Project 3 — Campaign Automation with Make

Section 10.8.3: Project 3 — Campaign Automation with Make

🎯 Learning Objective

Design and implement a complete Make scenario automating a sales or marketing campaign, integrating ChatGPT for personalization at scale. This project combines automation skills (Chapter 6) with prompt engineering (Chapter 2) in a high-ROI business use case.

Why AI Automation Is a Commercial Game-Changer

Traditional B2B commercial prospecting is a massive time sink: a salesperson spends on average 65% of their time on non-revenue-generating tasks (prospect research, email drafting, administrative follow-up). AI automation can reduce this time to ~20%, by automating research, qualification, and personalization.

The benefit is twofold:

1. →Volume: process 10x more prospects without hiring
2. →Quality: each email is personalized to the prospect's profile (no generic mail-merge)

An AI-personalized email has an open rate of 35-45% compared to 15-20% for a classic template email (Lemlist/Apollo 2025 data). The response rate goes from 2-3% to 8-12%.

The Project Brief

Context: you need to automate the commercial prospecting process of a B2B company, from lead generation to appointment booking.

Deliverables:

1. →Make scenario architecture (detailed diagram with each module)
2. →Functional Make scenario (screenshots + JSON export of the scenario)
3. →ChatGPT prompts used (documented with explanation of each variable)
4. →Test report with 20 executions (results + qualitative analysis)
5. →ROI calculation and optimization recommendations

Difficulty Level: ⭐⭐⭐⭐ (advanced — requires skills from chapters 2, 4, and 6)

Scenario Architecture

Phase 1 — Data Preparation (1h)

Step 1a: Create the Prospect Database

In Google Sheets, create a list of 20 prospects (fictional for testing, real for production) with the following columns:

Step 1b: Enrich the Profiles

Before writing to someone, you need to understand their context. Use ChatGPT to pre-enrich:

Based on this information about a B2B prospect, enrich the profile:

Name: {{firstName}} {{lastName}}
Position: {{position}} at {{company}} ({{industry}}, {{size}})
Source: {{source}}

Research and return:
1. The 3 main likely challenges for this type of position in this industry in 2026
2. The industry jargon this person probably uses daily
3. An ultra-personalized hook based on their lead source
4. The best time to contact them (insight on the persona's habits)

Step 1c: Build the Scoring and Personalization Prompt

This is the most critical prompt of the scenario — it determines the quality of the ENTIRE pipeline:

You are a B2B sales intelligence expert. Analyze this prospect profile and return ONLY valid JSON (no surrounding text):

Prospect: {{firstName}} {{lastName}} — {{position}} at {{company}}
Industry: {{industry}} | Size: {{size}} employees
Lead source: {{source}}
Likely pain point: {{painPoint}}

Our offer: [describe your product/service in 2 lines]

{
  "score": [1-10, based on: position/offer fit (40%), company size (20%), priority industry (20%), warm/cold source (20%)],
  "justification": "[score reason in 1 sentence]",
  "segment": "[decision_maker | influencer | user | no_fit]",
  "approach_angle": "[the unique personalized hook for this prospect — NOT generic]",
  "likely_pain_point": "[the SPECIFIC problem our solution solves for THEM]",
  "adapted_social_proof": "[which case study or metric would resonate most with this profile]",
  "personalized_email": {
    "subject": "[email subject, < 50 characters, personalized to name or company]",
    "body": "[email of 5-7 lines maximum. Human tone, NO visible framework. Start with a reference to the prospect's situation, not with 'let me introduce myself.' End with an open question, NOT with 'Would you be available for a call?']"
  },
  "follow_up_angle": "[different angle for the D+3 follow-up, NOT a simple 'just following up']"
}

Phase 2 — Building the Make Scenario (2h)

Step 2a: Create the Trigger

1. →Open Make.com → Create a new scenario
2. →Add the Google Sheets > Watch Rows module as trigger
3. →Configure: select your spreadsheet, the "Prospects" sheet, mode = "New rows only"
4. →Test: add a row to the Google Sheet → verify that Make detects it

Step 2b: Add the ChatGPT Module

1. →Add the OpenAI > Create a Completion module
2. →Model: choose GPT-4o (best quality/cost ratio for scoring, no need for GPT-5 here)
3. →System Message: paste the scoring prompt above
4. →User Message: map the Google Sheet variables: Prospect: {{1.LastName}} {{1.FirstName}} — {{1.Position}} at {{1.Company}}...
5. →Temperature: 0.3 (for consistent and reproducible scoring)
6. →Response Format: JSON object

Step 2c: Parse the JSON Response

1. →Add a JSON > Parse JSON module
2. →Source: the ChatGPT response ({{2.choices[0].message.content}})
3. →This gives you access to fields: score, personalized_email.subject, personalized_email.body, etc.

