Growth analytics case
This one prices the gap and designs the test, from public GA4 ecommerce data to the first variant worth shipping.
New users are 89% of visitors. Returning users prove the store can convert. The first-purchase experience is where the value is stranded.
The leverage score gap was not close: new users scored 238,902, about 8.9× the next segment. The low rate matters because it sits on top of 240,441 users, not because it makes an elegant chart.
Mobile and desktop cart-to-checkout rates were nearly identical: 44.7% for mobile and 44.3% for desktop. A mobile-only checkout redesign is too narrow as the first move.
The stronger read is structural first-purchase friction. The intervention helps a new visitor complete a first order before asking them to become a registered customer.
The dataset is bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*,
November 2020 through January 2021. Each layer can be inspected on its own.
Confirmed event names, date range, purchaser counts, and the purchase-based activation definition.
Built step-over-step conversion from session start to purchase, including checkout completion.
Created retention views that feed the Looker Studio cohort dashboard.
Ranked device, traffic medium, and new-versus-returning cuts by activation leverage.
Estimated 12,044 users per group for the +50% relative lift target.
Connected Looker Studio dashboards and generated a Claude-assisted impact memo with human review.
The conservative target is a +50% relative lift, moving new-user activation from 0.64% to 0.96%. At a $69.09 average order value, that is 769 incremental purchasers and $53,130 in revenue.
No changes. New users continue through the existing product and checkout path.
Select to model in the sizer
Make guest checkout the default for new users, remove account creation before purchase, and offer account creation after the order is complete.
Select to model in the sizer
After a new user views two or more products, show a free-shipping nudge for the first order. This waits until Variant B establishes the friction baseline.
Select to model in the sizer
Primary: new-user activation rate. Secondary: add-to-cart rate. Guardrails: revenue per activated user and account creation rate.
The artifacts show the decision path: public data to dashboard, dashboard to memo, memo to intervention, intervention to test plan.
Public Looker Studio dashboards connect to BigQuery views for funnel steps and cohort retention. They are linked rather than embedded to keep the portfolio fast.
Open dashboardsA polished PDF memo with clean Markdown source. It compresses the case into what a hiring manager can read in one sitting: what broke, why new users matter, what to test first, and how the $53,130 opportunity was calculated.
AI-generated, human-reviewed. The full one-page PDF covers seven sections: problem, data, segment, proposed intervention, impact estimate, implementation scope, and risks.
Technical enough to audit, but the structure points back to the business question: where should the team intervene first, and how would we know it worked?
sql/ |
Schema audit, funnel, cohort, segmentation, and Looker-ready views. |
|---|---|
notebooks/ |
Segment leverage ranking and sample-size calculations. |
findings/ |
Activation definition and the primary new-user drop-off finding. |
test_plan/ |
A/B test plan with variants, metrics, power, duration, and rollout rules. |
figma/ |
Variant B and Variant C intervention mockups. |
memo/ |
One-page impact memo in Markdown and PDF, plus a reproducible PDF build script. |
validate.py |
Environment, BigQuery view, and memo-output validation checks. |
This case runs the full loop: define activation, identify leverage, rule out tempting but weak explanations, size the business impact, choose the first test, and explain the risk in plain English.
Closing read
The recommendation is not "conversion is low." It is: new users are the highest-leverage segment, the first move is guest checkout, the upside is priced conservatively, and the test has enough sample to read.