Candle35/zando_ab_testing

End-to-end A/B testing analysis for Zando Ghana — data cleaning, EDA, statistical hypothesis testing, and business recommendation using Python. # Zando Ghana — A/B Testing & Statistical Analysis A full end-to-end A/B testing project analysing whether a redesigned product listing page increased purchase conversion rates for Zando Ghana, an online fashion and lifestyle retailer. --- ## Business Problem Zando Ghana's UX team redesigned their product listing page with larger images, a Quick View panel, and a more prominent Buy Now button. Before rolling it out permanently, they needed data-driven evidence that the new design actually increased the number of customers completing a purchase — and that the result was not due to chance. --- ## Dataset | Property | Details | |---|---| | Records | 4,000 sessions | | Period | September 1–30, 2024 | | Columns | 16 (session ID, user ID, date, device, group, purchase made, order value, pages visited, time on site, and more) | | Groups | Control — 2,038 sessions (old page) / Treatment — 1,962 sessions (new page) | --- ## What I Did ### Phase 1 — Data Audit Inspected the dataset for structure, data types, missing values, duplicates, and group balance before any cleaning or analysis. ### Phase 2 — Data Cleaning Resolved all data quality issues found: - Standardised inconsistent capitalisation in the `device` column - Converted `date` from string to datetime format - Filled missing values in `age_group` (120 rows) and `city` (80 rows) with their respective mode values - Converted `items_purchased` and `coupon_used` from float to integer - Identified that nulls in `order_value_ghs`, `items_purchased`, and `coupon_used` were correct by design — these fields are only populated when a purchase is completed, so non-purchasing sessions are intentionally empty ### Phase 3 — Exploratory Data Analysis Produced five visualisations with written business interpretations covering: - Experiment group size balance - Purchase conversion rate by group - Distribution of time on site (revealing non-normal skew) - Gender composition by group - Conversion rate by device type ### Pha …