Coupon Usage Analytics Project

End‑to‑end predictive analytics for coupon redemption in Nigerian retail

Phase 1: Data Analysis & Prompt Engineering

In Phase 1, I focused on loading, cleaning, and exploring Nigerian retail and e‑commerce coupon usage data.

• Tools used: Excel, MySQL, Python (Pandas, Matplotlib, Seaborn), AI prompt engineering.
• Actions:
  • Loaded the dataset into Python using Pandas, with initial exploration in Excel and MySQL.
  • Cleaned missing values and standardized column formats.
  • Explored coupon redemption trends by campaign, channel, discount type, and festive periods.
  • Visualized redemption rates and customer behavior patterns.
• Insights:
  • Redemption rates were higher for percentage discounts compared to fixed value coupons.
  • Campaign timing (month, weekday) influenced redemption likelihood.
  • Festive periods showed uplift in coupon usage.
• Outcome: Established a strong analytical foundation and identified variables likely to influence coupon redemption, setting the stage for predictive modeling. 👉 Full Phase 1 repository: Coupon Usage Exploratory Analysis

---

Phase 2: Predictive Modeling & Power BI Insights

🔹 Machine Learning Models

Implemented three models in Python:

• Logistic Regression
• Decision Tree
• Random Forest

🔹 Model Performance

The models were evaluated using standard classification metrics:

• Accuracy: 0.698
• Precision: 0.698
• Recall: 1.000
• F1 Score: 0.822

🔹 Performance Evaluation

• The model achieves perfect recall, meaning it successfully identified all positive (redeemed) cases without missing any.
• Precision and accuracy are moderate (~70%), indicating that while the model captures positives well, there is some misclassification of negatives.
• The F1 score (0.822) provides a balanced measure of precision and recall, showing strong overall performance and reliability for coupon redemption prediction.

🔹 Business Implication

• High recall ensures that no potential redeemer is overlooked, which is critical for maximizing campaign reach and customer engagement.
• The trade‑off in precision means some customers may be targeted who ultimately don’t redeem, but this is acceptable in marketing contexts where the cost of outreach is lower than the benefit of capturing all redeemers.
• Overall, the model supports aggressive targeting strategies, helping retailers increase redemption rates while providing a strong foundation for optimizing marketing spend.

---

🔹 Exported Outputs

• predictions.csv → actual, predicted, predicted probability, plus campaign/channel features.
• feature_importance.csv → Random Forest feature importance scores.

🔹 Power BI AI Visuals

Integrated outputs into Power BI:

• KPI cards for model metrics
• Feature Importance bar chart
• Key Influencers visual (AI-driven explanations)
• Predictions table with slicers for campaign, channel, and confidence bands

Validation: Power BI AI visuals confirmed Python insights — discount value, campaign type, and channel were the strongest drivers of coupon redemption.

---

📊 Dashboard Screenshots

• Model Performance KPIs
  
• Feature Importance Chart
  
• Key Influencers Visual
  
• Predictions Table

---

🚀 Impact

This end‑to‑end workflow demonstrates:

• Data cleaning and exploration (Excel, MySQL, Python)
• Predictive modeling (Logistic Regression, Decision Tree, Random Forest)
• AI‑assisted storytelling (Prompt engineering)
• Business intelligence visualization (Power BI)

Business Value: Delivered actionable insights for Nigerian retail and e‑commerce businesses, helping optimize coupon targeting, reduce marketing waste, and improve customer engagement.