Automation Basics with n8n - AI Administrator

What you will be able to do

Open an n8n trial and keep your work safe: export every day.
Say what an automation platform does, and how n8n differs from Make, Zapier and Power Automate.
Tell a trigger from an action, and follow items from node to node.
Read a workflow against a future-state map.
Build, test and publish a Form, Sheets, IF and Gmail workflow.
Add a guard, Retry On Fail and an error workflow.

Prerequisites: Module 4: Workflow Mapping II and your future-state map. You open the free 14-day n8n Cloud trial together at the start of the session (do not sign up in advance). Have an e-mail inbox you can open, and one Google account for both your sheet and the n8n connection.

Session	What happened, or happens
3	The leave request mapped as it really runs: 25 minutes of work, about 4 days of waiting, 8 hand-offs
4	The future state, an Automation lane, and one chosen first build
5 (this module)	That first build running in n8n, and your own first workflow
6	AI steps inside the workflow: classify, extract, draft

1. Open n8n together

Same clicks, same minute, everyone. In the live session the instructor signs up on screen at your pace; anyone stuck is moved to a breakout room by the co-host.

1

n8n.io/pricing, then Start free trial

n8n Cloud. No credit card required.

2

Your e-mail and a password

Later you connect one Google account to Sheets and Gmail: the same account that owns your sheet.

3

Verify

Open the verification e-mail (check spam and Promotions). Nothing after two minutes? Raise your hand in Zoom.

4

Name your workspace

It gets its own address, like yourname.app.n8n.cloud. Write it down.

5

Two onboarding questions

Answer or skip; they only tailor suggestions.

6

Start automating

You land on the overview. Stop there and wait for the group.

Video, recorded live: the real pricing page and sign-up form (nothing submitted), the trial counter, an empty canvas, and Export JSON / Import from file (1 min 22 s, narrated, with captions).

Three rules for the trial

1. It lasts 14 days, and then the workspace is deleted. 2. It allows 1,000 executions; every test run counts as one (the counter is always top left). 3. At the end of every session: the three-dots menu next to the workflow name, Export JSON. Your workflows then live in your Downloads folder, not only in n8n; Import, From file loads them into any workspace.

Sign-up sheet

Day3_S5_n8n_Signup_Sheet.pdf: the six clicks and the three rules on one page.

2. What an automation platform is

A visual editor where you connect boxes. Each box, a node, talks to one application or does one operation on data. You draw the Automation lane from Module 4 as a chain of nodes, and the platform runs it every time the trigger fires. No code, but real logic: branches, retries and error handling.

Platform	Strength	Watch out for	Cost model
n8n	Self-host or cloud; branching; code when needed; AI nodes	More technical settings	Per workflow execution (cloud); free self-hosted
Make	Visual scenarios, many apps	Operations add up fast	Per operation
Zapier	Largest app catalogue, simplest start	Price per task at scale	Per task
Power Automate	Inside Microsoft 365	Licensing tiers	Per user or per flow

The skills transfer: every one of these has triggers, actions, data mapping, conditions and credentials. This course uses n8n because it covers plain automation, AI steps and agents in one tool, and because it can run on your own server, which matters for data that must not leave the organisation.

Trigger (exactly one)

When does this run? A form is submitted, a schedule fires, an e-mail arrives, a webhook is called. In Module 4 terms: the start event of the future-state map. A second event needs a second workflow, or a trigger that covers both.

Actions (one or many)

What happens next, in order: read or write a sheet, send an e-mail, check a condition, call an API. In Module 4 terms: the steps in the Automation lane.

n8n's trigger picker: manually, on an app event, on a schedule, on webhook call, on form submission, when executed by another workflow, on chat message — "What triggers this workflow?" The trigger answers "when does this run?". This module's trigger is *On form submission*.

Data moves as items

Three rows flow into a Gmail node, which runs three times and sends three e-mails — Each node runs once per item it receives: three rows, three e-mails. Many "it sent 40 e-mails" accidents come from forgetting this. (Figure: AI Administrator course material.)

