Inspect
Check grain, repeat activity, dates and duplicate rows.
↗← ALL WORKThe right answer
RESEARCH NOTE / EXPLORATORY DATA ANALYSIS / 07
The right answer
was a better question.
An exploratory analysis that checked the dataset before choosing a model.
EXPLORE THE STORY ↓1 : 1customer-to-transaction ratio
99,457 transaction records
01
THE BUSINESS PROBLEM
Can this data explain repeat purchasing?
RFM needs variation in purchase frequency. This dataset had one transaction per customer.
CUSTOMER 01●→1 purchase
CUSTOMER 02●→1 purchase
CUSTOMER 03●→1 purchase
CUSTOMER 04●→1 purchase
Illustration of the observed structure: every customer has exactly one transaction. Frequency cannot separate repeat-purchase segments.
02From question
WHAT I DID
From question
to useful work.
Preserve
Keep valid transactions; exclude 119 missing ages only from demographic analysis.
↗Redirect
Flag the partial period and choose H&M for repeat-behaviour research.
↗03A modelling dead end
THE RESULTS
A modelling dead end
became a useful decision.
Repeat-purchase RFM was unsuitable. The analysis prevented a segmentation story the data could not support.
99,457
Transactions inspected.
0
Duplicate rows found.
1 purchase
Per customer. No frequency variation.
THE BUSINESS TAKEAWAY
Match the business question to the data’s grain before building a model.
↗THE EVIDENCE / GO DEEPER
LET’S DISCUSS YOUR PROJECT ↗Methods, context & full project notes ＋
The business question
Could this dataset support customer segmentation based on repeat purchasing, alongside a view of category and time-based shopping patterns?
The finding
Every customer had exactly one transaction. Frequency was therefore constant, making meaningful repeat-purchase RFM segmentation unsuitable for this dataset. This informed the choice of H&M for the repeat-behaviour and forecasting research.
Data preparation
Converted invoice dates, checked prices and quantities, and separated demographic missingness from transaction retention. The 119 missing ages were excluded from demographic analysis rather than causing valid transactions to be dropped. The partial March 2023 period was flagged before forecasting.
Why it matters
A technically valid model can still answer the wrong question. Inspecting grain and repeat activity first avoided presenting a segmentation result the data could not support.
Source: executed Istanbul research notebook. Academic analysis of the public Customer Shopping Dataset; findings are descriptive.
Alpha Global ↗
YOUR NEXT MOVE