Learning goals
- You can explain the logic of triggers and actions.
- You use an AI step with a fixed output format.
- You plan error handling, monitoring and documentation.
- You choose a suitable platform.
Trigger, action, data
Every automation follows the same pattern. A trigger starts it: a new email, a submitted form, a calendar event. Actions follow: create a record, send a message, file a document. In between, data flows from step to step.
An example: a contact form is submitted (trigger). An AI step assigns the request to a category. Depending on the category, it lands in the CRM or with service, and the responsible team gets a message in Slack or Teams (actions).
Delivery Hero shows what such a workflow can deliver. According to an n8n case study, lock-out time for around 800 account recoveries a month fell from 35 to 20 minutes, about 200 hours a month in total. The first workflow took five hours to build. Like every vendor case study, this is a company claim.
The AI step
AI is needed in workflows where unstructured text has to take a clear form. For the next step to process the answer, the output format must be fixed.
Assign the following request to exactly one category: invoice, complaint, appointment request, application, other. Reply with the category name only, no other text. Request: {{text}}For several fields, JSON works well. Always plan a path for unclear cases, such as the category "other" or the value "UNCLEAR", and route those cases to a person.
Running it reliably and choosing a platform
A workflow is only finished when it is documented and reports errors. For each workflow, record: purpose, owner, systems involved, data class, fallback plan. Set up a notification for failed runs.
Three questions decide the platform:
- Data sovereignty: Must data stay in-house? Then self-hosted n8n. The Community Edition is free; operation and updates are on you.
- Ecosystem: Do you work mainly in Microsoft 365? Then Power Automate.
- Team: Do business teams build without IT support? Then Zapier with its large app catalogue, or Make for visual, branching workflows.
Watch the pricing model: Zapier counts every step as a task, n8n every execution.
Your first workflow
Build a small workflow with one AI step.
Do the task in your day-to-day work
You do this task in your day-to-day work, not here in the course. Tick off the steps once you have done them.
What comes out of it: A running, documented workflow with a test log.
Enter your result here
Enter what came out of the task. Everything is saved automatically. Your results from modules 3 and 4 are already waiting here.
See an example
This is what a filled-in result could look like. Your entries stay unchanged.
Your first workflow
- Use case
- Draft replies to complaints
- Platform
- n8n
- Trigger
- New email in the service@ inbox
- AI step and categories
- Classify (delivery, product, invoice) and draft a reply
- Actions
- Create a ticket, save the draft in the ticket, notify the team lead for invoice issues
- Test result, e.g. 9 of 10 correct
- 18 of 20 classified correctly
- Owner
- Customer service team lead
- Data class
- Internal
- Failure notification set up
- yes
- Workflow documented
- yes
From module 3 Your top 3 open
Module 3 has no entries yet. Go to module 3 →
From module 4 One week, two routines open
Module 4 has no entries yet. Go to module 4 →
Used again in modules 7 and 8
Your result appears there automatically, and you can take it over with one click. You will also find all results in the workbook.
Quiz
1. What starts an automation?
2. Why does an AI step in a workflow need a fixed output format?
3. According to the n8n case study, how much lock-out time did Delivery Hero save per month?
4. Your workflow handles strictly confidential data and you have in-house IT capacity. Which option fits best?