What Is n8n and Should Your Team Use It

The short answer to what is n8n: a workflow automation tool built as a node canvas. You drag boxes that represent services, wire them together, and the data flows from one to the next. If you have used Zapier or Make, the shape is familiar. What makes n8n different is that you can run it on your own server and drop actual code into the middle of a workflow when no connector fits.

That combination is why it has become the default for teams building AI workflows, and why it is the wrong choice for teams that just want a form to hit a spreadsheet.

What is n8n, concretely#

A workflow starts with a trigger: a webhook, a schedule, a new row, an incoming email. Then it passes through nodes, each doing one thing: fetch a record, transform a field, call a model, branch on a condition, write to a database.

Three features distinguish it from the simpler tools:

Branching and loops. Real processes have exceptions. n8n models them properly rather than forcing a linear path.

A code node. JavaScript or Python inside the workflow. Every automation eventually hits something no connector covers, and twelve lines of code beats a five-node workaround.

AI and agent nodes. Native LLM steps and agent patterns, so a workflow can decide what to do next instead of following a fixed route. How to build an AI agent covers that design.

Self-hosting, and what it really costs#

The community edition runs on your own infrastructure under a fair-code licence that permits internal business use. This is the feature people move for, and it solves a genuine problem: data that cannot legally or contractually leave your building.

It is not free, though. Be honest about the trade:

A server. Small, but real, plus a database and backups.

Updates. New versions, security patches, occasional breaking changes.

Monitoring. If the instance is down, every workflow is down, and you find out when someone complains unless you set up alerts.

A person. Two to four hours a month, indefinitely, and a name against it.

For a team running thousands of executions a month or handling sensitive data, that trade is clearly worth it. For a team running two hundred, the cloud plan is cheaper once you price your own hours honestly.

How the pricing shape differs#

n8n cloud charges per workflow execution. One run is one unit whether the workflow has three nodes or thirty.

That matters because the competing model charges per step. Zapier counts tasks, Make counts operations, and both mean your bill grows as your workflows get more capable. In n8n, deepening a workflow is effectively free, which changes what you are willing to build.

Zapier vs n8n and Make vs n8n cover the full comparison, including where each of the others is the better answer.

Who should use it, and who should not#

Use n8n if: someone technical owns automation, your data has residency constraints, you are building AI workflows with branching, or your execution volume makes per-step pricing painful.

Do not use n8n if: a non-technical marketer is the only person who will ever touch it and your workflows are two steps long. Zapier will serve you better and the week you save is worth more than the ceiling you gave up.

The deciding question is always who maintains it in six months. AI workflow automation covers the ownership practices that keep any of these platforms alive.

The first workflow to build this week#

Do not start with the ambitious one. Start with something that runs often and fails visibly.

A good first build: an incoming enquiry is enriched with public information about the company, scored against a written rubric, written to your CRM, and posted to a channel with a drafted reply waiting for approval.

It touches everything you need to learn, including the parts people skip:

1. A webhook trigger.
2. An HTTP request to an external service.
3. A model call for the judgement step.
4. A branch for the case where enrichment fails.
5. A write to your CRM.
6. An error path with an alert.

Build steps 4 and 6 on day one, not after the first silent failure. AI lead generation covers the rubric this workflow applies, and n8n workflows covers the ones worth building after it.

Mistakes people make in their first month#

Skipping error handling. A workflow with no error path fails silently, and by then the manual habit is gone.

Building five at once. Build one, run it for thirty days, then build the next.

Storing credentials in nodes instead of the credential store. Fine until you export a workflow to share it.

No version control. Export your workflows and keep them somewhere. An instance that dies with no exports takes the work with it.

Automating an undocumented process. You will encode whatever the last person improvised. Write the numbered steps first.

The short version#

n8n is a node canvas you can self-host, priced per run rather than per step, with real code and native AI nodes. Choose it when someone technical owns automation or when data cannot leave your infrastructure. Build one workflow with its error paths, run it for a month, then build the second.

If you want that first system built on your accounts with your team trained to extend it, that is what my AI training packages cover, and the AI automation agency page is where I do this work for clients.