The output of the leave form in JSON view: one item with seven fields plus submittedAt and formMode — An item is a set of fields. One form submission is 1 item with 7 fields (plus `submittedAt` and `formMode`). Switch Schema, Table and JSON to read it; fields are what you drag into the next node.

The one habit that prevents most bugs

After adding any node, click Execute step and read its output panel before you add the next one.

Credentials: sign in once, reuse everywhere

A Gmail node with the Sign in with Google button in its Credential field — *Sign in with Google* creates a credential in n8n's credential store. Nodes refer to it; the secret never sits in a node.

Least privilege: grant the narrowest scope offered.
For production work, use a shared or service account, so the automation survives your holiday and your departure.
Never paste a password or API key into a node field or a sticky note. Module 11 builds on this.

Could not load list, 403 Forbidden, The caller does not have permission — The trap we hit while preparing this module: **403 Forbidden**, "The caller does not have permission". n8n was connected to a different Google account from the one that owns the sheet. Fix: open the credential, switch to the same account as the sheet, and tick every box on Google's consent screen. The sheet must also be a native Google Sheet, not an uploaded .xlsx (File, Save as Google Sheets).

3. Demo: the leave request runs

Module 4's first build, as a real workflow. In the session you watch; you do not build this one. It implements candidates 2 and 3 from your prioritisation matrix: look up the balance, run the policy checks, prepare the manager's decision card.

The Automation lane of the Module 4 future-state map, each step mapped to an n8n node — One step, one node. The form is the trigger; looking up the balance is a Google Sheets node; the policy checks and the decision are a code step and an IF; logging and e-mails are Sheets and Gmail. The manager's decision stays human. (Figure: AI Administrator course material.)

Video, recorded live in n8n Cloud: the form trigger and its two URLs, then three real test requests. Omar's passes and goes to the manager; Sara's (one day's notice) and Lina's (three days, two left) take the false path to HR, with the reason in plain words; every run is kept in Executions (2 min 7 s, narrated, with captions).

The Run policy checks code node: working days, notice needed, balance and quarter end — The policy checks, written once by the instructor: working days, notice needed (2 working days for 1 to 2 days of leave, 10 for longer), the remaining balance, and the last two weeks of a quarter. Every rule is one sentence of the leave policy: exactly the green steps of Module 4. You will not write code in this course; Edit Fields and IF do simple checks.

What the demo deliberately does not do

It does not approve or decline: the manager decides (Day 5 builds the approval gate). It does not chase the manager yet (candidate 4, a reminder, comes later). It does not read the free-text note (Module 6 adds the AI step). We built the high-value, low-risk candidates first.

Compare this with Module 3's current state: no re-typing, the checks run before the manager sees anything, and HR only sees the exceptions. A good map is the workflow design; when they disagree, update the map.

The demo workflow

Day3_S5_Leave_Request_Demo.json: import it with Import, From file, then paste your sheet URL into both Sheets nodes and connect your Google account (the red triangles show what is missing). AI Caravan Ltd., its staff and its leave policy are fictional.

4. Lab: request intake

Build it yourself: in the live session, 40 minutes in breakout rooms of three or four, everyone in their own workspace. A request form, logged to a sheet, with an IF, two e-mails and a guard.

Video, recorded live: the trigger, five fields, a test submission, the Sheets node mapped by dragging fields, and the IF are built from scratch; then the finished lab: e-mails, the guard, Mark incomplete, Retry On Fail, the error workflow setting, three test runs and Publish (2 min 39 s, narrated, with captions).

1

The sheet

Google Sheets, a new sheet Requests with the headers Timestamp, Name, Email, Department, Amount, Description, Status. A native Google Sheet, in the same Google account as n8n. Or start from Day3_S5_Requests_Template.xlsx and use File, Save as Google Sheets.

2

The trigger

A new workflow M5 Request intake. Add first step, On form submission. Fields: Name (required), Email, Department (dropdown: Finance, HR, IT, Operations), Amount (number), Description (textarea). Leave Email and Amount optional on purpose: the guard in step 6 needs something to catch. Execute step, submit the test form once, read the output.

