What is n8n? The complete guide for businesses

n8n is an open source workflow automation tool that connects your applications and runs sequences of tasks without human intervention. You pick a trigger, you chain nodes together, and the sequence runs every time the condition occurs. The name is pronounced « n-eight-n » and comes from « nodemation », a contraction of node and automation. Since 2023 n8n has also shipped native artificial intelligence capabilities, which lets it orchestrate a classic automation and an AI agent on the same canvas.

Tandem is an n8n partner agency and deploys the tool with its clients, across more than 40 SMEs and mid-caps, from Doctolib to InfraVia and on to PayFit, Nomination and KparK. This guide covers what our teams explain in a first meeting, namely the concepts you need, what n8n can do, what it really costs, and the limits that product demos leave out.

What is n8n exactly?

n8n is a fair-code licensed workflow automation platform, built by the German company of the same name since 2019. Its official documentation defines it as a tool that combines business process automation with AI capabilities. In practice n8n does three things. It connects applications that do not talk to each other natively, it moves and transforms data between them, and it triggers actions based on rules you define.

Two characteristics set n8n apart from the rest of the market. The first is self-hosting. You can install n8n on your own server, in France if you want to, and your data never leaves your infrastructure. The second is the billing model. n8n counts workflow executions rather than operations, so a thirty-node scenario costs exactly the same as a three-node one.

The official n8n catalogue lists more than 2,000 integrations as of 5 August 2026, and the project's GitHub repository passed 199,000 stars on the same date. On the market side, German software group SAP took a stake in n8n on 12 May 2026, taking its valuation to roughly 4.8 billion euros, with the stated intention of embedding the n8n canvas in its own agent builder Joule Studio (sources Bloomberg and Tech.eu, May 2026). For a European SME that news has one practical merit. It lowers the risk of betting on a tool that might vanish within eighteen months.

How does an n8n workflow work?

An n8n workflow always follows the same logic, a trigger and then a chain of nodes leading to a result. The trigger is the event that starts the sequence, for instance an incoming email, a submitted form, a new row in a spreadsheet, or simply a clock that fires every Monday at 8am. Each node that follows does exactly one thing, read a piece of data, call a model, write to a CRM, send a message.

Four notions are enough to read any existing workflow and modify one without breaking the rest.

• The node. This is the unit building block of a workflow, and it comes in four families. The trigger node starts the sequence, the action node does something in a service, the utility node transforms the data along the way, and the code node runs JavaScript when no integration fits.
• The trigger. It decides when the sequence starts. n8n handles webhooks, time-based schedules and application event listeners, which covers almost every need an SME has.
• The credentials. You store the access details for each application once, and all your workflows reuse them. This is also the part that deserves the most attention on the security side.
• The execution. Every run of the workflow is logged, with the input and output data of each node. You can see exactly where a scenario failed and replay it.

Do you need to code to use n8n?

No to get started, yes to go far. n8n sits in the low-code category rather than in strict no-code, and its visual editor lets you build a first useful workflow by dragging nodes and filling in fields, without writing a line of code. Most of the automations Tandem installs in SMEs rely on standard nodes and simple expressions.

The nuance arrives at the second tier. As soon as you handle nested data, cover edge cases or want to make a scenario reliable, a little JavaScript changes everything. n8n offers a Code node that accepts JavaScript and Python, and it is often the thing that unlocks the last five percent of a project. The learning curve is real, and it is the most frequent criticism of the tool. A business owner hoping to automate their back office alone over a weekend will be disappointed.

What is n8n used for? Five patterns installed in SMEs

The workflows that create the most value are never the most spectacular. Here are the five families Tandem finds engagement after engagement, with the gains teams actually report.

1. Triaging inbound requests. A model reads each email or ticket, classifies it by category and urgency, then drafts a reply the team validates. This pattern shows up in almost every engagement, because the volume is daily and the rule is easy to write.
2. Lead enrichment and scoring. A new contact triggers a public information lookup, a score, the creation of the CRM record and a personalised outreach sequence. Sales teams get back the time they spent copying and pasting.
3. Accounting and administrative back office. Supplier invoice processing, overdue payment reminders, reconciliations and accounting exports all fire on their own, with no manual re-entry.
4. Content and reporting production. A workflow collects, filters, drafts and then assembles a recurring deliverable, with a human review before anything goes out.
5. Multi-tool data consolidation. n8n pulls figures from the CRM, the billing tool and the ad platforms, aggregates them into a dashboard and raises an alert when a metric leaves its expected range.

One detailed example beats a list. On a sector news newsletter engagement, the n8n workflow chains a web research tool that surfaces relevant articles, a model that filters and drafts each section in an imposed format, then an email tool that stages the send with a human review before distribution. We documented that scenario step by step in our article on automating a newsletter with n8n. For a wider view, our inventory of AI use cases in business lists what works function by function.

n8n Cloud or self-hosted, which should you pick?

n8n comes in two deployment modes, and the choice hinges on how sensitive your data is rather than on price. The self-hosted community edition is free, with no execution limit and no restriction on available nodes, under a fair-code licence for internal use. You only pay for your server, and you take on the updates, the backups and the monitoring.

On the cloud side, the prices listed on the official n8n pricing page on 5 August 2026 start at 20 euros per month on annual billing for the Starter plan, with 2,500 monthly executions, then 50 euros per month for the Pro plan and its 10,000 executions. Both include unlimited users and workflows. The Business plan, at 667 euros per month, adds single sign-on and a self-hosted option under a commercial licence.

