AI Automation for Beginners:
Build Your First Automation This Weekend

Before you start: setting realistic expectations

Your first AI automation will not be perfect. It will probably have a few failure modes you did not anticipate. It may need prompt revisions after you see it run on real inputs. This is normal — every experienced automation practitioner's first automation had these same characteristics. The goal of your first automation is not perfection; it is one working automation that saves you real time and teaches you how this actually works in practice.

Realistic expectations for your first automation: 60-75% of outputs will be usable as-is or with minor edits (not 100%). You will find 2-4 edge cases in the first week that your prompt does not handle well. After 2-3 iterations of prompt refinement over the first two weeks, you will be at 80-90% quality. This is the normal learning curve — not a sign that you are doing it wrong.

The exact beginner stack: free to start

The tools you need, in order of setting them up:

1. Make.com account (free): Go to make.com and create a free account. The free tier gives you 1,000 operations per month and up to 15 minutes polling interval — enough for learning and low-volume first automations.

2. OpenAI account and API key: Go to platform.openai.com, create an account, add a payment method (required to generate API keys), and set a monthly spending limit of $10 (Settings → Limits → Monthly budget). Generate your first API key. Your first few weeks of learning will likely cost $1-3 in API fees.

3. Gmail account (you probably already have one): The Gmail trigger in Make.com is the most accessible real-world automation trigger to start with. Your existing Gmail account works perfectly.

4. Google Sheets (free): Create a new Google Sheet called "Automation Log." This will be your monitoring log — the place where every automation run records its results.

Your first automation: email classification in 3 hours

This is the automation most people build first, delivers the most immediate value, and teaches the most transferable concepts.

Step 1 (30 min): In Make.com, create a new scenario. Add a Gmail "Watch Emails" trigger. Connect your Gmail account via OAuth. Set it to watch your Inbox, maximum 5 emails per run, every 15 minutes. Click "Run once" — you should see one of your recent emails appear as data in Make.com. If you do, the trigger is working.

Step 2 (45 min): Add an OpenAI "Create a Completion" module. Connect your API key. Model: gpt-4o-mini. Temperature: 0.1. Max tokens: 150. In System message, paste your classification prompt (see template below). In User message, map the Gmail Subject and Body Plain fields using the data picker.

Step 3 (20 min): Add a Google Sheets "Add a Row" module. Connect your Google account. Select your Automation Log spreadsheet. Map: timestamp, email subject, OpenAI response. Click "Run once" and verify a row appears in your sheet.

Step 4 (15 min): Add a Gmail "Add Label" module. Connect your Gmail account. Map the label name to the category from the OpenAI response JSON.

Step 5 (30 min): Test thoroughly. Click "Run once" 3-4 times. Review the Google Sheet. Review the Gmail labels applied. Check that the OpenAI responses are correctly formatted JSON. Toggle the scenario Active.

Classification system prompt template:

You are an email triage assistant.
Classify each email into exactly one category.

CATEGORIES:
- BILLING: payment, invoice, subscription, charges
- SUPPORT: technical help, feature questions, bug reports
- SALES: new business enquiry, pricing request, demo request
- ADMIN: scheduling, logistics, internal coordination
- PERSONAL: social, personal relationships
- OTHER: anything that does not fit above

Return ONLY valid JSON: {"category": "CATEGORY", "urgency": 1-5, "summary": "max 10 words"}

Urgency: 1=no rush, 3=within 24 hours, 5=respond immediately
Return ONLY the JSON. No other text.

What to do in week two

After your first automation has run for one week: check the monitoring log. Look for any emails that were misclassified (you will know from the label applied). Note the specific types of email that were misclassified. Add handling for those types to your system prompt. Run the updated prompt on 10 historical examples to verify it handles the new cases correctly. Deploy the updated prompt. This is the normal iteration process — prompt refinement based on real production data is how automations improve from 70% to 90%+ quality.

Week two goal: add one more automation. The highest-value second automation for most people is meeting summarisation (if you have regular meetings with transcripts available) or weekly report generation (if you produce a weekly status update). Both follow the same pattern as email classification: trigger → AI process → output → log.

FAQ