Conversations about automation usually start in the wrong place. People ask "should we set up n8n or Make?" when it isn't even clear yet which task is going to be automated. Picking the tool is the easiest part of the job. The hard part is finding the repetitive task that eats hours out of your team's day, and building it so that you'll notice when it breaks.
In this post I'll explain what AI automation is, how n8n works and how "free" it really is, how it differs from Zapier and Make, and how WhatsApp automation works. At the end there's a short checklist that shows whether the task you have in mind is ready for automation.
The least discussed part of that definition is the second sentence. In my view, most workflows don't need AI at all. You don't need a language model to add a row to a spreadsheet when a form comes in. If you hand that job to AI, you pay a model fee on every run, and every now and then you get a strange error nobody can explain.
AI earns its keep when you put it where content needs to be understood. Telling whether an incoming email is a complaint or an order, say, or reading the amount off a PDF invoice. A classic rule can't do those things; a model can. Just don't forget it can also get things wrong.
Tasks that happen often and whose rules can be written down. To figure out whether a task is worth automating, you have to look at each one individually; the first week of an AI automation project is spent on exactly this screening.
Entering web form requests into the CRM
Several times a day · clear rule
Sending the shipping tracking number to the customer
Every order · needs a WhatsApp template
Classifying incoming invoices
AI reads · amount goes through human approval
Weekly sales report
Data in two systems · one doesn't connect externally
Annual dealer contract
Once a year · different every time
Let me explain why the contract was ruled out. For a task done once a year, the setup and maintenance effort you'd spend automating it is greater than the time you'd save. You don't need a spreadsheet to work that out.
Now you might say, "our process goes differently every time." In that case, you need to write the process down first. Automating a task that has no rules does nothing except speed up the mess. If you have no idea where to start, and AI use across your team is scattered too, that's more of an AI consulting question.
n8n is an automation tool for connecting apps to each other and building workflows. You build the workflow on screen by linking boxes together, and each box is a step. Its most distinctive difference from similar tools is that you can install it on your own server.
Let me give you a concrete example. A request from someone who fills out the quote form on your website could go through this workflow.
AI runs in only the second step of this workflow. The rest is plain rules. The step people skip most often is the last one. One day, one of the connected apps renews its password or changes its form, and the workflow stops. Without an error notification, you only find out when a customer calls asking "what happened to my quote?"
A broken workflow is a problem. A workflow that doesn't tell anyone it's broken is a much bigger one.
At most businesses, n8n automation examples fall into a few categories. Pushing new orders into a spreadsheet and the accounting system, reading and classifying incoming invoices, a sales report that arrives on its own every Monday morning, a notification sent to the customer once the package ships. None of them is flashy. The ones that work usually look just like these.
You can use the version you install on your own server for your company's internal operations without paying a license fee. n8n's cloud version and some enterprise features, however, are paid.
Let me be frank: the line "n8n is free" is only half true. n8n's code is distributed under a license called the Sustainable Use License. The license allows the software to be used for your own internal business purposes or for non-commercial personal use. Offering n8n to others for a fee isn't covered by that permission, and enterprise features are tied to a separate license. What's free is the license. The server, updates, backups and security are on you, and someone actually has to do those things.
On the cloud side, according to n8n's pricing page, plans are split by the number of workflow executions per month. An execution means the workflow running once from start to finish; it doesn't matter how many steps are inside it. Plan names and limits change from time to time, so I'm not putting numbers here. I recommend checking that page once more before you set things up; I check it on every project too.
If you want to start fast and simple, Zapier makes more sense; if control and keeping your data on your own server matter, n8n does. Make sits somewhere in between. The real difference is in how they charge, and that difference becomes clearer as your volume grows.
| Zapier | Make | n8n | |
|---|---|---|---|
| What it charges for | Tasks: every successfully completed action step | Credits: most action steps cost 1 credit | In the cloud, monthly workflow executions; self-hosted, no license fee but server costs |
| Free starting option | Limited free plan | Limited free plan | Self-hosted community edition, free trial in the cloud |
| Ease of setup | Easiest | Medium | Requires technical knowledge |
| Where data is processed | On Zapier's servers | On Make's servers | On your own server if you want |
| Best for | Low-volume workflows with few steps | Moderately complex workflows | High volume, sensitive data, technical support |
The billing units are listed on each tool's own pricing page: Zapier defines a task as "a unit of work completed successfully" and doesn't count triggers; Make deducts one credit for most operations. So every time a multi-step workflow runs on Zapier or Make, it uses roughly as many units as it has steps, while on n8n cloud it counts as a single execution.
