A/B Testing Lab - Marketing Science Dashboard

An interactive, end-to-end A/B testing framework with Streamlit
Un dashboard interactivo de A/B Testing para Marketing Science

---

Dashboard Pages

Page	What it does
Dataset	Load synthetic or your own CSV data, EDA, timelines
Experiment Design	Sample size calculator, MDE sensitivity, timeline estimator
SRM Validation	Sample Ratio Mismatch detection (Chi-Square test)
Statistical Analysis	Frequentist Z-test + Bayesian Beta-Binomial + Segment analysis
Business Report	Executive summary, revenue projections, downloadable results

---

English

What is A/B Testing?

A/B testing is the practice of running controlled experiments to measure the causal impact of a change - a new landing page, button color, email subject line, pricing strategy, etc. Instead of guessing what works, you let the data decide.

This dashboard implements a complete, production-grade A/B testing workflow:

Define experiment -> Calculate sample size -> Run experiment  
Validate integrity (SRM) -> Statistical test -> Business decision

Features

• Synthetic data generator - realistic e-commerce dataset (LATAM market, 45K users, 14 days) out of the box
• Upload your own CSV - plug in any binary A/B test dataset
• Sample size calculator - with power/alpha controls and sensitivity curves
• SRM Check - detects broken experiment assignment before you even look at results
• Frequentist Z-test - confidence intervals, p-values, achieved power, peeking warnings
• Bayesian analysis - posterior distributions, P(B>A), expected loss framework
• Segment analysis - forest plots by device, country, or any categorical column
• Revenue impact projection - monthly & annual uplift based on your traffic and AOV
• Export results - download full report as CSV

Quick Start

# Clone
git clone https://github.com/JulianDataScienceExplorerV2/Marketing-AB-Testing.git
cd Marketing-AB-Testing

# Install dependencies
pip install -r requirements.txt

# Run the app
streamlit run app.py

Then open http://localhost:8501 in your browser.

Project Structure

Marketing-AB-Testing/
├── app.py                      # Main Streamlit dashboard (5 pages)
├── src/
│   ├── statistics.py           # Sample size, SRM, Z-test, Bayesian
│   └── data_generator.py       # Synthetic e-commerce data
├── data/                       # Drop your CSV files here
└── requirements.txt

Tech Stack

Tool	Purpose
Streamlit	Interactive web dashboard
Scipy / Statsmodels	Statistical tests, power analysis
NumPy	Bayesian sampling, numerical ops
Plotly	Interactive charts (forest plots, posteriors, CI viz)
Pandas	Data manipulation

---

Espanol

Que es A/B Testing?

El A/B testing es la practica de correr experimentos controlados para medir el impacto causal de un cambio - una nueva landing page, un boton de diferente color, el asunto de un email, una estrategia de precios, etc. En vez de adivinar que funciona, dejas que los datos decidan.

Este dashboard implementa un flujo completo de A/B testing listo para produccion:

Disenar experimento -> Calcular tamano de muestra -> Correr experimento  
Validar integridad (SRM) -> Test estadistico -> Decision de negocio

Funcionalidades

• Generador de datos sinteticos - dataset de e-commerce realista (mercado LATAM, 45K usuarios, 14 dias)
• Sube tu propio CSV - conecta cualquier dataset de A/B test binario
• Calculadora de tamano de muestra - con controles de power/alpha y curvas de sensibilidad
• SRM Check - detecta asignacion rota antes de mirar resultados
• Z-test frecuentista - intervalos de confianza, p-values, power alcanzado
• Analisis bayesiano - distribuciones posteriores, P(B>A), expected loss
• Analisis por segmentos - forest plots por device, pais, o cualquier columna categorica
• Proyeccion de impacto en revenue - uplift mensual y anual segun tu trafico y ticket promedio
• Exportar resultados - descarga el reporte completo como CSV

Inicio rapido

# Clonar
git clone https://github.com/JulianDataScienceExplorerV2/Marketing-AB-Testing.git
cd Marketing-AB-Testing

# Instalar dependencias
pip install -r requirements.txt

# Correr la app
streamlit run app.py

Luego abre http://localhost:8501 en el navegador.

---

Conceptos clave / Key Concepts

Concept	Description
MDE	Minimum Detectable Effect - the smallest lift worth detecting
SRM	Sample Ratio Mismatch - mismatch in group sizes signals a broken experiment
Power	Probability of detecting a real effect (typically 80%)
a (alpha)	Significance level - acceptable false positive rate (typically 5%)
Peeking	Stopping early when p < a - inflates false positive rate
Expected Loss	Bayesian metric: how much conversion you risk by picking the wrong variant

---

Made with code by JulianDataScienceExplorerV2

Data-driven decisions, not gut feelings.