Step 2d: Route by Score

1. →Add a Router with 3 branches:
   • →Branch 1 (filter: score > 7) → hot prospects → immediate email
   • →Branch 2 (filter: score between 4 and 7) → warm prospects → nurturing
   • →Branch 3 (filter: score < 4) → archive

Step 2e: Configure Email Sending (Branch 1)

1. →Module Gmail > Send an Email
2. →To: {{1.Email}} (from Google Sheet)
3. →Subject: {{3.personalized_email.subject}} (from parsed JSON)
4. →Body: {{3.personalized_email.body}}
5. →Important: add a footer with unsubscribe link (GDPR compliance)

Step 2f: Configure Nurturing (Branch 2)

1. →Module Trello > Create a Card (or any other management tool)
2. →Board: "Prospects to Nurture"
3. →Card name: {{1.LastName}} — {{1.Company}} (Score: {{3.score}})
4. →Description: {{3.approach_angle}} + {{3.likely_pain_point}}

Step 2g: Configure Tracking (All Branches)

1. →Module Google Sheets > Update a Row at the end of each branch
2. →Added columns: "AI Score", "Segment", "Action", "Processing Date", "Email Sent (Y/N)"

Step 2h: Add Automatic Follow-up

1. →After the Gmail module, add a Sleep module of 3 days (259200 seconds)
2. →Then a second OpenAI module with a follow-up prompt:

Write a follow-up email for this prospect who didn't respond to my first email.

First email sent:
Subject: {{initial_subject}}
Body: {{initial_body}}

Prospect profile: {{name}} — {{position}} at {{company}}
Initial angle: {{approach_angle}}
Suggested follow-up angle: {{follow_up_angle}}

Constraints:
- Max 4 lines
- Different tone from the first email (if the first was formal, be more casual)
- Bring NEW value (an article, a data point, a case study)
- End with a short open question
- DO NOT say "I'm just following up" or "further to my previous email"

3. →Then a second Gmail > Send an Email module to send the follow-up

Phase 3 — Testing and Measurement (1h)

Step 3a: Controlled Execution

Execute the scenario on your 20 test prospects. Do NOT launch in automatic mode right away! Use "Run once" mode and verify each execution:

1. →Is the scoring consistent? (a SMB CEO in your target industry should score > 7)
2. →Is the personalization truly personalized? (differs between prospects)
3. →Is the email sendable as-is? (no [placeholder], no robotic phrasing)
4. →Does the router send to the correct branch?

Step 3b: Evaluation Grid per Prospect

Key Metrics to Calculate:

• →Scoring agreement rate: % of prospects where AI Score and Human Score are within the same range (±2 points)
• →Average email quality: average rating on generated emails
• →Real personalization rate: % of emails that mention a prospect-specific element (not generic)
• →Error rate: % of emails with errors (wrong name, fabricated information)

Phase 4 — Optimization and ROI (1h)

Step 4a: Iterate on the Scoring Prompt

If the scoring agreement rate is < 70%, improve the prompt:

• →Add examples: "A Marketing Director in a 50-200 person SaaS SMB = score 8. An intern = score 2."
• →Specify the weightings: "Position accounts for 40% of the score, industry 30%, size 20%, source 10%"
• →Test with the same 20 prospects and compare

Step 4b: Improve Personalization

If the emails are too generic:

• →Add to the prompt: "The email MUST mention at least 1 specific element: the exact industry, a known challenge for this position, or the lead source."
• →Test: ask a colleague to read 5 emails without seeing the names → can they guess who each email is for?

Step 4c: ROI Calculation

Calculate the ROI of this prospecting automation scenario:

Test data:
- 20 prospects processed
- Total setup time: [X] hours
- Time per prospect with the scenario: [X] min (estimate: verification + final personalization)
- Time per prospect WITHOUT scenario (manual method): [X] min

Costs:
- Make subscription: [free tier or X€/month]
- OpenAI API (GPT-4o): ~€0.005 per prospect (scoring prompt)
- Human time: [hourly rate]

Projections for 500 prospects/month:
1. Time saved in hours
2. Total cost (Make + API + human supervision time)
3. Cost per qualified prospect sent
4. ROI in % vs. manual method
5. Breakeven: at how many prospects does the scenario become profitable?

Typical Calculation Example:

Step 4d: Optimization Recommendations

Document possible improvements:

• →Short term: refine the scoring prompt, add examples, expand data sources
• →Medium term: add a response detection module (Gmail > Watch Emails) to automatically update the tracker
• →Long term: integrate a CRM (HubSpot, Pipedrive), add behavioral scoring (email opens, link clicks)