What is n8n? A plain-English guide for business owners, not developers

Search for what n8n is and you get two kinds of answer: a vendor page saying “workflow automation for technical teams”, and a forum thread that assumes you already know what a node is. Neither tells you the thing you actually want to know, which is whether this tool will solve your problem or eat your next three weekends.

This is the version we would give a client before quoting anything: what n8n is, what it genuinely does better than the alternatives, where it will frustrate you, what “free” really means, and five workflows a normal small business runs on it. It ends with the section most guides skip — the point at which you should stop building it yourself.

Last updated: 7 September 2026.

What n8n is, in one paragraph

n8n is a workflow automation platform. You open a canvas, drag on boxes called nodes, join them with lines, and each node does one thing: watch a mailbox, call an API, run a condition, write a row, ask a language model a question. When a trigger fires — a form submission, an incoming email, a webhook, a schedule — n8n walks the graph from left to right, passing data from one node to the next. That is the whole mental model, and it is the same one Zapier and Make use. What differs is the depth you can go to before the tool runs out of room.

n8n meaning: where the name comes from

The name is a numeronym for “nodemation” — node plus automation — abbreviated the way “internationalization” becomes i18n: keep the first and last letter, count the eight in between. It is pronounced n-eight-n, and the company writes it lowercase in every position, including the start of a sentence. That is not a typo you are seeing across their docs; it is the spelling.

It is also worth knowing that n8n is source-available rather than open source in the strict sense. The code is public and you can read, modify, and self-host it, but it is published under the Sustainable Use License rather than a permissive one. That distinction has practical consequences, covered further down, and it is the single most commonly misunderstood thing about the product.

What n8n is genuinely good at

Four things, and it is worth being specific because each one maps to a type of buyer.

Long workflows without a cost penalty. n8n bills per workflow execution. One run is one billable unit whether the workflow contains four nodes or forty. Make and Zapier both bill per step — per operation and per task respectively — so on those platforms every validation check, every log line, every retry costs money. That changes what you build. On per-step billing you quietly delete the error handling that makes a workflow debuggable, because each check has a price. On n8n those steps are free, so the workflow you end up with is the one you actually wanted. The arithmetic is stark: a 12-step workflow running 3,000 times a month is 3,000 billable units on n8n and 36,000 on Zapier.

An escape hatch that actually works. Every visual builder eventually meets logic it cannot express in dropdowns. The question is what happens next. On n8n you drop in a code node, write a few lines of JavaScript (Python is available too), and carry on inside the same workflow. On the alternatives you are pushed out to an external function service, a webhook to something you host yourself, or a workaround built out of three nodes pretending to be one.

Self-hosting as a first-class option. n8n runs on your own infrastructure — the standard route is a Docker container on a small server — which means the data in your workflows never leaves machines you control. For anyone handling personal data, medical records, or client financials under a contract that specifies where data lives, this is not a preference, it is the requirement that eliminates most of the market.

AI steps that behave like normal steps. A model call is just another node, so classification, extraction, summarizing, and drafting slot into a workflow next to the CRM write and the Slack message. n8n goes furthest of the three mainstream platforms on agent-style workflows where a model is given tools and allowed to choose between them. If your automation involves reading unstructured text and deciding something, this matters.

What n8n is not good at

This is the section vendor pages do not write, and it is the one that saves people money.

It is not a no-code tool, whatever the category page says. You can build a first workflow without code. You cannot run a production workflow without meeting expressions, JSON, API authentication, and the difference between an item and an array. The people who abandon n8n are almost never defeated by the canvas; they are defeated by a webhook returning a nested object in a shape the next node refuses to accept.

It is not good at being handed over. A workflow graph with twenty nodes, three branches, and a code step is readable by the person who built it and opaque to everyone else. If your plan is “our office manager will maintain this”, n8n is the wrong choice and Zapier is the right one — a linear list of steps in plain language is a genuinely better artifact for a non-technical owner, even though it is a worse tool.

It is not good at very small jobs. If the automation is one trigger and one action, use the free tier of whatever is simplest and stop reading. And if it is a single scheduled job between two systems you already control, a small script on a cron schedule is cheaper and less fragile — you are otherwise paying a subscription for connectors and a UI you do not need.

