Lesson 1 of 6 · 15 min · Checked on 6 October 2026
Serious automation: Zapier, Make and n8n
By the end of this lesson you will be able to:
- Describe an automatic flow as a trigger, actions and connections between services
- Tell Zapier, Make and n8n apart by what matters: how you design, who hosts it and how much control it gives
- Design a flow with AI that has a log, human approval and a plan for when it fails
Sixty sign-ups in one weekend
The cultural association of the freguesia (parish) opened sign-ups for the May walk. In two days, sixty came in. Graça, the secretary, spent Sunday copying names from her email into a spreadsheet, replying to each person and sorting out the questions: "can I bring the dog?", "is there a ride back?".
In Level 2 you automated a small flow. Now the step up is to connect several services to each other (the form, the AI, the spreadsheet, the phone) so that repeated work does itself. Graça is left with only what really needs a person.
The concept: trigger, actions and connections
All automation tools rest on the same idea:
- Trigger: the event that starts everything. "A reply arrived on the form."
- Actions: the steps that come next, in order. "Ask the AI to classify", "write a line in the sheet", "send an alert".
- Connections: for each service, you give permission once, by signing in with your account. From then on, the tool reads and writes there on your behalf.
Each time the flow runs, it is called an execution. The tool keeps a log of all of them.
The three best known
Zapier, Make and n8n do all of this. They differ in the way you work:
- Zapier: you design the flow mostly as a list of steps, from top to bottom. It is the most direct way to start. It runs on the servers of the company that makes it.
- Make: you design on a board, with circles joined by lines. It makes paths that split easy ("if it is a question go this way, if it is a simple sign-up go that way"), and so is changing the data midway. It also runs on the company's servers.
- n8n: also a board, with boxes joined together. The big difference: you can use it on their service or install it on a computer of your own (self-hosting). It gives more control over where the data ends up, and it asks for more technical knowledge from whoever installs and maintains it.
None of them is "the best". Ask yourself: do I need something very simple? Do my flows have a lot of "ifs"? Is it important that the data does not leave home? For prices and plans, see the Current state page.
This is how it looks on 6 October 2026; your screen may differ. Look for the button with the same name.
Tip
Before you open any tool, write the flow on paper, one step per line. If you cannot explain it on paper, the tool will not understand it either.
A complete flow, step by step
Graça's flow, in plain words:
1. Trigger: new reply on the form. The form has name, contact, number of people and a free field, "Remarks".
2. Action: the AI classifies and writes a draft. The flow sends the AI only the "Remarks" field, always with this fixed request:
You receive the remarks from a sign-up for a walk organised by an association.
1. Classify them into one of these categories: NO_REMARKS, QUESTION, SPECIAL_REQUEST, OTHER.
2. If it is QUESTION or SPECIAL_REQUEST, write a short, friendly reply in European Portuguese. Do not promise anything that is not in these rules: [paste the rules of the walk here].
3. Answer only in this format: CATEGORY | REPLY
Notice that the name and the contact do not go to the AI. They are not needed to classify.
3. Action: a line in the spreadsheet. Name, contact, number of people, category and the draft reply. The sheet becomes the official list.
4. Action: an alert on the phone. If the category is QUESTION or SPECIAL_REQUEST, Graça gets a message: "Sign-up with a question. Draft ready in the sheet." If it is NO_REMARKS, the flow sends the confirmation with a fixed text that Graça wrote and reviewed once.
Of the sixty sign-ups, Graça only has to look at the ten that need her.
Example
Mrs Lurdes writes in the remarks: "I am 74 and I walk slowly, can I come?" The AI classifies it SPECIAL_REQUEST and suggests this reply (translated here): "Of course you can! The route is 6 km and we go at the group's pace." Graça reads it, sees that the route is 9 km and not 6, corrects it and sends it. This is what approval is for.
Four good practices
One flow, one task. Sign-ups in one flow, shop orders in another. When one fails, you know where to look.
Look at the log. All these tools show the executions: what went in, what came out of each step, where it stopped. In the first weeks, have a look every day.
A person approves what goes out. A fixed text that you wrote can go out by itself. A text written by the AI on the spot stays as a draft until someone reads it.
Think about the day it fails. The AI service may be down, the email connection may expire, someone may rename the sheet. Ask: if step 2 fails, is the sign-up lost? In the design above, it was, because the line is only written in step 3. The fix is to save first, think later: write the line in the sheet right after the trigger and only then call the AI. And turn on the tool's error alert (all of them have a way of warning you when a flow fails).
Watch out
Each service the data passes through (the form, the automation tool, the AI, the sheet) receives a copy. If it is personal data of members or customers, the association or the company has to know where it goes and under what contract. Send the AI only the fields it needs. This is not legal advice.
Tip
When a connection asks for permission to touch "all your files", see whether there is an option to give access to just one folder or one sheet. Less access, less damage if something goes wrong.
Exercise: your flow on paper
Review this design of an automation. Tell me: where data can be lost if a step fails; where text written by the AI goes out without a person reading it; and what data I am sending to the AI without need.
[paste your flow here, one step per line]
Exercise
Choose a repeated task from your work: quote requests, bookings, sign-ups, complaints.
How I know it worked a page of 4 to 7 lines in which the data is saved before the first AI step, nothing written by the AI goes out without approval, and every failure has an answer ("it stays in the sheet, I check at the end of the day").
If it didn't work
- Did it end up with more than ten steps? Most likely it is two flows. Separate them.
- Do you not know what happens when it fails? Open the tool's help and look for "error" or "failed execution": all of them explain where to look and how to run it again.
- Are you going from paper to the tool? First use a form and a test sheet with invented data, before you connect to the real list.
Videos
A no-code introduction to n8n: what it is, how it works, when to use it, example nodes, building a flow and an AI flow.
An official n8n video: builds a Q&A AI agent with a knowledge base in two workflows (ingest and agent), then tests it and reviews past executions.