Olist Commerce Analytics
Decision-oriented analysis of Brazil's Olist e-commerce dataset covering 96k orders, analyzed for commercial performance and customer experience recommendations.
2026-06-12
Concept
The public Brazilian E-Commerce dataset by Olist is a rich analytics case: orders, items, payments, reviews, and customers. This project builds decision-oriented analysis; not just visualizations, but answers to concrete business questions.
Business Question
Where should an e-commerce operator focus to improve commercial performance and customer experience: category mix, geographic coverage, repeat purchasing, or delivery reliability?
Approach
- Analytical mart: one row per order as the foundation; one-to-many tables aggregated before joining to prevent KPI inflation.
- Metric discipline: commercial KPIs use delivered orders only; partial months excluded from trends.
- Data contracts: metric ranges tested with pytest to protect quality.
Key Findings
- 96,478 delivered orders generated R$13.22M in item revenue.
- 8.1% of orders were late; late orders averaged 2.57/5 stars vs 4.29/5 on time.
- Only 3.0% of customers purchased twice, yet drove 6.1% of orders.
- São Paulo contributed 38.3% of revenue; health_beauty led categories at 9.3%.
Reflection
The primary recommendation: prioritize delivery-exception monitoring, then build category-aware repeat-purchase cohorts. Analytics quality comes from data discipline, not chart count.