Problem
The portfolio scenario asks whether a shorter checkout flow might affect conversion. It is a simulated business question, not work commissioned by or performed for an actual retailer.
Data
The repository contains 12,000 session records split evenly between control and treatment groups. The notebook describes the input as simulated Australian retail data. Fields used in the analysis include conversion, session revenue, device type, state, and traffic source.
Approach
I summarized conversion and revenue by test group, used a two-proportion z-test to compare conversion rates, and explored results by device type, state, and traffic source. The repository includes a notebook, a Python script, summary files, and visuals.
Technology
Python, pandas, NumPy, SciPy, Matplotlib, and Seaborn.
Solution
The repository contains a notebook and Python script for the comparison, along with saved group and segment summaries and supporting visuals.
Result within the simulated sample
The saved summary reports a 9.18% conversion rate for the control group (551 of 6,000 sessions) and 11.18% for the treatment group (671 of 6,000). The repository reports a two-proportion test p-value of approximately 0.0003 for these sample results.
These figures are outputs from simulated data. They do not show that a real checkout change increased conversion, establish a forecast, or support a real-world rollout.
Methodological learning
A small p-value in simulated data is not evidence of real business impact. A real experiment would require verified data provenance, valid random assignment, an appropriate analysis plan, and business assumptions supported by actual traffic data.