n8n for marketers: a hands-on review of the automation and AI agent builder

Formally, n8n is an open-source workflow automation platform (fair-code license, more on that below). In practice, it is an environment where a marketer builds integrations between tools that don't see each other out of the box – a Marquiz quiz and amoCRM, an ad account and Google Sheets, a CRM and a Telegram bot – without needing a developer.

• Over 400 native integrations – CRMs, messengers, spreadsheets, email, databases, payment systems. If there is no ready-made node, the universal HTTP node connects any service with an open API, including regional tools that never make it into Western catalogs.
• Triggers of any kind – a webhook (form submission), a schedule (report every morning), an event in a service (new spreadsheet row, new email).
• Logic inside the workflow – conditions, branching, loops, error handling, data transformation. This is what separates n8n from simple if-this-then-that connectors.
• Two ways to run it – n8n cloud (sign up and go) or self-hosted on your own server (free, data never leaves your infrastructure).

An important detail for teams on a tight budget: the self-hosted version is not stripped down in terms of workflow features – the license restrictions concern reselling, not your internal use.

All three tools solve the same problem – connecting services into chains. The difference is the billing model, and for complex scenarios it is decisive. Zapier and Make bill per operation: every workflow step burns a "task" or an "operation". n8n counts full workflow executions: one run of a chain is one execution, no matter how many steps it contains.

The takeaway from the table is simple. If all you have is a couple of primitive form-to-spreadsheet connections, Zapier is faster and easier – n8n is overkill there. As soon as you get chains with branching, data enrichment and AI steps, the economics flip: in n8n workflow complexity does not raise your bill, while in Zapier and Make every added step multiplies operation spend.

n8n has been developing its AI side since 2023, and by 2026 it is a full agent builder: the platform ships with over 70 AI nodes based on the LangChain framework – models, memory, vector databases, tools, output parsers.

• The AI Agent node is the centerpiece. You connect an LLM to it (OpenAI, Anthropic, Google, or a local model via Ollama), give it a set of tools and a system prompt – and the agent decides on its own which tools to call to reach the goal.
• Agent tools are any n8n nodes: a CRM lookup, a spreadsheet query, an HTTP request, sending a message. In other words, the agent gets access to all 400+ integrations on the platform.
• Memory – the agent keeps conversation context: chat history can be stored in a database, so you can pick up a conversation with a client a day later without losing the thread.
• Vector databases and RAG – you load a knowledge base (property descriptions, internal policies, price lists), and the agent answers from it instead of making things up.

In practice this means a chain like "chatbot qualifies a lead against a knowledge base and logs the result in the CRM" is assembled in one window from ready-made blocks. I covered the roles such agents fill in marketing in the AI agents article linked above; n8n is the most accessible way for a marketer to build their first agent by hand.

Below are the setups marketing teams build most often, including in real estate. These are architecture templates, not promises of results: every project's numbers will be its own.

1. Lead processing: forms → CRM → notifications

A webhook receives a lead from any source (website, quiz, social lead form), the workflow normalizes the data, checks for duplicates by phone number, creates a CRM deal tagged with the source, and sends the manager a lead card in Telegram. An optional AI step: while the manager is busy, the agent's first message asks the client about budget and timeline.

2. Ad account reports in Telegram

On a schedule, the workflow pulls stats from ad accounts and the CRM, matches spend against leads by campaign, calculates cost per lead, and posts a morning summary to the team chat. An AI step is useful here as an analyst: the LLM comments on anomalies – where spend grew but leads didn't.

3. Content pipeline

From a row in a content plan spreadsheet, the workflow generates a post draft via an LLM following your tone guide, attaches an image, sends it to the editor for approval in Telegram, and publishes across channels after a button press. The human stays the editor; the routine of tending the publishing queue moves to automation.

4. Database enrichment and reactivation

The workflow goes through dormant deals in the CRM, segments them by reason for refusal, and launches targeted touches: a selection of new properties matching the old request, an email about a price change. Without automation, teams almost never get around to tasks like these.

A marketer can build the first such pipeline alone. When there are many of them and they span the whole funnel – from ads to closed deals – it becomes an engineering job: the Matveo team builds such systems end to end, and n8n is one of the working tools there alongside custom development.

n8n cloud pricing at the time of publication (billed annually; monthly billing is roughly 20% more expensive):

Active workflows are unlimited on all plans – you pay only for executions. There is a free trial, no card required.

The self-hosted option is free: you install n8n on your own server (usually via Docker; a VPS costing a few dollars a month is enough) and get unlimited executions. The license is called the Sustainable Use License – it is fair-code, not classic open source. For a user the rule is simple: using n8n for your own business's internal tasks and building workflows for clients as a consultant is fine; what's forbidden is reselling n8n itself – for example, hosting it and charging users for access.

n8n is harder than Zapier, and pretending it takes five minutes would be dishonest. What it asks of a marketer:

• Process logic. Being able to break a task into trigger → conditions → actions and think through the error branch: what happens if the CRM doesn't respond or the lead has no phone number.
• A basic grasp of APIs and JSON. Data travels between nodes as JSON, and you will have to look at it. You don't need to program, but you do need to read a data structure.
• Patience for debugging. The first workflow comes together in an evening, but it only becomes stable after a week of real leads and fixing edge cases.
• For self-hosted – a server. Docker, a domain, HTTPS, backups. Either your own hands or a technical specialist's help at the start.

Who should skip n8n: if your entire automation need is one form-to-email connection, the built-in integrations of your website builder will do. And the reverse: if the company's processes are undocumented and leads get lost at the level of internal rules, no tool will fix that – put the system itself in order first. I wrote about this in the article on bringing AI into a business: the sequence process → data → automation applies to n8n too.

1. 01Pick one process with a measurable outcome – most often lead processing or the morning report.
2. 02Build the workflow in the cloud on a trial and run it on test data.
3. 03Run it in parallel with the manual process for a week or two and compare speed and losses.
4. 04Only then expand: add AI steps, a second chain, a move to self-hosted as volume grows.