All analytics work
Analytics project, 2024
Brazilian e-commerce, end to end
Sales patterns, customer behaviour and logistics efficiency pulled out of a full marketplace dataset.
A marketplace dataset with orders, reviews, payments, sellers and geolocation is one of the few public sets rich enough to ask operational questions rather than toy ones.
The work joined those tables into one grain and then looked at the parts that actually move a marketplace: delivery time against review score, seller concentration, category seasonality, and where the logistics network adds days it does not need to.
The most useful finding was the least surprising one. Late delivery predicts a bad review far more strongly than price or category, which makes logistics the cheapest lever on satisfaction.
Results
- Multi-table join across orders, reviews, payments and geolocation
- Delivery lateness isolated as the dominant driver of review score
- Category seasonality and seller concentration mapped