That doesn't mean n8n is cheaper in every case. For a simple workflow that rarely runs, Zapier's ease of use is often worth more than the effort of setting up and maintaining n8n. I can't tell you which one will cost less without seeing your workflow.
It's when a WhatsApp message goes out automatically after an event, like a package shipping or an upcoming appointment. This isn't done with the regular WhatsApp Business app, but with the WhatsApp Business Platform (API) and message templates approved by Meta.
There are two rules you need to know up front. The first is consent. Meta requires businesses to get permission from users before sending them messages. The second is cost. Since July 1, 2025, Meta has been charging per message. Marketing templates are always paid. Utility templates are free if they're sent within the 24-hour window that opens when the customer messages you, and paid outside that window. Free-form replies you send when a customer writes to you are also free.
Messages like shipping notifications usually fall under utility templates, but Meta has the final say on which category a template lands in. Checking the category after approval is the business's responsibility. I won't guess on this; you need to check each template separately.
If what you're looking for is an assistant that answers customer questions, that's an AI chatbot job. I explain how it works and where it can go wrong separately in the chatbot guide.
You start with a single task. First the steps are written down, then the workflow is built and tested without touching the live system, and only after that does it go live. You can use the sequence below for the first workflow you build yourself.
When building your first workflow
Do the task by hand once more, but take notes
Write down every click, every copy-paste. The workflow is built from this list.
Collect the exceptions
What happens if the address is missing? What if the same customer writes twice? It's usually these edge cases that break automations.
Limit the AI step
The model goes only where something needs to be read or classified. Fields like amounts and payments get a human approval step.
Test without touching production
Test with real sample data but without affecting the live system. Send bad data on purpose and see what the workflow does.
Set up error notifications, then go live
Don't switch the workflow on until it's clear who gets notified if a step fails. Check the execution logs every day for the first two weeks.
A single, simple workflow is usually up and running in somewhere between a few days and two weeks. Setups that connect several systems take longer, because every piece of software exposes a different door to the outside world. Some have a ready-made integration, some have none at all.
One more note. The workflow has to write data somewhere, and most of the time that place is a CRM. If your requests are currently sitting in a message inbox or a notebook, some workflows may first require CRM consulting to set up a place where customer records are collected.
Think of one task you're considering automating and check off the list for that task.
Is this task ready for automation?
0/7Check the items with the task you have in mind.
The item people get stuck on most is the fourth one. Some software doesn't expose a proper connection point, and then the task can only be automated with fragile workarounds. Knowing this before setup is better than getting stuck halfway.
If you've decided to build it yourself, the steps above are a good enough start. If several systems need to be connected, or there's no one to take over maintaining the workflow, working with an AI automation team both speeds things up and keeps the workflow from quietly stopping.
The three most common are n8n, Make and Zapier. The choice depends on how many times a month the workflow will run and where you want your data processed. Who will maintain it matters at least as much.
Not for simple workflows; the interface works on a drag-and-drop basis. Installing it on your own server, updating it and debugging complex workflows, however, do require technical knowledge. If nobody on your team can do that, the cloud version or Make is an easier start.
There are two separate line items. One is tool and usage fees (the automation tool and the AI model, plus a server if needed); the other is setup and maintenance effort. The number of workflows and the run volume both push these up; one workflow and ten connected workflows are not the same budget.
Most automation tools process data on their own servers, and a large share of those are located abroad. AI steps also send content to the model provider. Cutting personal data down to the fewest steps necessary is a good start. In Turkey, I recommend doing the legal assessment with your KVKK (Turkey's Personal Data Protection Law) advisor.
A single, simple workflow is up and running in somewhere between a few days and two weeks. Jobs that connect several systems progress in stages, with workflows added one after another. What determines the timeline most is how easily the software you need to connect can be connected.
Last updated: September 27, 2026. I update this guide as the tools' pricing and licensing change.
On October 8, 2026, Google introduced a Gemini work agent that runs with its own email address and audit log; pricing and a general availability date were not announced.
From a workflow that writes form responses to a spreadsheet to appointment reminders on WhatsApp, here are the ten workflows businesses most often build with n8n, each with its trigger, its steps and what to watch out for.
September 25, 2026: On September 24, Adobe opened Photoshop, Lightroom, Express and Firefly tools to all Gemini plans and added Acrobat to its Claude connector. What changes for small businesses, and where should you be careful?
We use cookies to improve the site and measure advertising performance. Analytics and advertising cookies run only if you accept them. Details are in our Privacy Policy and Data Protection Notice.
