GloBox A/B Test Analysis

🧪 The Project

Analyze the results of an A/B test for a website banner for GloBox, a fictional company, and provide a data-backed recommendation to stakeholders on whether to launch the new banner experience.

---

🏢 Business Context

GloBox is primarily known for boutique fashion items and high-end decor. However, their food and drink offerings have grown significantly in recent months. To increase visibility and drive revenue in this category, they launched an A/B test on their mobile website, featuring a promotional banner at the top.

---

🎯 Objectives

• Boost visibility of the Food & Drink section
• Increase revenue

---

📊 Key Metrics

• Conversion Rate / Conversion Lift
• Average Amount Spent per User

---

🧠 Process Overview

1. Data cleaning, transformation, and exploratory analysis were conducted in Beekeeper Studio using PostgreSQL.
2. Cleaned datasets were exported to Tableau Desktop for visual exploration.
3. The same dataset was also analysed with Python for:
   • Hypothesis testing
   • Effect size estimation
   • Confidence intervals
4. A separate export was created to analyse novelty effects.
5. Final recommendations and visual insights were compiled in the attached slide deck.

---

🔍 Key Insights

• 📈 The treatment group showed a significant increase in conversion rate, but no meaningful change in revenue.
• 🌍 Highest conversion rates were seen among:
  • Users from Mexico
  • Male users
  • Android users
• 🕒 No novelty effects were observed over time.

---

✅ Recommendation

Since only one of the two business goals was met, I recommended re-running the test with adjusted targeting and design, as detailed in the slide deck and technical report.

---

🗂 Files Included

• README.md – This document
• EDA.sql – All SQL queries used for:
  • Exploratory analysis
  • Conversion metrics
  • Group comparisons
  • Tableau/Python exports
• Globox_statistical_analysis.py – Python script containing:
  • Hypothesis testing on conversion rates and average spend
  • Confidence interval calculations
  • Effect size estimation using Cohen's h and d
  • Visualisations with error bars
• Slide deck – Final summary presentation (attached separately)

---

🛠️ Tools Used

• PostgreSQL (via Beekeeper Studio)
• Tableau Desktop
• Python
• GitHub

---

🔐 Database Access & Notes

⚠️ Some DDL operations (like CREATE TABLE) may fail when run directly due to a lack of schema privileges on the hosted Neon DB.
All SQL was written, tested, and exported using Beekeeper Studio.

🔗 Connection Info (Read-Only)

'postgres://Test:bQNxVzJL4g6u@ep-noisy-flower-846766-pooler.us-east-2.aws.neon.tech/Globox'