← StatisticsCompleted
Exploratory data analysis
This analysis explores the NordHome retail dataset — customers, orders, order items, payments and returns — to understand how spending is distributed, how customer groups differ and where returns deserve a closer look. The aim is not to prove a hypothesis but to find patterns that lead to better questions.
Python · Pandas · NumPy · Matplotlib
View the complete analysis on GitHub →01 — Project at a glance
The scope of the dataset
Counts come from the raw files. Duplicates and data-quality issues were cleaned or flagged before the analysis.
The dataset is synthetic, so differences below are exploratory signals, not real-world benchmarks.
02 — Analytical questions
What questions are we trying to answer?
The analysis covers 21 business questions across six domains. Six of them are followed in sequence below — each answer shapes the next one: from the shape of spending, to who spends, to where returns deviate.
The full notebook: 21 business questions across 6 domains
- 01 — Revenue1. Monthly revenue trend · 2. Revenue by country and channel · 3. Return/refund/cancellation impact on revenue · 4. Quarterly seasonality
- 02 — Customers1. Distribution by market and country · 2. Distribution by age group · 3. Loyalty vs. non-loyalty value · 4. Repeat vs. one-time buyer value · 5. Revenue per customer by age group
- 03 — Products1. Revenue and units by category · 2. Top 10 products by revenue · 3. Return rate and refund value by category · 4. Catalog vs. realized margin by category
- 04 — Payments1. Payment status breakdown · 2. Unpaid risk by payment method
- 05 — Returns1. Return volume and reasons by category · 2. Return rate trend year over year · 3. Return rate by channel and country
- 06 — Marketing1. Channel clicks and conversion rate · 2. Top-performing campaigns · 3. Loyalty vs. non-loyalty conversion by channel
Six of these 21 questions are walked through in detail below, in the order they were asked.
Spending is strongly right-skewed
Most orders sit in a narrow range, while a small number extend far beyond it: the median order is €658, the mean €787.
I kept the 717 flagged orders (2.5%) in — excluding them moves the average order value by only 6.3%.
→ So who is doing the spending? Next: repeat customers versus one-time buyers.
Repeat customers show a substantially different spending pattern
85.2% of customers ordered more than once. They spend €2,845 on average across 3.6 orders, versus €800 for one-time buyers — about 3.6× more per customer.
→ Repeat buying separates customers clearly. Can a simple demographic such as age do the same?
Age groups show limited differentiation
Average revenue per customer runs from €2,465 (50–59) to €2,617 (30–39) — a spread of roughly 6%. Age alone does not separate customers by spend.
→ If customer traits explain little, the focus moves to where orders come back: returns by channel and country.
Order return rates cluster around 19% across channels and countries
Across 40 channel × country combinations, order return rates run from 16.0% to 22.8% (average 19.0%). The highest cell, Marketplace in the Netherlands, rests on roughly 750 orders and may be sampling noise; a proportion test would be the next step.
→ Return rates look similar across segments. Do they at least stay stable over time?
Return rates increased in H1 2024
The item return rate held at 8.2–8.7% from 2021 to 2023, then reached 9.7%; the revenue return rate moved from about 4% to 4.5%. A signal to investigate, not a conclusion.
→ Before asking which products might drive returns, one more basic question: which products carry the revenue?
A small group of products dominates the top revenue range
Gifts account for 6 of the 10 best-selling products, including the top three. The ten sit in a tight band: the best earns about 1.6× the tenth (€144K vs €92K).
→ These deep dives are brought together in the key findings below.
03 — Key findings
What stood out in the data?
- 01
Spending is right-skewed
A few very large orders and customers stretch the average (median €658 vs mean €787). They are signals to investigate, not errors.
From deep dive 01
- 02
Repeat customers behave differently
€2,845 versus €800 per customer, about 3.6× more. The data shows that it matters, not why.
From deep dive 02
- 03
Age, channel and country barely differentiate
Age moves revenue per customer by only about 6%; return rates cluster around 19% across 40 channel × country cells.
From deep dive 03 · 04
- 04
Returns are rising; revenue concentrates in few products
The item return rate rose from 8.2–8.7% to 9.7% in H1 2024, and Gifts hold 6 of the top 10 products. Signals, not conclusions.
From deep dive 05 · 06
04 — From findings to new questions
What should we investigate next?
- 01 — High-value customers · findings 01, 02What characteristics distinguish high-value customers?
- 02 — Product performance · finding 04Why are certain products or categories concentrated among top-revenue products?
- 03 — Return behaviour · findings 03, 04What factors might explain differences in return rates?
- 04 — Customer segmentation · findings 02, 03Would behavioural segmentation reveal stronger differences between customers?
05 — Methods
How the analysis was performed
- Data preparationLoaded the raw NordHome tables and checked their structure and row counts.
- Data cleaningRemoved duplicates and flagged data-quality issues; excluded cancelled, returned and refunded orders where a chart requires it.
- Exploratory analysisDescriptive statistics across six areas — revenue, customers, products, payments, returns and marketing — through 21 business questions.
- Distribution analysisHistograms of order value and revenue per customer; median compared with mean.
- Group comparisonsRepeat versus one-time buyers, age groups and top products.
- Relationship analysisReturn rates by channel × country and over time.
- Outlier analysisIQR method (Q3 + 1.5 × IQR) to flag orders and customers outside the typical range; flagged cases were kept in the data.
06 — Conclusion
Where the investigation ends — for now
- 01
What did the EDA reveal?
Spending is skewed, and repeat customers spend about 3.6× more per customer than one-time buyers. Age, channel and country hardly separate customers, and returns edged up in H1 2024.
- 02
What did it not explain?
Why repeat customers spend more, what defines a high-value customer, and what drives the rise in returns. EDA shows patterns, not causes — and the dataset is synthetic.
- 03
Why is further analysis needed?
Each open question calls for a targeted follow-up — segmentation, product and category drill-downs, and tests of whether differences are statistically meaningful — before any business recommendation.