An SME that moves quotes, contracts and client data through its workflows is better off self-hosting in Europe, because compliance with the General Data Protection Regulation becomes far easier to demonstrate when nothing leaves your infrastructure. A team with no systems skills will gain more by starting on the cloud, even if it migrates later. The cost of a server is not paid in euros alone, it is also paid in maintenance hours.

Tandem hosts and maintains client workflows in both configurations. The details of that service are on our n8n agency page.

n8n and AI agents, the real shift

This is the capability that made n8n take off. The tool natively integrates language models, vector stores for retrieval augmented generation (RAG) and the MCP protocol, so the same canvas orchestrates your classic automations and your AI agents. Selecting an AI Agent node gives you a pre-wired agent skeleton, with its trigger, its memory and its tool list. All that remains is writing the instructions and connecting the applications.

The MCP protocol pushes that logic one step further. Community MCP nodes let an n8n agent query a tool server, discover the available actions on its own, then fill in the parameters with no extra prompt and no code. Two reservations still keep the mechanism away from sensitive use. MCP servers are often hosted locally, and the protocol sets no standard for authentication or fine-grained permissions.

The ease of starting an n8n agent creates a trap. An autonomous agent impresses in a demo and disappoints in production, because it picks its own path and is inherently less predictable than a workflow. Companies want control, measurability and repeatability first.

Across more than 40 companies supported, Tandem sees a four-level progression, the assistant you call on demand, the simple automation with no intelligence, the workflow that inserts a model in the middle of a deterministic sequence, then the autonomous agent. The vast majority of projects that return on their investment stop at level 3. We set out those trade-offs in our comparison of workflows, assistants and AI agents and in our analysis of the difference between an AI agent and an AI workflow.

An agent sells the dream. A workflow runs the business.

The limits of n8n that product demos leave out

n8n is a tool, not a magic solution. Four limits deserve to be stated before you commit.

1. The learning curve is steep. Expressions, nested data handling and wiring the AI blocks take hours of practice. Plenty of business owners give up after twenty-odd hours without a single workflow in production.
2. Self-hosting has a hidden cost. The licence is free, but someone has to run the server, the version upgrades, the backups and the monitoring. Without that someone, you find out about the outage at the same time as your customer.
3. Model costs sit on top of the listed price. A workflow that calls a language model on every run consumes tokens billed by the model vendor, whatever your n8n plan. At high volumes that line often exceeds the subscription itself.
4. Governance becomes critical as you grow. Thirty workflows built by five different people, with no naming convention and no review, produce exactly the tangle you were trying to avoid. Who builds, who approves, who monitors, those three questions get settled before the tenth workflow.

n8n, Make or Zapier?

The difference comes down to three points. Zapier is the easiest to pick up and the most expensive at scale, because it bills per operation. Make offers a very readable visual editor and a good power-to-price ratio, with no self-hosting option. n8n asks more effort upfront and returns self-hosting, a predictable per-execution cost and a distinctly more advanced AI layer. For a European SME that handles client data and wants to industrialise, that last profile wins most often. The detail of the pricing grids, the billed units and the hosting constraints is covered in our n8n vs Make vs Zapier comparison.

Will Claude Code and dev agents replace n8n?

No, and Tandem sees two complementary tools instead. Adoption of development agents such as Claude Code, Codex or Antigravity has accelerated since late 2025, sometimes faster inside companies than an internal AI assistant. A dev agent reproduces in two days a scenario that used to take two weeks, data aggregation, complex business logic and integrations included. The result is real.

The trap is mistaking a scenario that runs locally for a scenario in production. When an automation breaks at 3am, you do not want a terminal. You want visual logs, automatic retries, alerts and reliable execution. That is exactly what the platform provides, and it is what makes n8n hard to replace.

• What the platform provides. Deployment without multiplying micro-services, monitoring and error handling, credentials that stay on the platform side, and building blocks reusable from one workflow to the next.
• What a dev agent implies. Deploying and maintaining code, handling observability by hand, and industrialising the infrastructure around the flow.

The combination that works today splits the roles. The dev agent generates the logic and the custom nodes, then builds the internal interface that makes the scenario usable by everyone. n8n runs it, monitors it and holds the credentials.

The value is not in the workflow that runs. It is in the tool the teams actually use.

Where to start with n8n

Four steps take you from curiosity to a first useful workflow, and the order matters more than the speed. Keep the cheat sheet below to hand, it gathers the vocabulary, the key nodes and the structure of an agent on a single page.

1. Pick a tedious, frequent task. A task that takes ten minutes a week does not deserve a workflow. Aim for what comes back every day and nobody enjoys doing.
2. Write the sequence on paper before opening the tool. The trigger, the steps, the edge cases and the expected result. A poorly framed scenario produces a workflow that falls over at the first surprise.
3. Start on the cloud with the free trial. Fourteen days are enough to validate your first scenario without standing up a server. No need to start from a blank page either, the n8n community library held 11,129 workflow templates on 5 August 2026. The hosting question gets settled afterwards, once the real need is known.
4. Ship one workflow to production, completely. With its error handling, its alert and its documentation. One reliable workflow beats ten abandoned prototypes.

Start small, then widen. That holds for agents as much as for classic workflows, and getting familiar with the tool matters as much as the outcome of the first scenario.

Should you adopt n8n in your company?

Yes, if you have repetitive processes, applications that do not talk to each other, and at least one person willing to spend time on it. n8n is today the most complete automation tool for a European SME or mid-cap, because it combines self-hosting, a predictable cost and an AI integration its competitors are only starting to match.

No, if you are looking for a tool that runs with nobody looking after it. In that case the question is not which tool but how the work is framed. Our AI audit exists for exactly that, mapping processes and quantifying the gain before a single workflow gets built.

The licence is free. Your team's time is not.