Self-hosting is not good at looking after itself. Free means no licence fee, not no cost. The upgrades, the backups, the certificate that expires, the container that stops overnight and takes the workflows with it silently — all of that is now yours. Nothing about n8n is unusually fragile here; this is simply what running your own software costs.

It is not the place to learn what to automate. No platform tells you which process is worth automating, and choosing the wrong process is a far more expensive mistake than choosing the wrong tool.

Self-hosted vs n8n Cloud

There are two ways to run it, and the choice is usually simpler than people make it.

The default answer is n8n Cloud. Three reasons override it: data residency, private-network access, or execution volume high enough that server costs beat the subscription.

Run the arithmetic before choosing self-hosting on cost grounds, because it rarely wins at small-business scale. The bare infrastructure floor is low — a small server sufficient for modest volume is roughly $10–25 a month — but infrastructure is not what self-hosting costs. Once the operational work is priced honestly (version upgrades, backup restore tests, patching, incident response, monitoring), a small production setup runs about $310 a month all in against a Cloud subscription starting in the low tens of euros. That figure and the arithmetic behind it are in the real cost of n8n, which costs the same stack three ways. The short version: self-hosting is cheaper on the licence line and more expensive in total until your Cloud tier would exceed roughly $310 a month, or the maintenance genuinely disappears into work an ops team already does.

The licence caveat almost everyone gets wrong

n8n’s Community edition is published under the Sustainable Use License. In plain terms:

• Permitted: using it inside your own business, for free, including a for-profit company, at any scale, on your own servers. Modifying it for internal use. There is no fee and no vendor-imposed execution cap.
• Restricted: turning n8n itself into the thing you sell. Hosting an instance for clients, reselling access, white-labelling it, or embedding it inside a product you charge for.

The line is between using it to run your operations and distributing it as a service. Automating your own invoice intake on a self-hosted instance is squarely fine. Running a client’s workflows on your instance and billing them for the automation is exactly the case the licence is written to catch — n8n sells a separate embed arrangement for that, and the right move is to ask them rather than to assume.

Two practical consequences. First, “n8n is open source and free” is true enough for internal use and misleading if you have a service business in mind, so read the licence yourself before building a business model on top of it. Second, when an agency tells you they will host your workflows on their own instance, that is a question worth asking about — including of us. We build on your infrastructure or your Cloud account, and the credentials and workflows end up in your accounts.

Five n8n examples a small business would actually run

Not demos. These are the shapes that recur across real projects, with the part that breaks named honestly in each.

1. Inbound lead capture and routing. A form submission or an email to sales@ triggers the workflow. n8n enriches the record from a lookup service, deduplicates it against the CRM, creates or updates the contact, posts a short summary to a team channel, and sets a timer that nudges the owner if nothing has happened in 48 hours. What breaks: duplicates. Two submissions from the same person with a typo in the email are two contacts unless you decide, explicitly, what counts as the same person.

2. Supplier invoice intake. Invoices land in a dedicated mailbox. n8n pulls the attachment, has a model extract supplier, date, total, tax, and line items, validates the total against the purchase order, writes a draft entry into the accounting system, and pushes anything that fails validation into a human review queue. What breaks: the source documents. Twelve regular suppliers sending structured PDFs is a routine build. Two hundred irregular suppliers sending phone photos is a different project at three times the price.

3. Client onboarding kickoff. A signed contract fires a webhook. n8n creates the project, the shared folder with the right permissions, the draft invoice, the kickoff calendar invite, and the welcome email sequence — then posts one message confirming all five happened. What breaks: partial failure. Step four failing after steps one to three succeeded leaves a half-onboarded client, so this workflow needs to know how to report exactly where it stopped.

4. Support and review triage with approval. Incoming tickets or public reviews are classified by topic and sentiment, matched against a knowledge base, and a draft reply is written and placed in front of a person to approve, edit, or discard. Nothing is sent automatically. What breaks: nothing technical — the approval queue is the point. A workflow that publishes replies on its own is how automation loses a company’s trust permanently, and one bad public response costs more than the workflow saved.

