What is n8n and what can it do?

A workflow in n8n is a chain of nodes on a canvas. The first node is a trigger: a webhook, a schedule, a new email, a new row in a database or an event in a connected app. Each node after it does one job: fetch a record, transform data, check a condition, call an API, run an AI model, send a message. Data passes from node to node as JSON, and you can see exactly what each node received and produced on every run.

Three features matter most for business use:

• The Code node. When no pre-built node does what you need, you write a few lines of JavaScript or Python inside the workflow. This is the escape hatch that stops you hitting a wall.
• Error workflows. You can point any workflow at a separate error workflow that fires when something fails, so a broken step becomes a Slack message to a named person rather than silence.
• AI nodes. n8n has built-in nodes for calling language models, including an AI Agent node that can use tools you give it, so an AI step can sit inside an otherwise rule-based flow.

n8n is distributed under its own fair-code licence rather than a standard open source licence. In practice that means you can self-host it for your own business at no licence cost, while paid cloud and enterprise plans add hosting, support and team features.

Example 1: invoice chasing

• Trigger: daily schedule at 9am.
• Steps: pull unpaid invoices from the accounting tool → filter to those past due → check whether a reminder went out in the last seven days → send the next reminder in the sequence → log it against the client record → if the invoice is past the final reminder, post it to the finance channel for a phone call.
• Output: nobody has to remember who owes what. The finance lead only sees the accounts that need a human.

Example 2: enquiry triage with an AI step

• Trigger: new email in the sales inbox.
• Steps: AI node classifies the email (new enquiry, existing client, supplier, spam) and extracts the fields into a fixed format → a check node confirms the output is valid → enquiries create a CRM deal and get a drafted reply for approval → existing clients route to their account manager → anything the model is unsure about goes to a review queue.
• Output: a sorted inbox and a draft waiting, instead of a pile.

Example 3: a content pipeline

• Trigger: schedule.
• Steps: pull from several news feeds → merge and remove duplicates → rank → write a draft against a style guide → generate an image → publish through the CMS API.
• Output: a published article, on a schedule. This is the shape of a real editorial content engine built in n8n for a digital publishing business.

Choose n8n when at least one of these is true:

• The workflow has branching logic, loops or several systems in the middle.
• It includes AI steps that need checking, retries and fallbacks.
• Volume is high enough that per-task pricing on other platforms would add up.
• The data is sensitive enough that you want it on your own server.
• You want to own the workflow logic rather than rent it.

If the job is "when a form is submitted, add a row to a sheet and send a Slack message", Zapier will do it faster with less setup. n8n earns its place when the process is something your business depends on.

Self-hosting is free in licence terms, not in effort. Someone has to look after the server, keep n8n updated, back up the database that stores your workflows and credentials, and watch disk space as execution logs build up. For heavier loads, n8n's queue mode splits work across separate worker processes so one long job does not block everything else.

The cloud version removes the server work and bills by workflow executions rather than by individual steps. Either way, the part that takes real effort is not hosting. It is designing workflows that behave when an API is down, a field is missing or a login expires.

• Expired credentials. Connected accounts lose access when a password changes or a token is revoked. Prevention: an error workflow that names the failing credential in the alert.
• Silent partial failures. A node returns an empty list and the workflow "succeeds" having done nothing. Prevention: explicit checks that expected data is present, with an alert when it is not.
• Runaway executions. A loop with no exit condition or a webhook that fires repeatedly. Prevention: execution timeouts and deduplication on incoming events.
• Version drift. An n8n update changes how a node behaves. Prevention: test updates on a staging copy before production, and pin versions.

Say you run a 20-agent property brokerage in Dubai. Leads come from portals, your website, Instagram and WhatsApp. Today an admin copies each one into the CRM, guesses which agent should get it, and sends a WhatsApp introduction when she gets a moment.

In n8n, each lead source feeds a webhook. The workflow normalises the data, checks for duplicates against the CRM, uses an AI step to read free-text messages for budget, area and buy-or-rent, and assigns the lead by area and agent availability. It sends an approved WhatsApp template within minutes, logs everything, and alerts the team leader if the assigned agent has not responded within the hour.

The admin moves from copying leads to checking the review queue and keeping listings up to date. The brokerage hosts n8n on its own server because client contact data stays in-house.

n8n is our default for workflows a business depends on, alongside Zapier for simple connections and Claude Code when logic has to be written from scratch. We map the process first, then build in n8n against your real data so you see it working early.

Every n8n build we hand over includes an error workflow wired to a named owner, input checks at the start of each run, deduplication on incoming events, and documentation that explains each workflow in plain English. If you self-host, we set up the server, backups and update routine, or we run it on n8n's cloud if you would rather not hold a server. The workflows are yours either way.