n8n is a €20 tool that quietly hires you a junior engineer
n8n runs on a €15–25/month VPS. The cloud plans start around €24/month. So when a founder hears "workflow automation," they think: software license, set it and forget it, ROI in three months. What they actually bought was the ability to hire a junior engineer — one that doesn't show up on the payroll but needs someone to patch workflows when an API changes at 2am.
The confusion is real. The license is cheap. The ownership cost is not. We budget 2–4 hours per month of engineering time per 10 live workflows. That's not "checking in." That's actively maintaining deployed software. Each workflow is a flowchart that runs on a schedule or a trigger, but the nodes themselves are no-code only in the sense that you're not writing Python. The logic between them is software, with all the fragility that entails.
What n8n actually automates, in plain language
n8n moves data between tools without a human copy-pasting. CRM to spreadsheet to invoicing to email. It runs nightly reports, weekly digests, monthly reconciliations on a schedule nobody has to remember. It catches a new form submission, qualifies the lead, and notifies sales in Slack — all before anyone notices. It drafts a reply, summarises a document, classifies a ticket, and then hands it to a human for approval. This last pattern — AI-in-the-loop — is where n8n shines for SMBs.
What it does not do well: real-time chat UIs, heavy data processing, anything a user clicks on directly. Think of Zapier as a taxi: you pay per ride. n8n is a leased car: fixed cost, you drive it. If you're running a few critical workflows that touch 5,000+ records monthly, the per-ride model gets expensive. If you're just triggering an email or syncing a spreadsheet column, Zapier wins on simplicity.
The best mental model: if you can describe the work as a flowchart with no branching, or maybe one or two decisions, n8n will feel natural. If you need to loop over data, make decisions based on external state, or chain five different API calls, you've entered the zone where n8n works but requires someone who understands software.
The honest monthly bill: hosting, AI calls, and the 2am phone call
Self-hosted n8n costs €15–50/month on a VPS you manage. n8n Cloud runs €24–120/month depending on execution volume. Then come the LLM calls. OpenAI or Anthropic typically adds €30–300/month for an SMB running a few AI workflows. But here's the real shock: connectors to paid APIs — Uberall, Apollo, enrichment tools — often dwarf the n8n bill itself. A single data-enrichment API might cost €200–500/month.
Add maintenance. If you outsource it, budget €200–800/month for someone to check logs, patch workflows when APIs change, and handle the 2am alert. If you don't outsource, that's 4 hours of your ops person's week — roughly €500–800/month in salary cost. Most founders don't count that labor, which is why they think n8n is cheap.
The hidden killer: silent failures. A workflow stops firing on Tuesday, nobody notices until Friday, and you've lost three days of data or three days of leads. The cost of that failure — lost bookings, missed opportunities, out-of-sync records — almost always exceeds the entire monthly bill. That's why we now ask every prospective client: do you have error alerting that reaches a human, or just a log file nobody reads?
When n8n alone is enough (and when it genuinely isn't)
n8n alone is enough if you have under 10 workflows, standard SaaS connectors exist (Slack, Stripe, HubSpot, Gmail), and one tech-comfortable person internally owns them. Think internal reporting, lead routing, simple notifications, calendar sync. These workflows are stateless, fire rarely, and don't touch customer data.
n8n genuinely isn't enough for anything customer-facing where a failure costs revenue or trust. It isn't enough for AI agents that make decisions, loop over data, or call several tools in sequence — those need error handling, versioning, and observability that n8n doesn't provide out of the box. It isn't enough for multi-location, multi-language systems, or anything touching more than ~5,000 records at scale. Our process automation service starts where DIY n8n ends: when the cost of failure exceeds the cost of hiring us to build properly.
The telltale sign you've outgrown DIY: your workflow has more than 30 nodes or more than 2 branches. At that point, you're no longer building a workflow — you're building software, and you need a team, not a tool.
What a 14,552-listing sweep looks like built on this kind of stack
Radical Storage had health data on roughly 400 of their listings — about 3% of their network of storefronts. They contacted us thinking they needed a single n8n workflow to fill the gap. What they actually needed was a system. We built a sweep that now monitors all 14,552 storefronts across 239 cities in 48 countries. It surfaced 3,271 listings silently invisible on Google. Bookings were being lost with no error anywhere. Fixes are now ranked by the revenue sitting behind each listing: €1.9M/yr mapped to specific repairs. That's the gap between n8n and production: not a workflow, a monitored system with priorities. You can read the full case study on our /work page for the architecture detail, but the engineering lesson is this: once silent failures become visible and tied to revenue, the cost of owning that system in-house usually exceeds the cost of outsourcing it.
How to tell a good n8n build from a bad one before you pay for it
Ask any vendor — or test yourself — against these seven questions. Does it have error alerts that reach a human, not just a log file? Are credentials stored in n8n's vault, not pasted into nodes where they can leak? Is there a staging instance, or does the vendor edit live workflows? Are workflows version-controlled in Git, or is the only backup "in n8n"? Is there documentation a non-developer can read to understand what each workflow does? Does the vendor quote a maintenance retainer, or disappear after handover? Does anyone own the 2am alert?
If you answer "no" to any of those, you've hired a consultant to build you a time bomb. A good n8n build should feel boring. Boring means it runs without you, fails audibly, and can be handed to your ops team tomorrow.
If you already have an n8n instance collecting dust, send us the three workflows you tried to build and we'll tell you in a week which ones are worth finishing, which ones belong in a different tool, and what the real monthly run-cost looks like. Book that scoping call at /contact.