3

Log the row

The + on the form node, Google Sheets, Append row in sheet. Sign in with Google (the account that owns the sheet), pick the document and the tab, then drag each field from the input panel onto its column; submittedAt goes to Timestamp. Status: type the fixed value New. Execute step, then check the sheet.

4

Branch on amount

An If node: drag Amount in, operator Number, is greater than, value 500. Make sure it is a Number operator, not String. Test once above and once below 500.

5

Two e-mails

Gmail, Send a message on each branch: "Request received - needs approval" on true and "Request received - being processed" on false. To: drag Email; a short message using Name and Amount.

6

The guard

An If right after the form: Email is not empty AND Amount is greater than 0. True continues to the sheet. False: Edit Fields sets Status to Incomplete, then a second Sheets node appends the row. No e-mail.

7

Retry and an error workflow

In each Gmail node's Settings: Retry On Fail, Max. Tries 3, Wait Between Tries 5000 ms. Then a second workflow, Error notifier: an Error Trigger, then Gmail to you. Attach it in the lab workflow's menu, Settings, Error Workflow.

8

Test, then publish

Three tests: an amount of 200, 900, and one left blank. Three rows (New, New, Incomplete) and two e-mails. Then Publish (top right), copy the production form URL, and open Executions. Finally: menu, Export JSON.

Trap 1: an empty number is 0, not empty

Tested on 11 October 2026: when Amount is left blank, the n8n form sends Amount = 0. A guard Amount is not empty lets it through, logs it as New and sends an e-mail. Use Amount is greater than 0. The lesson in itself: test the empty case on purpose, do not assume.

The Request intake test form with Name, Email, Department, Amount and Description — The form you build. Name and Department required; Email and Amount optional on purpose, so the guard has something to catch.

The finished lab workflow in the Executions view, run 3 taking the incomplete path — The finished lab, run 3: the blank amount takes the guard's false path, Mark incomplete and Log incomplete. Every run is kept in Executions.

The Requests sheet with three rows: New, New and Incomplete; the e-mail column is blurred — Three tests, three rows: 200 = New (being processed), 900 = New (needs approval), blank = Incomplete with no e-mail.

In your breakout room

Everyone builds in their own workspace; one person shares their screen at a time.
Stuck for more than three minutes? Ask for help (button at the bottom of Zoom).
Execute step after every node, and read the output panel before adding the next.
Most common issues: the wrong Google account (403), an uploaded .xlsx, not clicking Execute before submitting the test form, an IF set to String instead of Number.
Done early? Add the Error notifier, then help someone in your room.

Deliverable

A published M5 Request intake workflow with three test runs (small, large, incomplete), the Requests sheet showing three rows, and the exported JSON file. Optional: the Error notifier attached. Saved as M5-request-intake. Handout: Day3_S5_Lab_Handout.pdf. The solution, to compare after you have built your own: Day3_S5_Request_Intake_Solution.json.

5. Make it safe, then publish

Gmail node Settings: Retry On Fail on, Max. Tries 3, Wait Between Tries 5000 ms, On Error Stop Workflow — Retry On Fail: Max. Tries 3, Wait 5000 ms. Most failures are a mail server or an API that hiccups once. *On Error* below decides what happens after the last try: stop the workflow (default) or continue.

When it still fails: the error workflow

Build once: a separate workflow, Error Trigger, then Gmail to you with the workflow name and the execution link; publish it. Attach to every workflow: workflow menu, Settings, Error Workflow. A silent failure becomes an e-mail, and the link opens the failed run in Executions.

Testing

Execute step runs one node with the data before it; Execute workflow runs the whole thing once. The test URL listens for one submission. Test runs show in Executions with a flask icon.

Published

Publish (top right) makes it live; the production URL works for everyone, any time. From n8n 2.43, a workflow with unconnected required inputs cannot be published. Change something? Test again, publish again. (Older n8n versions and tutorials say Active; same idea.)

