Module 2: Automation Fundamentals
AI Automation Services
Tiago Forte (of "Building a Second Brain" fame) put it perfectly: "Don't automate a broken process. Fix the process first, then automate it." I've seen automation consultants spend 20 hours building an elaborate workflow that automated a process the business shouldn't have been doing in the first place.
The Automation Stack (What You Need to Know)
Layer 1: Triggers
Everything starts with a trigger — something that kicks off the automation. Common triggers:
- A form is submitted
- An email arrives
- A new row appears in a spreadsheet
- A webhook fires from another app
- A schedule (every day at 9am)
- A new record is created in a CRM
Layer 2: Logic
After the trigger, you need logic — decisions about what happens next:
- If/then branching (if the email is a complaint, go left; if it's a question, go right)
- Filters (only process emails from @company.com)
- Loops (do this for each item in a list)
- Error handling (if something fails, do this instead)
Layer 3: AI Processing
This is where it gets interesting. Between trigger and action, you insert AI:
- Text analysis: Read an email and determine sentiment, category, or urgency
- Content generation: Draft a response, create a summary, write a description
- Data extraction: Pull structured data from unstructured text (invoice number from an email, key details from a document)
- Classification: Sort items into categories based on content
- Translation: Convert between languages
Layer 4: Actions
What happens at the end:
- Send an email or Slack message
- Update a CRM record
- Create a document
- Post to social media
- Add a row to a database
- Trigger another automation
The Three Automation Platforms You Should Know
Make.com (Primary — Learn This First)
The most powerful no-code automation platform for AI work. Visual builder, excellent AI integrations, handles complex logic well. Most AI automation agencies use this as their primary tool.
Strengths: Visual flow builder, powerful data manipulation, great API support, AI module built-in.
Weaknesses: Learning curve is steeper than Zapier, occasional reliability issues.
Cost: Free tier (1,000 operations/month), then $9-16/month for most use cases.
Zapier (Secondary — Know This Too)
The most popular automation tool. Simpler than Make, but less powerful for complex AI workflows. Some clients will specifically request Zapier because they know it.
Strengths: Massive app library (6,000+ integrations), simple interface, most clients know the name.
Weaknesses: Expensive at scale, less flexible for complex logic, AI features are catching up.
Cost: Free tier, then $19.99+/month.
n8n (Advanced — Optional)
Open-source, self-hosted automation platform. Full control, no per-operation pricing, very powerful. Best for clients who want to own their infrastructure.
Strengths: Self-hosted (no per-operation costs), extremely flexible, open-source.
Weaknesses: Requires technical setup, smaller app library, less polished UI.
Cost: Free (self-hosted) or cloud plans from $20/month.
Building Your First AI Automation (The Pattern)
Almost every AI automation follows this pattern:
1. Trigger: Something happens (new email, form submission, scheduled time)
2. Fetch data: Get the relevant information
3. AI processes: Send data to ChatGPT/Claude with a specific prompt
4. Parse response: Extract the structured output from the AI
5. Act: Do something with the result (send email, update CRM, create document)
6. Log: Record what happened (for debugging and reporting)
Example — AI Email Classifier:
1. Trigger: New email arrives in support inbox
2. Fetch: Get email subject and body
3. AI: "Classify this email as: Complaint, Question, Feature Request, or Spam. Extract the customer's name and the main issue in one sentence."
4. Parse: Extract classification and summary from AI response
5. Act: Route to correct team in helpdesk software, create ticket with AI summary
6. Log: Record in Google Sheet for reporting
That automation takes 2-3 hours to build and saves a support team 5-10 hours per week. At £2,000 to build, the client breaks even in their first month.
Common Mistakes in Automation Building
Mistake 1: Over-automating. Don't automate every edge case. Build for the 80% of cases, and have exceptions flag for human review. Trying to handle 100% of cases makes automations fragile and expensive.
Mistake 2: No error handling. Automations fail. APIs go down. Data formats change. Build in error notifications (Slack/email alert when something fails) and retry logic.
Mistake 3: No documentation. If you can't explain what an automation does in plain English, it's too complex. Document every automation so anyone can maintain it.
Mistake 4: Ignoring rate limits. AI APIs have rate limits. If you process 1,000 emails at once, you'll hit the limit. Build in delays and batch processing.
---
Automation Design
Design an AI automation for this business process: Business: Online tutoring company Process: When a student submits a homework assignment (uploaded to Google Drive), the system should: - Read the document - Check it against the assignment rubric - Generate feedback comments - Score it on a 1-100 scale - Email the student their feedback and score - Update a tracking spreadsheet Design the automation: 1. Which platform (Make.com, Zapier, or n8n) and why 2. Step-by-step flow with each module/action 3. The AI prompt for grading and feedback 4. Error handling for common failure points 5. Estimated build time and suggested price to charge
Process Mapping
Help me map a business process for automation. The business: A recruitment agency that receives 200+ CVs per week via email for various open positions. Current manual process: 1. Admin opens each email 2. Downloads the CV 3. Reads it and decides which job role it fits 4. Enters key details into their ATS (name, email, experience, skills) 5. Sends an acknowledgement email to the candidate 6. Flags promising candidates for the recruiter Map this process for automation: - Which steps can be fully automated? - Which steps need AI? - Which steps should remain human? - What's the trigger and what are the actions? - What data needs to flow between steps? - Where are the likely failure points?
AI Prompt Engineering for Automation
Write an AI prompt for use inside a Make.com automation. The prompt needs to: Task: Classify incoming customer support emails for an e-commerce store Categories: Order Status, Returns & Refunds, Product Questions, Complaints, Spam/Other Required output: JSON format with fields: category, urgency (low/medium/high), customer_name, order_number (if mentioned), one_line_summary The prompt must: - Be clear and unambiguous - Handle edge cases (emails that fit multiple categories) - Always return valid JSON - Include 2 example inputs and outputs for few-shot learning - Be optimised for GPT-4 (concise but thorough)
1. Sign up for Make.com (free tier).
2. Complete the "Getting Started" tutorial on Make.com.
3. Build this automation:
- Trigger: Receive a webhook (use Make's built-in webhook module)
- AI step: Send the incoming data to OpenAI with a prompt that summarises and classifies it
- Action: Send yourself a Slack message or email with the AI's output
4. Test it by sending a test webhook with sample data.
5. Screenshot your working automation — this is the start of your portfolio.
---
- 1**Fix the process before automating it** — automating a broken process just makes it break faster
- 2**Every AI automation follows the same pattern:** Trigger → Fetch → AI Process → Parse → Act → Log
- 3**Learn Make.com first** — it's the most powerful platform for AI automation
- 4**Build for 80%, not 100%** — handle common cases automatically, flag edge cases for humans
- 5**Error handling isn't optional** — every automation needs failure alerts and retry logic