📊 SuperStore Sales & Profitability Analysis

An end-to-end exploratory data analysis of ~9,800 orders from a US retail superstore (2015–2018), going beyond a revenue-only view to find where the business is actually profitable — and where it isn't. Includes a Jupyter notebook, a reusable analysis script, and a live interactive Streamlit dashboard.

▶ Live dashboard  •  📓 Notebook  •  📄 One-page summary 📊 View on Kaggle

---

Table of Contents

• Project Purpose
• Key Findings
• Interactive Dashboard
• Repository Structure
• Tech Stack
• How to Run
• Data & Methodology
• Further Improvements

---

📌 Project Purpose

Most public "Superstore" EDA projects stop at "which category sells the most?". This one goes a step further and asks the questions a business stakeholder actually needs answered:

• Which customer segments and individuals generate the most revenue and lifetime value — not just the most orders?
• Which product categories are genuinely profitable, and which are quietly losing money?
• Does discounting help or hurt the bottom line, and at what threshold does it flip?
• Which regions look strong on sales but are actually unprofitable?
• How has the business trended year over year, and what's the seasonal pattern?

---

💡 Key Findings

Metric	Value
Total Sales	$2.26M
Total Profit	$278,979
Overall Profit Margin	12.3%
Orders / Unique Customers	4,922 / 793
Repeat Customer Rate	98.4%

1. Furniture is a volume trap. It's the #2 category by revenue but dead last by profit — Tables (-9.0% margin) and Bookcases (-3.0% margin) are outright loss-making.

2. Discounts above ~20% destroy profit. Average profit per order goes negative once discount exceeds 0.2, and the 30-50% band loses ~$157 per order on average.

3. Texas, Ohio, Pennsylvania and Illinois are profitability blind spots — all top-10 states by sales, all net-negative on profit.

4. Growth resumed strongly after a 2016 dip: sales fell 4.3% in 2016, then grew 30.6% (2017) and 20.3% (2018). A SARIMA forecast projects the next 6 months, including the seasonal Jan/Feb dip seen every prior year.

5. Customer segmentation & CLTV: Consumers are the largest segment by headcount and revenue, but Corporate customers carry the highest average lifetime value per customer.

Full workings, RFM customer tiering, shipping-mode analysis, and the discount/geography/time-series breakdowns are in the notebook.

---

🖥️ Interactive Dashboard

A Streamlit dashboard lets you filter by date range, segment, category and region, with live KPIs, profitability breakdowns, a US profit/sales choropleth, and time-trend charts.

streamlit run dashboard/app.py

Or run it as a container, no local Python setup needed:

docker run -p 8501:8501 ghcr.io/subhranshupan/super-store-data-analysis-project:latest

(Published automatically by .github/workflows/release.yml on every version tag - see Packages.)

Live demo: deployed for free on Streamlit Community Cloud - link added here once deployed (see the Deployment step in Further Improvements below).

---

📁 Repository Structure

.
├── .github/workflows/
│   ├── ci.yml                     # lint + test + smoke-run on every push/PR
│   └── release.yml                # on version tag: publish wheel/sdist + Docker image
├── .streamlit/
│   └── config.toml                # dashboard theme, used on Streamlit Cloud too
├── superstore/                    # shared package - single source of truth
│   ├── data.py                    # load_data / clean_data
│   └── forecast.py                # forecast_monthly_sales (SARIMA)
├── data/
│   ├── superstore_sales.csv       # cleaned, enriched dataset
│   └── README.md                  # data dictionary + enrichment methodology
├── notebooks/
│   └── SuperStore_Sales_Analysis.ipynb
├── scripts/
│   └── run_analysis.py            # computes metrics + saves charts, using superstore/
├── dashboard/
│   └── app.py                     # Streamlit dashboard (Customers/Products/Geo/Trends/Forecast)
├── tests/
│   └── test_data.py, test_forecast.py, conftest.py   # pytest suite (9 tests)
├── reports/
│   ├── metrics.json               # all computed KPIs, machine-readable
│   ├── rfm_customer_segments.csv  # per-customer RFM scores/tiers
│   ├── subcategory_profitability.csv
│   └── Executive_Summary.pdf      # one-page recruiter-friendly summary
├── assets/images/                 # chart exports used in this README
├── Dockerfile / .dockerignore     # containerized dashboard, published to GHCR on release
├── CHANGELOG.md
├── requirements.txt
├── requirements-dev.txt           # + pytest, ruff, build
├── pyproject.toml                 # packaging + pytest/ruff config
└── LICENSE

---

⚙️ Tech Stack

• Python (Pandas, NumPy) — data cleaning and analysis
• Matplotlib / Seaborn — static charts
• Plotly — interactive choropleth, sunburst and treemap visualizations
• Statsmodels (SARIMA) — monthly sales forecasting
• Streamlit — interactive dashboard
• Jupyter Notebook — primary analysis narrative
• pytest + ruff — automated tests and linting
• GitHub Actions — CI on every push/PR, releases (PyPI-style package + Docker image) on every version tag
• Docker / GitHub Container Registry (GHCR) — the dashboard ships as a container

---

🚀 How to Run

git clone https://github.com/SubhranshuPan/Super-Store-Data-Analysis-Project.git
cd Super-Store-Data-Analysis-Project
python -m venv .venv && source .venv/bin/activate   # optional but recommended
pip install -r requirements.txt          # or requirements-dev.txt to also get pytest/ruff

# Reproduce all metrics + chart images (incl. the forecast chart)
python scripts/run_analysis.py

# Explore the full narrative notebook
jupyter notebook notebooks/SuperStore_Sales_Analysis.ipynb

# Launch the interactive dashboard (Customers / Products / Geography / Trends / Forecast)
streamlit run dashboard/app.py

# Run the test suite / linter
pytest
ruff check .

---

🗂️ Data & Methodology

The base dataset ships with order, customer, product and geography attributes plus Sales, but not Profit, Discount or Quantity. Those three fields were sourced from the public "Sample – Superstore" reference dataset and merged in on Row ID, after verifying an exact match on Product ID/Customer Name/Sales for every row. Full details in data/README.md.

---

🔭 Further Improvements

• Sales forecasting: a SARIMA model forecasts sales 3-12 months ahead with an 80% confidence band — see the notebook's Forecast section or the dashboard's Forecast tab.
• Automated tests: tests/ (pytest) covers the cleaning and forecasting logic in superstore/ — 9 tests, run with pytest.
• CI: .github/workflows/ci.yml lints with ruff, runs the test suite, and smoke-runs scripts/run_analysis.py on every push/PR to main.
• Shared module: cleaning/loading logic lives once in superstore/data.py and is imported by the notebook, scripts/run_analysis.py, and dashboard/app.py — no more duplicated logic across the three.
• Packaging & releases: superstore builds as an installable wheel/sdist (pyproject.toml), and .github/workflows/release.yml publishes both the Python package (attached to the GitHub Release) and a Docker image of the dashboard (to GHCR, under this repo's Packages tab) on every vX.Y.Z tag. See CHANGELOG.md for the v1.0.0 notes.
• Deploy the dashboard publicly: the app, Docker image and .streamlit/config.toml theme are all ready — once this repo is pushed to GitHub, deploy for free at share.streamlit.io pointing at dashboard/app.py on main, and add the live link at the top of this README. This is the one step that needs a GitHub push + your own Streamlit account, so it's left for you to click through.

---

📬 Contact

Subhranshu Panda — GitHub · open to Data Analyst / Data Science part-time and internship roles.