Define the grain before the chart
Separate event facts, order facts and identifiable users. A transaction with multiple item events counts once as an order. Missing session IDs in Retailrocket mean a visitor funnel, not a session funnel.
Data engineering / Documented data analysis
Product Analytics Warehouse
A reproducible event pipeline connects local validation, BigQuery models and dashboards for purchase funnels and return visits.
01 / THE PROBLEM
Raw events do not explain where customers leave or whether they return. Reliable analysis needs explicit grains, deduplication rules and metric definitions before a dashboard can be trusted.
02 / FOLLOW THE WORKFLOW
A simplified example of the documented flow. Sample outputs are authored illustrations, not live model predictions.
Documented dataset count
2,756,101 source event rows
Start from anonymized Retailrocket activity.
9 BigQuery SQL views
Events → orders → visitors
Create consistent facts and dimensions.
Visitor-based, across the observation period
View → Add to cart → Transaction
Count visitors in timestamp order.
Observed association, not a causal result
97.65% did not reach cart after viewing
Explain the largest observed drop.
Step 1 of 4 · Events
03 / ENGINEERING
Separate event facts, order facts and identifiable users. A transaction with multiple item events counts once as an order. Missing session IDs in Retailrocket mean a visitor funnel, not a session funnel.
The local standard-library Python pipeline validates inputs and upserts stable event identities. The synthetic fixture is clearly separated from authentic Retailrocket and GA4 analysis.
The same Retailrocket rows were loaded into BigQuery. Nine SQL views implement facts, dimensions, daily activity, week-one retention and the ordered visitor funnel; documented results agree with the local run.
04 / EVIDENCE & LIMITS
The source spans 3 May–18 September 2015 UTC and contains 17,672 distinct transaction IDs. Of 1,404,179 visitors who viewed an item, 32,934 subsequently added to cart. A complete-week cohort starting 4 May had 3,425 of 71,389 visitors return in the following calendar week (4.80%). These are dataset findings, not business improvements caused by the project.
Retailrocket does not provide checkout events, session IDs, prices, revenue or acquisition channels. None are inferred. The funnel can span products and sessions across the observation period. BigQuery sandbox tables may expire; checked-in reports preserve the evidence.
05 / READ THE SOURCE
Based on public repository documentation reviewed on 7 November 2026. Results are attributed to the project’s recorded evaluations; they were not rerun for this portfolio.
TOOLS BEHIND THE WORK
Explore the stack. Drag the ribbon or select a tool.