5. Nightly reporting rollup. On a schedule, n8n pulls spend from the ad platforms, pipeline from the CRM, and revenue from the billing system, normalizes the currencies and date ranges, writes a single tidy table, and posts a five-line summary to a channel each morning. What breaks: definitions. Three systems will disagree about what a “conversion” is, and the workflow cannot resolve that for you — a human has to pick one definition before anything is built.

Notice the pattern. In every case the technical work is the easy half, and the expensive half is deciding what happens when reality does not match the happy path.

The realistic learning curve

Here is the honest timeline, from watching people go through it.

• First afternoon. You will have a two-node workflow running. This part is genuinely easy and it is why people get optimistic.
• First week. You meet expressions and data shape. Your API returns a nested object, the next node wants a flat one, and you spend an evening learning the difference between an item and a list. This is the wall, and it is where most people quit.
• Second week. Something useful is running on real data. You are pleased with it. It has no error handling.
• First month. It broke. A credential expired, an API rate-limited you, a record arrived with a missing field. You now understand that the workflow is 30% features and 70% failure handling, and you start again with that in mind.
• Third month. You are competent, you have five or six workflows, and you are the only person in the company who can fix any of them. That last fact is the real cost, and it is not on any pricing page.

The community forum and the shared template library shorten the first two steps considerably — there is a public workflow for most common shapes, and starting from one beats starting from a blank canvas. Nothing shortens the fourth. Exception handling is craft, learned by having things break.

So, is n8n worth it? A test you can run without buying anything

Four questions. If you answer yes to two or more, n8n is a reasonable bet. If you answer yes to none, use something simpler and spend the saved time on the process instead.

1. Do your workflows have more than about six steps, or run more than a few thousand times a month? Multiply steps by monthly runs. If the product is above roughly 10,000, per-step billing will hurt and n8n’s model is worth having. The full comparison of how each platform meters you is in n8n vs Make vs Zapier.
2. Will at least one step need logic a dropdown cannot express? Custom scoring, a bespoke transformation, a call to an API with no ready-made connector. If yes, you want the code node.
3. Is there a rule about where your data can live? A contract, a regulator, or a system that only exists on your internal network. If yes, self-hosted n8n is one of very few options.
4. Is there one person who will own these workflows for the next year, and do they enjoy this kind of problem? If no, this is the answer that outranks the other three. n8n in the hands of nobody is worse than Zapier in the hands of your office manager.

That fourth question is the one that decides most projects, and it is the one nobody asks themselves before installing.

When to delegate instead

There is a specific moment where building it yourself stops being the cheaper option, and it is recognizable. Any two of these mean the maths has already flipped:

• You are three weekends in and the happy path still is not reliable. Your time has a cost even when it is not invoiced, and the third weekend is usually where the total passes the price of having it built.
• The workflow now touches four or more systems. Cost does not scale linearly with system count; it scales closer to quadratically, because every pair of systems is a new place where data can disagree.
• It has to work when you are on holiday. A workflow with exactly one person who understands it is an outage waiting for a calendar clash.
• You are automating something with money or a regulator attached. Payments, payroll, personal data, anything with a retention requirement. The design changes rather than getting decorated afterwards, and retrofitting it usually means rebuilding.
• You have inherited a half-built tangle. Someone left, and there are eleven workflows, four of them disabled, none documented. Untangling that is a specific job and it is faster than restarting from zero — usually.

For reference, so you can compare against your own time: our process audit is $490 (the market charges $2,500–8,000) and is credited against implementation. A workflow build starts at $1,900 where the market is $6,000–15,000, and ongoing monitoring and extension is $690 a month against a market rate of $3,000–8,000. Those numbers are published rather than quoted after a discovery call, and the reasoning for sitting at the bottom of the market is on the n8n agency page.

Two caveats on delegating. Do not hire anyone before you can state the process in writing, step by step, including what happens when a record is missing or malformed — otherwise you are paying someone to discover your own process at their rate. And whoever builds it, insist the workflows, credentials, and documentation live in your accounts. Anything else is a subscription wearing a project’s clothes.

And if n8n turns out to be the wrong platform for what you are doing, that is a perfectly good outcome of asking. We build on all three mainstream platforms and hold no reseller agreement with any of them, so the recommendation follows the process — sometimes it is Make, sometimes Zapier, occasionally a fifteen-line script and no platform at all. That is the first thing the AI automation service engagement settles, before anyone builds anything.