n8n automation: what it is and where it fits

n8n automation is building workflows in n8n, an open-source tool where nodes pass data between apps, APIs and AI models. A trigger starts the run, each node does one job, and you can add custom code when a node falls short. It suits people who want control and predictable cost on multi-step work. It is a poor fit if you want zero setup.

How an n8n workflow runs

An n8n workflow is a trigger followed by connected nodes. The trigger is a webhook, a schedule or an event from another app. Each node then fetches, transforms, filters or sends data. UI Bakery's guide to n8n describes it as "a flexible, open-source automation platform that lets you build backend workflows by connecting APIs, services, and data into one system."

The useful part for a marketer is that you can add conditions, loops and branches, and drop in JavaScript when a visual node cannot do the job. That makes it closer to a small backend than a trigger-action toy.

Where n8n pricing differs from Zapier and Make

n8n bills by workflow execution, not by step. The n8n blog's comparison of AI workflow tools says its cloud plan starts at $20 a month, billed annually, for 2,500 workflow executions with unlimited steps, while others charge variable credits per step. The same page says Make charges credits for each step, including error handling, and that AI steps can consume more than one credit.

Plans and prices change, so check the current pricing page before you commit. The decision rule I use: if your workflow has more than a handful of steps, or loops over a list of records, a per-execution model usually costs less. If it is two steps and runs rarely, price barely matters and ease of setup wins.

Self-hosting is free for the software, but you pay in server time and upkeep. UI Bakery's guide says a small VPS typically costs $5 to $20 a month plus engineering time. The engineering time is the real cost. If nobody on your team will patch and back up a server, use the cloud version.

For a head-to-head on a solo marketer's budget and time, see n8n vs Zapier for a solo marketer.

What n8n AI agents add to a workflow

An AI agent in n8n is a node that uses a language model to classify input, pick a path or draft output, instead of following a hardcoded rule. Typical uses are sorting support tickets, routing leads and drafting replies. The n8n workflow template library shows the pattern in practice: its HR category lists 335 templates, many of them CV screening and scheduling flows built on OpenAI, Gemini or Claude plus Gmail and Google Sheets.

My position: use the model for judgment calls and plain nodes for everything deterministic. The n8n blog makes the same point about review, saying you can implement a human-in-the-loop review of outputs before taking final actions. I agree, and it matches how I build: a real input, an output someone can inspect, and a person approving before anything goes out. For more on this, read what n8n AI does and how to use it.

Who should not use n8n

n8n is the wrong pick when the learning curve costs more than the automation saves. The n8n blog itself admits it is not the easiest tool on its list to start with. Zapier suits non-technical teams running simple automations, and it has far more built-in integrations.

Skip n8n if:

• the task has one or two steps and runs a few times a month
• nobody can maintain a server and you do not want cloud billing
• the process is not yet clear enough to write down

The last one matters most. Automating a muddled process gives you a faster muddle. If you are unsure which task to automate first, one question to rank automation work is a good place to start.

A scoping checklist for your first n8n workflow

Scope the first workflow on paper before you open the editor. Copy this and fill it in.

WORKFLOW SCOPE
Name:
Trigger (webhook / schedule / app event):
Input example (paste one real record):
Steps in plain words (1 line each):
Which steps need AI judgment? (default: none)
Output (where it lands, in what format):
Human review before it ships? (yes/no, who):
Failure path (what happens on an error, who is told):
Volume per month (executions):
Success check (what you will inspect after week one):
Kill rule (when you turn it off):

Run it on ten real records by hand before you trust it. Read every output yourself for the first week. Then decide whether the review step stays.

If you want the same plain-file approach for other work, the jakebauman.io shop has workflow kits in three groups, and the free starters are the place to begin. Paid kits are one payment with no renewal.

Next step: pick one repetitive task you did this week, fill in the scope block above, and see whether it fits on one page. If it does, build it.

How this post was made: drafted with AI assistance (Claude Sonnet 5.5) from the sources linked above, then checked by Claude Opus 5.5 and an automated quality gate that matched every figure and quotation to its source on 2026-11-01.