 30-90 day rhythm, because vendor pricing pages move and because the agentic-step feature surfaces are evolving fastest of all. The three things that would flip the recommendation:

• n8n’s licensing terms shift. The Community Edition is on the Sustainable Use License, which restricts commercial hosting-as-a-service but allows internal commercial use. A material change to this license affects the self-hosted recommendation directly.
• Zapier closes the agentic-runtime gap. If Zapier Agents adds model substitution and local-tool registration, the data-residency and extensibility arguments for n8n weaken at the SMB end of the market.
• Make.com publishes a self-hosted tier. Currently no such tier exists. If it ships, the cleanest “I do not want to manage Docker, I do want data residency” path becomes Make.com self-hosted, and n8n’s structural advantage narrows.

We will re-test against vendor pricing pages and feature surfaces on or before 26 May 2026. If any of the three preceding conditions has triggered, this claim moves to Partial.