Sales Data Analysis
→
Why This Project?
When you use someone else’s data with little to no documentation and no one around to answer questions, you quickly learn to rely on exploration, experience, and clear logic. This project was a perfect example: the raw transaction logs contained non-transaction rows, ambiguous return and cancellation flags, and inconsistent formatting. Tackling these challenges sharpened my ability to make evidence-based decisions, apply systematic cleaning steps, and document every assumption for future users.
Data
This project analyzes the UCI Online Retail II dataset, which contains over a million transactions from a UK‐based e-commerce store spanning 2009–2011. Each record includes an invoice number, stock code, product description, quantity, unit price, customer ID, country, and transaction date. Beyond straightforward sales, the data contains canceled and returned orders, sample fees, and non‐product line items.
Methodology
Data decisions:
- Transactions were determined to be returns if the stock code started with C or the quantity was a negative value.
- When the descriptions differed for the same stock code, the most common description was assigned to that stock code.
- Transacions with a price of 0 were removed.
- Stock codes that didn't follow the standard format (##### or #####@) were evaluated on a case-by-case basis and dropped when appropriate (e.g, Bank Charges, Amazon Fees).
2 variables were calculated:
- sales: Quantity multiplied by Price.
- IsReturn: Binary indicator (1 if the transation is a return, otherwise 0)
Results
1. Data Overview
- Total transactions analyzed: 1,043,657
- Time span: Dec 2009 – Dec 2011
2. KPIs
- Total customers: 5,877
- Total orders: 46,971
- Total sales: $20,067,435.59
- Total return sales: $729,878.09
3. Key Insights
- Monthly sales trends: peak in November each year, low in February
- Return concentration: The top 5 products returned make up 38% of sales returned