AI automation is a chain of steps that starts with an event (a trigger) and includes at least one AI model step that reads, sorts or writes text. The best first project is “form → AI summary → spreadsheet → email or Telegram”. You can build it in one evening in Zapier (simplest), Make (visual and flexible) or n8n (most control, and it can run on your own server).
Process automation is not new: forms have filled spreadsheets and emails have gone out automatically for years. What is new is a step that understands free text: it can tell that a customer is complaining, summarise a long email and draft a reply.
What is AI automation, and how is it different?
Every automation has three parts:
- Trigger: the event that starts the flow, such as a new form response, a new email or every Monday at 9:00.
- Actions: the steps that follow: add a row, send an email, create a CRM record.
- Logic: conditions and branches. If urgent, go one way; if not, another.
A regular step behaves the same way every time. An AI step is probabilistic: it can return slightly different answers for the same text and sometimes gets it wrong. So an AI step always needs a clear instruction, a fixed output format and a check.
The vocabulary differs: Zapier calls a flow a “Zap”, Make a “scenario”, n8n a “workflow”. The idea is the same.
n8n vs Make vs Zapier: comparison table
Data as of October 2026. Prices and limits change often, so check the official pages before you choose.
| Zapier | Make | n8n | |
|---|---|---|---|
| What you pay for | Tasks: each successful step | Credits: most module actions cost 1 | Executions: one full workflow run |
| Free option | 100 tasks a month, two-step Zaps only | 1,000 credits a month | Community Edition on your own server |
| Self-hosting | No | No | Yes |
| Learning curve | Easiest | Medium | Steepest |
| AI step | AI by Zapier (Pro and up) | AI modules, Make AI Agents | AI Agent node, many models |
| Best for | Quick, simple flows | Complex visual scenarios | Control, data stays with you |
Two details change the choice. First, the Zapier free plan only allows two-step Zaps (a trigger and one action), and our example has five steps, so you will need the Pro plan. Second, n8n Community Edition is free, but running the server (Docker or npm) is on you, and its Sustainable Use License allows free use for internal business purposes. n8n Cloud can be tested at no cost, with a limited number of executions.
Your first workflow, step by step
The scenario: your website has a “Contact us” form. You get 10–30 requests a day: orders, questions, complaints, all read by one person. The goal: AI sorts and summarises each one, everything is logged in a sheet, and urgent requests send the manager a Telegram message.
Step 1. Draw the process on paper. Before opening any tool, write down what comes in, what should happen and who checks the result. If you cannot draw it, you cannot automate it.
Step 2. Prepare the sheet. In Google Sheets, create columns: date, name, email, message, type, summary, urgent, status.
Step 3. Trigger. Choose “new form response” (Google Forms, Tally, Typeform or n8n’s own form). Submit one test response so the tool can see the fields.
Step 4. AI step. Add a model step (AI by Zapier, a Make AI module or an n8n LLM node) with an instruction like this:
“You are a customer service assistant. Read the message and return only JSON, no other text. Fields: type (exactly one of: order, question, complaint, other), summary (one sentence, max 20 words), urgent (true if the customer mentions a deadline, damage or a refund, otherwise false). If unsure, set type to other. Message: …”
A correct answer looks like this:
{"type": "complaint", "summary": "The customer received a damaged order and asks for a refund.", "urgent": true}Clear categories, a fixed format and a default value (“other”) make the AI step stable. More on this in our guide to writing a good prompt.
Step 5. Write to the sheet. Add “create row in Google Sheets” and map each column: name and message from the form, type, summary and urgency from AI.
Step 6. Branch. Condition: if urgent = true, a Telegram bot sends the manager the summary and a link to the sheet (you create the bot in Telegram with @BotFather). If not, the customer gets an automatic email: “We’ve received your message and will reply within 24 hours.”
Step 7. Test. Before switching it on, send 10 test messages: empty, very long, in another language, spam, ambiguous. See where AI slips and improve the instruction.
Step 8. Go live and monitor. For the first month, check the sheet once a week: take 20 random rows and see whether AI got the type and urgency right. If the same mistake repeats, add an example to the instruction.
Why not use n8n’s AI Agent node here? An agent decides for itself which tool to use and in what order. In our task the steps are known in advance, so a simple LLM step is cheaper and more predictable.
What does the AI step cost? API keys
One detail surprises beginners: a ChatGPT or Claude subscription and their API are different things. In Make and n8n you often connect a model with an API key, billed separately by the amount of text processed (tokens). Sorting short messages is cheap, but set a spending limit in the provider’s dashboard so a looping flow cannot burn your budget.
Error handling: what happens when something breaks?
Automations break quietly: the form changed, a password expired, AI returned prose instead of JSON. Set up error handling from day one:
- n8n: in the workflow settings, choose an error workflow that starts with the Error Trigger node and sends you the details (which workflow, which node, what message).
- Zapier: add an error handler to a step. The Zap splits into a success path and an error path, and the error path runs only when that step fails. This is available on Professional plans and above.
- Make: add an error handler route to a module and decide whether the scenario stops, skips or retries.
For the AI step, add your own check too: if the answer is not valid JSON or the type is not on the list, log the row with status “needs review” for a person to look at.
Data privacy in automation
Every step means your data passes through another service: the form provider, the automation platform, the AI model provider, the spreadsheet. A few rules:
- Minimum data: send AI only the message text, not ID numbers or card details.
- Least privilege: connect accounts that can only reach the sheet and mailbox they need. OWASP recommends the same: an AI system should have no more permissions than its task requires.
- Passwords and keys: only in the tool’s credentials section, never in a text field or a sheet.
- Your own server: self-hosted n8n keeps data on your infrastructure, but the text still goes to the AI provider if you use a cloud model.
For the first two weeks, let AI prepare drafts only and have a person press send. Count how often you corrected the AI. Only when corrections are rare should the flow be allowed to send by itself.
When is automation not worth it?
Automation is not always the answer. Do not start if a task happens twice a month, if the process itself is still unclear, or if a mistake causes serious harm (contracts, money transfers). For ideas on what is worth automating, see our list of 15 AI use cases in business.
When the sequence of steps can no longer be written in advance and the system has to decide which tool to use, you move from automation to an agent. Anthropic’s engineers draw exactly this line: a workflow follows a predefined path, while an agent directs its own process. Details in our article on AI agents.
Frequently asked questions
Is n8n free?
n8n Community Edition is free if you install it on your own server (Docker or npm) and use it for internal business purposes. n8n Cloud is paid and billed by executions. If you do not want to run a server, n8n Cloud or Make is an easier start.
Which should a beginner choose: Zapier, Make or n8n?
For a simple flow you want quickly, Zapier. For many branches and steps where price matters, Make. For full control, your own server or complex AI agents, n8n. Many companies use two of them side by side.
Do I need to code to build an n8n workflow?
Not for basic flows: all three tools are visual. Understanding JSON and simple expressions helps a lot, especially in n8n.
Does AI automation work with Georgian text?
Yes. Current models sort and summarise Georgian messages well. Write the category names and instructions clearly and check results by hand in the first weeks, especially with informal or mixed-language messages.
Where can I take an AI automation course?
The official docs (n8n Docs, Zapier Help Center, Make Help Center) are a good start. If you want to build flows on your own tasks with a group and a trainer in Tbilisi, our course AI for Business: Automation and Agents covers Zapier, Make and n8n.
Next step
Pick one process that repeats every day and build the flow above with a test form. If you want it, and a few more, running in your own accounts by the end of a course, look at AI for Business: Automation and Agents or contact us for a free consultation. If the whole team needs automation, see the corporate program and our other courses on the AI courses page.
Sources
- n8n Plans and Pricing, n8n
- Error handling, n8n Docs
- Sustainable Use License, n8n on GitHub
- Zapier pricing and plans, Zapier
- Set up custom error handling, Zapier Help Center
- Make pricing, Make
- Building effective agents, Anthropic
- LLM06:2025 Excessive Agency, OWASP GenAI Security Project
