01 / Retail analytics
A public transaction dataset, an explicit simulated extension, and a reproducible route from data preparation to business measures.
Historical data from one UK non-store retailer, December 2010 to December 2011. The final December contains nine days.
Start here Try switching Evidence scope to distinguish observed transactions from the simulated business case.
Loading verified historical aggregates…
Returns retain their negative value. Source lines with non-positive prices or zero quantity are excluded under the documented cleaning rules.
Customers purchasing in both months divided by the previous month's identified purchasing customers. Anonymous sales are excluded.
The chart shows eight products from the scope's full-period top-twenty set, ranked by selected-period revenue. It is not a newly calculated top twenty for every filter.
Store, category, cost, campaign and web-session fields were generated with a fixed seed. They demonstrate analytical design; they are not evidence of actual omnichannel operations.
Fixed full-period combined scope: 1 December 2010 through 9 December 2011. The filters above do not change these tables. Margin uses simulated costs; attribution does not establish campaign lift.
Recency, frequency and monetary segmentation on identified customers in the combined data. A segment's monetary share uses the segment-monetary denominator.
The work connects a governed input, deterministic transformations, explicit metric definitions and a Power BI reporting specification.
The project examines revenue, repeat purchasing, retention and product performance. Simulated costs and channels support a separate business-case exercise. It does not claim measured savings, causal marketing lift or predictive customer lifetime value.
The pipeline retains returns, removes exact duplicate source rows under an explicit rule, quarantines invalid-price and zero-quantity lines, and keeps anonymous revenue separate from identified-customer measures. With no original line identifier, exact-duplicate removal can also remove a legitimate repeated item.
The October portfolio check ran the copied Python pipeline against the checksum-matched original source. The sales fact, ten selected tables and browser data matched the retained baseline byte for byte. This did not rerun PostgreSQL, native Power BI or a fresh dependency installation.
The original native reporting artifacts retain their historical acceptance. A corrected website does not renew native Power BI acceptance.
The source package contains code, definitions, tests and the browser aggregates. Obtain the original UCI workbook using its recorded source and checksum.
Python package and static reports. Full replay needs the original UCI workbook.
Python for replay. A PDF viewer for retained reports; SQL/DAX definitions are included.
Original UCI workbook required; small synthetic smoke fixture included.
Portable Python pipeline, SQL/DAX, locked environment, selected tables and verification instructions. Includes a small synthetic smoke fixture.
Download source ZIPZIP · 112 KBThe retained Power BI report as a PDF. Static report; the browser explorer above is a separate implementation.
Open report PDFThe retained decision memo, with assumptions and proposed actions. Suggested actions are not achieved outcomes.
Open memo PDF