step1_business_understanding.md

🎯 Step 1: Business Understanding & Hypothesis Framing

🔍 Business Problem

PowerCo, a major gas and electricity utility provider for small and medium-sized enterprises, is experiencing customer churn. With increased competition in the energy sector, customers now have more options than ever, making retention a challenge. PowerCo has engaged data scientists to diagnose the root causes of churn and recommend data-driven solutions.

⚡ Understanding Churn

Churn occurs when customers leave PowerCo to opt for a competitor’s services. The key question to investigate is:
➡️ What factors influence customer churn?
One primary hypothesis is that price sensitivity plays a significant role in customer retention.

📌 Key Hypothesis

We hypothesize that: ✅ Customers are highly price-sensitive, meaning an increase in pricing could result in higher churn rates.
✅ Other factors such as customer service quality, energy type (clean vs. non-clean), contract terms, and geographical location might also contribute to churn.

📊 Required Data

To test our hypothesis, we need access to: 📂 Customer Data – Industry, historical electricity consumption, contract duration, date joined, and other customer characteristics.
📂 Churn Data – Information on whether a customer has left PowerCo.
📂 Historical Price Data – The prices charged for electricity and gas at different time intervals.
📂 Competitor Pricing Data (if available) – To compare PowerCo’s pricing strategy with competitors.

Dataset Overview (Business Perspective)

To predict customer churn in an energy services company, we use two main datasets:

1. Client Data

This dataset contains detailed information about each client company, including:

• Customer Profile & Activity:

  • Category of business activity (activity_new)
  • Sales channel (channel_sales)
  • Whether the client uses gas (has_gas)

• Consumption History:

  • Electricity and gas consumption over the past 12 months (cons_12m, cons_gas_12m)
  • Consumption from the last month (cons_last_month)

• Contract Lifecycle:

  • Dates of activation, renewal, end of contract, and last modification
    (date_activ, date_renewal, date_end, date_modif_prod)

• Forecasts & Pricing:

  • Forecasted consumption and discount metrics (forecast_cons_12m, forecast_cons_year,
    forecast_discount_energy)
  • Forecasted prices and meter rent (forecast_meter_rent_12m, forecast_price_energy_off_peak,
    forecast_price_energy_peak, forecast_price_pow_off_peak)

• Financial Metrics:

  • Gross and net margins, total net margin, and number of active products
    (margin_gross_pow_ele, margin_net_pow_ele, net_margin, nb_prod_act)

• Customer Tenure:

  • Number of years the client has been with the company (num_years_antig)

• Target Variable:

  • churn: whether the client is expected to leave in the next three months

Note: Some fields (e.g., sales channel, campaign origin) are anonymized to preserve privacy,
but may still hold predictive power.

---

2. Price Data

This dataset provides historical pricing information, including:

• Energy Prices:

  • Variable prices for off-peak, peak, and mid-peak energy usage
    (price_off_peak_var, price_peak_var, price_mid_peak_var)

• Power Prices:

  • Fixed prices for power in the same three time periods
    (price_off_peak_fix, price_peak_fix, price_mid_peak_fix)

• Keys:

  • Linked via id (client identifier) and price_date (date of pricing record)

These datasets together enable the development of a machine learning model to identify at-risk customers and guide retention strategies.

🛠️ Approach to Analysis

1️⃣ Define Price Sensitivity – Establish a metric to quantify how customer churn correlates with price changes.
2️⃣ Data Preparation & Feature Engineering – Clean the data, handle missing values, and engineer relevant features.
3️⃣ Exploratory Data Analysis (EDA) – Identify patterns in the data, visualize trends, and detect key correlations.
4️⃣ Modeling Churn Probability – Use binary classification models like Logistic Regression, Random Forest, or Gradient Boosted Machines to predict churn probabilities.
5️⃣ Model Selection & Evaluation – Compare different models based on accuracy, interpretability, and effectiveness in explaining churn.

🎯 Expected Outcome

By completing this analysis, we aim to: 🚀 Understand the key drivers of churn.
🚀 Determine the extent to which price sensitivity influences customer retention.
🚀 Provide actionable insights to PowerCo to mitigate churn through pricing strategies and other retention initiatives.

---

📌 Next Step: Proceed to Step 2: Exploratory Data Analysis 🔍