Introduction

PowerCo, a client of Boston Consulting Group (BCG), aims to analyze its customer dataset to identify patterns and predict which customers are likely to churn within the next three months. The company seeks to implement a data-driven churn prediction model to proactively address customer attrition and improve retention strategies.

To achieve this, PowerCo has requested a thorough analysis of historical customer behavior, contract details, energy consumption trends, and pricing factors to build a predictive model that accurately identifies customers at high risk of churning. The model will help PowerCo make targeted interventions before customers leave, minimizing revenue loss and improving customer satisfaction.

Additionally, PowerCo is exploring whether offering a 20% discount to customers predicted to churn would be a profitable retention strategy. The key question is whether the increased retention rate and extended customer lifetime value (CLV) would outweigh the revenue loss caused by the discount.

This project will help PowerCo make data-driven retention decisions, optimize pricing strategies, and ultimately improve profitability while reducing customer churn.

About this file

client_data.csv

• id = client company identifier

• activity_new = category of the company’s activity

• channel_sales = code of the sales channel

• cons_12m = electricity consumption of the past 12 months

• cons_gas_12m = gas consumption of the past 12 months

• cons_last_month = electricity consumption of the last month

• date_activ = date of activation of the contract

• date_end = registered date of the end of the contract

• date_modif_prod = date of the last modification of the product

• date_renewal = date of the next contract renewal

• forecast_cons_12m = forecasted electricity consumption for next 12 months

• forecast_cons_year = forecasted electricity consumption for the next calendar year

• forecast_discount_energy = forecasted value of current discount

• forecast_meter_rent_12m = forecasted bill of meter rental for the next 2 months

• forecast_price_energy_off_peak = forecasted energy price for 1st period (off peak)

• forecast_price_energy_peak = forecasted energy price for 2nd period (peak)

• forecast_price_pow_off_peak = forecasted power price for 1st period (off peak)

• has_gas = indicated if client is also a gas client

• imp_cons = current paid consumption

• margin_gross_pow_ele = gross margin on power subscription

• margin_net_pow_ele = net margin on power subscription

• nb_prod_act = number of active products and services

• net_margin = total net margin

• num_years_antig = antiquity of the client (in number of years)

• origin_up = code of the electricity campaign the customer first subscribed to

• pow_max = subscribed power

• churn = has the client churned over the next 3 months

price_data.csv

• id = client company identifier

• price_date = reference date

• price_off_peak_var = price of energy for the 1st period (off peak)

• price_peak_var = price of energy for the 2nd period (peak)

• price_mid_peak_var = price of energy for the 3rd period (mid peak)

• price_off_peak_fix = price of power for the 1st period (off peak)

• price_peak_fix = price of power for the 2nd period (peak)

• price_mid_peak_fix = price of power for the 3rd period (mid peak)

Note: some fields are hashed text strings. This preserves the privacy of the original data but the commercial meaning is retained and so they may have predictive power

📚Analysis Report

Click here to read report

📙Code Notebook

Click here to read notebook