This project is about analyzing a dataset provided by an e-commerce marketplace called Olist to answer the CEO’s question:
How to increase customer satisfaction (so as to increase profit margin) while maintaining a healthy order volume?
This project demonstrates skills about:
- Data sourcing from SQL/CSV files using pandas
- Feature engineering using pandas
- Data visualization & Decision Science using pyplot, seaborn and plotly
Olist is a leading e-commerce service that connects merchants to main marketplaces in Brazil. They provide a wide range of offers including inventory management, dealing with reviews and customer contacts to logistic services.
Olist charges sellers a monthly fee. This fee is progressive with the volume of orders.
Here are the seller and customer workflows:
Seller:
- Seller joins Olist
- Seller uploads products catalogue
- Seller gets notified when a product is sold
- Seller hands over an item to the logistic carrier
Customer:
- Browses products on the marketplace
- Purchases products from Olist.store
- Gets an expected date for delivery
- Receives the order
- Leaves a review about the order
Clone the project and install it:
git clone git@github.com:aduverger/olist.git
cd olist
pip install -r requirements.txt
make clean install test # install and testThe dataset consists of 100k orders from 2016 and 2018 that were made on the Olist store, available as a csv on Kaggle.
Download the 9 datasets and store them in your data/csv folder.
The data is provided through csv files yet the structure is similar to SQL. You can refer to this schema for details:
You just have to run the notebook that is inside the notebook folder to see how we used data science as a tool to identify bad sellers from the Olist store.
Alternatively, you can directly look at the presentation made with Jupyter.
Here are some examples of the insights that have been made using seaborn and plotly:
Why we should remove underperforming sellers:
Old or new friends ?
Slow delivery ?
Sellers behaviour pattern:
