My contribution
Used Python to examine customer groups and SQL to reconcile the reported totals.
← All work
Customer Segmentation & BI · Independent project
Customer segmentation & spending concentration
I studied purchases from 440 wholesale customers to find who spent the most and how their buying patterns differed. I checked the totals and suggested questions for future marketing tests.
440customer records
42.9%of spend from the top 20%
2purchasing mix segments
UCI wholesale customers / Annual spending
42.9%of recorded spending came from
the highest spending 20% of customers.
88 of 440 records · Historical public data
Project overview
Independent analysis of historical wholesale customer purchasing data.
Methods
Segmentation, data validation, and analytical storytelling.
Purpose
Helps a team define a more focused customer question and a testable next step.
Terms explained
- Customer segmentation
- Grouping customers by a shared characteristic, such as the types of products they buy.
- SQL and Python
- SQL works with records in a database. Python is a programming language used here to study patterns and compare groups.
- Dashboard
- A report with charts and numbers that can be explored in one place.
01 · The business question
How do customers differ in what they buy, and where is annual spending concentrated?
02 · My contribution
- Used Python, a programming language for analyzing data, to group customers by the types of products they bought. Compared five settings for a grouping method called K means.
- Used SQL, a language for working with database records, to check that customer totals and product category totals matched.
- Built an interactive report in Plotly to show spending patterns and suggest marketing tests for different customer groups.
03 · The finding
The 88 customers who spent the most made up 20% of the records and accounted for 42.9% of annual spending. This tells us where spending was concentrated. Grouping customers by what they bought answers a different question.
04 · The decision
Try different marketing messages for groups with different buying patterns. Compare each message with a standard version before deciding whether it changed customer behavior.
Learning notes
Two useful views of the same customer.
- Purchasing mix and spending concentration answer different questions. A large customer is not automatically a distinct behavioral segment.
- Reconcile category totals and customer totals before presenting a percentage. The denominator is part of the finding.
A next questionUse the purchasing mix groups to frame a message test, then compare responses against a defined control.
05 · Work delivered
- Customer segmentation analysis
- SQLite reconciliation
- Plotly dashboard and proposed tests
Explore the public data
Who accounts for the spending?
Choose a customer type or region to compare how much customers spent and what they bought.
Customer records440
Total annual spending14,619,500
Top 88 customers’ share42.9%
Annual spending by product category
Recorded monetary units (m.u.). The source does not identify a currency.
| Category | Relative size | Spending (m.u.) |
|---|---|---|
| Fresh | 5,280,131 | |
| Milk | 2,550,357 | |
| Grocery | 3,498,562 | |
| Frozen | 1,351,650 | |
| Detergents & paper | 1,267,857 | |
| Delicatessen | 670,943 |
Ranks customers by total spending across six categories within the selected group. The customer count is rounded up when the selected percentage is not a whole number. Customer type and region come from the original records. They are not groups predicted by the analysis.
Source: Margarida Cardoso, Wholesale customers (2013), UCI Machine Learning Repository · DOI: 10.24432/C5030X · CC BY 4.0. Source records are unchanged. Filters and aggregate views were added for this portfolio.
See the method up close
Related methods.
Working examples with their data and assumptions.
Explore related fields
Follow the methods across other projects and working examples.
Context & measurement notes
An independent study of historical public data. The spending result describes these records. It is not business growth or a prediction of future purchases. The interactive report below groups records using the original customer and region labels.