Home About Expertise Projects Blogs Contact
Industry SimulationMachine LearningCompleted

BCG X
Customer Churn Analysis

A client-facing data science simulation for BCG X, EDA, feature engineering, and Random Forest modelling to predict customer churn for XYZ Analytics, delivered with an executive summary.

PlatformForage · BCG X
CompletedOctober 7, 2025
TasksEDA · Feature Engineering · Modelling · Executive Summary
ToolsPython · Pandas · NumPy · Scikit-learn
50%
Churn Recall Achieved
5
Simulation Tasks
RF
Random Forest Model
Oct 2025
Completed
🏅
BCG X · Forage Certificate
Completed: October 7, 2025
Verification Code: fiQNa9XpJ4L97CSyJ
Verify ↗
01, Simulation Overview

Working like a BCG data scientist

The BCG X Forage simulation places you inside a real consulting engagement. The client is XYZ Analytics, a company experiencing customer churn. Your job: investigate the data, build a model, and deliver insights to senior leadership in the form they actually want, a concise executive summary, not a Jupyter notebook.

💼
The consulting context
BCG data scientists don't just build models. They frame business problems, translate technical findings into business language, and recommend specific actions. This simulation tests all three, not just the modelling.
02, Task Breakdown

5 tasks, end to end

01
Background, Frame the Problem
Understood the business context: XYZ Analytics is losing customers. Identified what client data is needed, what questions to investigate, and what success looks like for a churn prediction model.
Business framing · Hypothesis definition
02
EDA & Data Cleaning
Loaded customer data with Pandas. Ran descriptive statistics, checked for missing values, and visualised distributions using Matplotlib. Identified key behavioural and demographic signals.
Pandas · NumPy · Matplotlib · describe()
03
Feature Engineering
Created new features from existing columns to improve model signal. Handled categorical variables with encoding. Removed low-signal features.
get_dummies() · Feature creation · Correlation analysis
04
Random Forest Modelling
Trained a RandomForestClassifier on the engineered dataset. Achieved 50% recall on churn prediction, meaning the model catches 1 in 2 churners, a meaningful improvement over random.
RandomForestClassifier · train_test_split · classification_report
05
Executive Summary
Wrote a concise summary translating model findings into business language, what drives churn, confidence in the model, and 3 specific actions XYZ Analytics should take.
Business writing · Insight communication
03, Model Results

What the model found

MetricValueInterpretation
Churn Recall50%Model correctly identifies 1 in 2 churners
Model TypeRandom ForestHandles non-linear patterns, ranks feature importance
Class ImbalancePresentMajority class (retained) dominates, standard churn challenge
Key FeaturesUsage, tenure, plan typeBehavioural signals outperform demographics
📋
The executive summary discipline
The most valuable part of this simulation was learning to write an executive summary. Consultants don't present confusion matrices to CEOs, they present a business problem, a confidence level, and a recommendation. That translation is a skill.
04, Key Takeaways

What consulting data science actually looks like

🎯
Problem framing comes before coding
BCG spends more time understanding the business question than writing code. A wrong framing produces a technically correct but useless model.
📝
The summary matters as much as the model
A 50% recall model delivered with a clear business narrative is more valuable than a 75% recall model with no actionable recommendations.
🔄
Churn is inherently class-imbalanced
Real churn datasets always have far more retained customers than churners. The model must be evaluated on churn recall specifically, not overall accuracy.
⚡
Speed and structure over perfection
Consulting simulations mirror real engagements: tight timelines, structured deliverables, and a senior audience who wants the bottom line first.
05, Tech Stack
Python 3PandasNumPyMatplotlibScikit-learnRandom ForestExecutive Reporting
← Back to Projects View on GitHub ↗