Understanding Gain and Lift Charts: Essential Tools for Marketing Campaign Optimization

What Are Gain and Lift Charts?

Gain and lift charts are powerful visual tools that help marketers and data scientists evaluate the effectiveness of predictive models, particularly for classification problems with imbalanced datasets. These charts are especially valuable in marketing campaigns, where response rates typically hover around 1-2%.

Why They Matter

For businesses running targeted marketing campaigns, every contact costs money. These charts help answer critical questions like:

• “If we only contact the top 20% of customers, what percentage of potential responders will we reach?”
• “How much more efficient is our model compared to random selection?”
• “At what point do we hit diminishing returns in our campaign?”

How Gain and Lift Charts Work

Both charts rely on the same fundamental process:

1. Model Scoring: Your predictive model assigns each customer a probability score (likelihood to respond)
2. Sorting: Customers are ranked from highest probability to lowest probability
3. Decile Creation: This sorted list is divided into 10 equal groups (deciles)
4. Cumulative Measurement: We track positive responses as we move through these ordered deciles

Let’s look at each chart in detail:

Gain Charts Explained

A gain chart shows the percentage of total positive responses captured within a percentage of the population.

How to Read a Gain Chart

• X-axis: Percentage of customers contacted (by decile)
• Y-axis: Percentage of total positive responses captured
• Baseline (diagonal line): Represents random selection
• Gain curve: Shows your model’s performance

Gain Chart Example

Imagine a bank running a telemarketing campaign for term deposits with 10,000 customers, where historically only 500 customers (5%) subscribe.

Decile	Customers	Actual Subscribers	Cumulative Subscribers	Gain (%)
1	1,000	200	200	40%
2	1,000	100	300	60%
3	1,000	75	375	75%
4	1,000	50	425	85%
5	1,000	25	450	90%
6	1,000	20	470	94%
7	1,000	15	485	97%
8	1,000	10	495	99%
9	1,000	5	500	100%
10	1,000	0	500	100%

The gain chart shows that by contacting just the top 30% of customers identified by our model, we can reach 75% of all potential subscribers!

A lift chart shows how much better your model performs compared to random selection at identifying positive responses.

How to Read a Lift Chart

Lift Charts Explained

• X-axis: Percentage of customers contacted (by decile)
• Y-axis: Lift value (model effectiveness compared to random)
• Baseline (horizontal line at 1.0): Represents random selection
• Lift curve: Shows your model’s performance relative to random

Lift Chart Example

Using the same banking example:

Decile	Gain (%)	Expected Random Gain (%)	Lift
1	40%	10%	4.0
2	60%	20%	3.0
3	75%	30%	2.5
4	85%	40%	2.13
5	90%	50%	1.8
6	94%	60%	1.57
7	97%	70%	1.39
8	99%	80%	1.24
9	100%	90%	1.11
10	100%	100%	1.0

The lift chart reveals that our model is 4 times more effective than random selection when targeting the top 10% of customers!

Business Applications

Marketing Campaign Optimization

By using these charts, marketers can:

• Target only the most responsive segments to maximize ROI
• Reduce campaign costs by avoiding unlikely responders
• Test multiple models to find the best predictor of customer behavior

Real-World Example

A retail bank running a term deposit campaign with a 2% baseline response rate used a predictive model to identify high-potential customers. By targeting only the top 30% of customers, they:

• Reached 75% of potential subscribers
• Reduced campaign costs by 70%
• Increased the effective response rate to 5%
• Achieved the same number of conversions while spending significantly less

How to Create Gain and Lift Charts in Python

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.metrics import roc_curve, auc

def plot_gain_lift_curves(y_true, y_proba):
    # Sort by probability in descending order
    sorted_indices = np.argsort(y_proba)[::-1]
    y_true_sorted = y_true[sorted_indices]
    
    # Total positive examples
    total_positives = np.sum(y_true)
    
    # Calculate cumulative gains and lift
    population = len(y_true)
    decile_size = population // 10
    gains = []
    lifts = []
    
    for i in range(1, 11):
        cutoff = min(i * decile_size, population)
        positives_found = np.sum(y_true_sorted[:cutoff])
        gain = positives_found / total_positives * 100
        lift = (positives_found / cutoff) / (total_positives / population)
        
        gains.append(gain)
        lifts.append(lift)
    
    # Create gain chart
    plt.figure(figsize=(12, 5))
    
    plt.subplot(1, 2, 1)
    plt.plot([0, 100], [0, 100], 'r--', label='Baseline')
    plt.plot([i*10 for i in range(11)], [0] + gains, 'b-', marker='o', label='Model')
    plt.title('Cumulative Gain Chart')
    plt.xlabel('Percentage of Customers Contacted')
    plt.ylabel('Percentage of Positive Responses')
    plt.grid(True)
    plt.legend()
    
    # Create lift chart
    plt.subplot(1, 2, 2)
    plt.plot([i*10 for i in range(1, 11)], [1] * 10, 'r--', label='Baseline')
    plt.plot([i*10 for i in range(1, 11)], lifts, 'g-', marker='o', label='Model')
    plt.title('Lift Chart')
    plt.xlabel('Percentage of Customers Contacted')
    plt.ylabel('Lift Value')
    plt.grid(True)
    plt.legend()
    
    plt.tight_layout()
    plt.show()

Conclusion

Gain and lift charts transform how marketers approach campaign targeting. By visualizing model performance across different customer segments, these charts help businesses make data-driven decisions that maximize results while minimizing costs.

The next time you’re planning a marketing campaign, consider using these powerful tools to identify your highest-value prospects and optimize your targeting strategy.

---