Customer Shopping Behavior Analysis

Two Streamlit dashboards built on a 3,900-row retail order file. One looks at what sells. The other looks at what the discount programme is costing.

The repo also carries the cleaning notebook and the SQL used for the same questions against a database copy of the table.

---

Overview

The dataset records one apparel order per row: who bought, what they bought, what they paid, how they paid, whether a discount was applied, and the rating they left afterwards. There is no cost or margin column, so every money figure here is order value rather than profit. That limit shapes what the second dashboard can honestly claim, and it is called out on the page itself.

Work order was: clean the file in pandas, answer the business questions in SQL, then put the two questions worth watching over time into an app.

What is in the repo

File	What it does
app.py	The Streamlit app. Both dashboards live here.
customer_shopping_behavior.csv	Raw order export, 3,900 rows, 18 columns.
Customer_Shopping_Behavior_Analysis.ipynb	Cleaning and profiling steps in pandas.
customer_behavior_sql_queries.sql	Ten business questions answered in SQL.
requirements.txt	The three packages the app needs.
README.md	This file.

The two dashboards

1. Product and Category Performance

Answers where the revenue comes from and whether the products earning it are the ones customers like. Top twelve products by revenue, a category split with share labels, the seasonal mix, and a scatter of average rating against revenue so an item that sells well but rates poorly stands out. A sortable product table sits underneath for anyone who wants the numbers rather than the picture.

2. Discount Impact

Answers whether the discount is buying anything. It compares average order value with and without a discount, ranks products by how often they go out at a reduced price, breaks the discount rate down by subscription status, loyalty band and age group, and closes with a short list of products that are both heavily discounted and rated below the store median.

Both tabs share one set of sidebar filters: season, category, gender and age group. Change a filter and every number on both pages moves with it.

Running it

git clone https://github.com/Chetan04040/customer-analysis.git
cd customer-analysis

pip install -r requirements.txt
streamlit run app.py

The app opens at http://localhost:8501. Keep customer_shopping_behavior.csv next to app.py; the script looks for it there rather than in whatever folder you launched from.

Streamlit 1.49 or newer is needed for the width="stretch" chart argument. On an older version, swap those for use_container_width=True.

To reproduce the SQL side, load the cleaned frame into Postgres or MySQL with the to_sql cell at the end of the notebook, then run customer_behavior_sql_queries.sql against the resulting customer table.

What the numbers say

Revenue is $233,081 across 3,900 orders, so the average order sits at $59.76.

Clothing takes 44.7% of revenue and Accessories another 31.8%. Outerwear is the tail at 7.9%. No single product dominates: the top five items together account for 22% of revenue, which is what you would expect from a catalogue where every item is priced in the same $20 to $100 band.

The discount finding is the more interesting one. 43% of orders carry a discount, and those orders average $59.28 against $60.13 for full-price orders. The discount is not buying a bigger basket. It is about 85 cents smaller.

Splitting by segment shows why the rate is so high. Every single subscriber order is discounted, all 1,053 of them. Among non-subscribers the rate is 21.9%. So the headline 43% is really a subscription perk applied automatically plus a much smaller set of genuine promotions. Subscribers are 27.0% of orders and 26.9% of revenue, and they spend slightly less per order than everyone else, which makes the perk hard to justify on order value alone.

Ratings sit in a narrow band, 3.62 to 3.86 out of 5, with Gloves and Sandals at the top and Shirt and Jeans at the bottom. Nothing is broken, but the products in the watch list at the bottom of the second tab go out at a discount roughly half the time while rating below the store median. Those are worth a pricing look before the next campaign.

Notes on the data

• 37 review ratings were blank. Each is filled with the median rating for its category so the rating charts keep a full row count.
• Promo Code Used matched Discount Applied on every single row, so it is dropped during load.
• Column names are lowercased and snake_cased to match the SQL.
• Age groups are quartiles of the age column: 18-31, 32-44, 45-57, 58-70.
• Loyalty bands come from previous_purchases: 1 is New, 2 to 10 Returning, above 10 Loyal.
• The file has no dates, no cost of goods and no discount amount. Anything presented as a trend over time or as margin would be invented, so neither appears.

Licence

MIT.