Capstone project for Statistics for Data Work

Did the new checkout really work?

After the one-page checkout shipped, Mangrove ran a 16-day test of express checkout for returning customers (one tap to reorder with the saved delivery slot and payment), from Thursday 1 October to Friday 16 October 2026, with customers split 50/50. The results: control 40,210 customers and 4,825 orders (12.0% conversion), express checkout 40,530 customers and 5,147 orders (12.7%). By platform: Android control 24,100 customers and 2,795 orders, Android new 24,350 and 3,040; iOS control 16,110 and 2,030, iOS new 16,180 and 2,107. Guardrails: order cancellations 3.1% against 3.2%, average basket $18.40 against $18.25, on-time delivery 91% in both. The team looked at the dashboard every day, and the period includes a month-end salary weekend. Check how the test was run, estimate the effect with its uncertainty, review the guardrails and the segments with care, and write the recommendation Imran will take to the leadership meeting. Build a small simulated dataset that matches these totals if you want to work from rows.

What you hand in

A notebook or spreadsheet with the analysis: the checks, the effect with its interval, the guardrails and the segment view
A list of checks on how the test was run (sample ratio, dates and whole weeks, peeking, the salary weekend), each with its result
A one-page result for leadership: the effect with its confidence interval, what it means in orders a day, the guardrails, the decision and what to test next

Check your work

Tick each item when your project meets it. Be strict: this is the rubric the reviewer uses.

Sample ratio is checked with a test (the split is close enough to 50/50) and the run is reviewed: 16 days that are not whole weeks, daily peeking and the salary weekend
The effect is reported with uncertainty: about +0.7 points (12.0% to 12.7%, about +5.8% relative) with a 95% interval of roughly +0.25 to +1.15 points, not only a p-value
The arithmetic is right and shown, in a spreadsheet or Python, so a reviewer can rerun it
Guardrails are reviewed with their size and uncertainty, and the small cancellation and basket changes are judged against the gain
The platform split is treated with caution: the Android and iOS effects are compared with their own intervals and not presented as a discovery after looking at many slices
The recommendation matches the evidence, translates the effect into orders and money, and names a follow-up (a holdout or a check after launch)

0 of 6 checked

Model answer

The model answer opens once your capstone passes the review, so you can compare your work with an outline of a strong submission.

Hand in your capstone

Your links are checked automatically, then the work is scored against the rubric above, item by item, with comments. A pass counts toward the course certificate and shows on your profile.

Loading…