Silent success is the real enemy

A failed run you hear about is fine. An automation that fails silently is worse than no automation: people stop checking because they think it works.

Before you close the tab

Menu, Export JSON, for every workflow you built today. The trial deletes the workspace after 14 days; your JSON files are your copy.

Knowledge check

Pick one answer per question, then check your score. These mirror the style of the final exam.

1. How many triggers does a workflow have?

None Exactly one One per action As many as there are apps

Why: Every workflow starts with one trigger; everything after it is an action. A second event needs a second workflow or a different trigger type.

2. A schedule trigger fetches 40 rows from a sheet. How many times does the next node run?

Once 40 times, once per item Twice It depends on the credential

Why: Data flows as items and each following node runs once per item, so 40 rows means 40 executions of the next node.

3. What is the single most effective habit for avoiding broken workflows?

Naming nodes in capitals Running each node after adding it and reading its output panel Using only Gmail nodes Activating before testing

Why: Most failures are a field that is missing or renamed a few nodes back. Checking each node's output catches it immediately.

4. Which credential practice is correct?

Paste the API key into a sticky note on the canvas Use your personal login so it is easy to remember Use a service account with the narrowest scope and store it in the credential store Share one credential with everyone by email

Why: Credentials should be least-privilege, owned by a shared or service account, and never live in node fields or notes.

5. What is the purpose of an Error workflow?

To stop all workflows when one fails To run automatically when any node fails and notify someone with the execution link To delete failed executions To retry the workflow forever

Why: The error workflow turns a silent failure into a notification, which is the first step of monitoring.

Self-check

Answer in your own words first, then open the model answer.

1. Map each node in the lab to a step or tag on your Module 4 future-state map.

Form Trigger = the trigger; Sheets append = the removed REKEY step; IF on amount = a green rule-based decision; Gmail = the DRAFT step with fixed text for now; the empty-field IF = a guard that did not exist before and should be added to the map.

2. Why put an IF that checks required fields before the email node rather than after it?

Because sending is irreversible. A guard before the step that leaves the organisation prevents 'Dear ,' emails; a check afterwards can only report the damage.

3. What would you change to move this workflow from Gmail and Sheets to Microsoft 365?

Swap the Gmail node for Microsoft Outlook and the Google Sheets node for Microsoft Excel 365 or SharePoint; the trigger, IF logic, data mapping and error handling stay identical.

Summary

Key takeaways

One trigger, then actions; data moves as items and each node runs once per item.
Every node implements one step of the future-state map; the human decisions stay human.
Execute step after every node and read its output; test the empty case on purpose.
Credentials live in the credential store, on the same account as your data.
Retry, an error workflow, then publish. Export your JSON every day.

Next, Module 6: an inbox triage: an AI step reads each incoming e-mail and labels it billing, technical or sales, with an "unsure" route that hands the doubtful ones to a person.

Latest reading: n8n and automation at scale (2026)

n8n 2.43.0 release (n8n release notes, 6 October 2026). Stronger guardrails and better publishing: a workflow with unconnected required inputs can no longer be published. The red triangles in the demo now block Publish.
No-Code Workflow Automation at Scale for IT Operations Teams (Intellistack, 16 September 2026). When building outruns governance you get workflow sprawl and unknown owners: why every automation needs an owner (Module 4) and an error workflow (this module).
In The Loop 2026, n8n's first conference (n8n, 13 to 14 October 2026, Berlin). Talks and workshops on production automation and agents; the recordings are a good follow-up to Modules 5 and 6.

References

n8n documentation: docs.n8n.io. Nodes, items, credentials, error workflows and publishing.
n8n Cloud pricing and free trial: n8n.io/pricing (checked 11 October 2026).
Hammer, M. (1990). Reengineering Work: Don't Automate, Obliterate. Harvard Business Review, 68(4), 104-112. Remove before you automate.

All n8n screens are real, from the course trial workspace (October 2026); personal e-mail addresses are blurred. AI Caravan Ltd., its employees and its leave policy are fictional. Further reading: n8n for intelligent workflows.