Cohort Analysis

Overview

Cohort analysis tracks groups of users with shared characteristics over time, revealing patterns in retention, engagement, and lifetime value.

When to Use

• Measuring user retention rates and identifying when users churn
• Analyzing customer lifetime value (LTV) and payback periods
• Comparing performance across different user acquisition channels or campaigns
• Understanding how product changes affect different user groups over time
• Tracking engagement patterns and identifying early warning signs of churn
• Evaluating the long-term impact of onboarding improvements or feature releases

Core Concepts

• Cohort: Group of users sharing a characteristic (signup date, region, etc.)
• Cohort Size: Initial group size
• Retention Rate: Percentage remaining active
• Churn Rate: Percentage who left
• Retention Curve: How cohort degrades over time

Cohort Types

• Acquisition Date: Users grouped by signup period
• Behavioral: Users grouped by actions taken
• Revenue: Users grouped by purchase value
• Geographic: Users grouped by location
• Demographic: Users grouped by characteristics

Implementation with Python

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

# Create sample user lifecycle data
np.random.seed(42)

# Generate user data
n_users = 5000
users = []

for user_id in range(n_users):
    signup_month = np.random.choice(range(1, 13))
    lifetime_months = np.random.poisson(6) + 1

    for month in range(1, lifetime_months + 1):
        users.append({
            'user_id': user_id,
            'signup_month': signup_month,
            'month': month,
            'active': 1,
        })

df = pd.DataFrame(users)

# Add derived columns
df['cohort_month'] = df['signup_month']
df['cohort_age'] = df['month']  # Could be day, week, etc.
df['date'] = pd.to_datetime('2023-01-01') + pd.to_timedelta(df['signup_month'] * 30, unit='D')

print("User Data Summary:")
print(df.head(10))

# 1. Cohort Table (Retention Matrix)
cohort_data = df.groupby(['cohort_month', 'cohort_age']).agg({
    'user_id': 'nunique'
}).reset_index()
cohort_data.columns = ['cohort_month', 'cohort_age', 'unique_users']

# Create pivot table
cohort_pivot = cohort_data.pivot(index='cohort_month', columns='cohort_age', values='unique_users')

print("\nCohort Sizes (Raw User Counts):")
print(cohort_pivot)

# 2. Cohort Retention (as percentage of cohort size)
cohort_size = cohort_pivot.iloc[:, 0]
retention_table = cohort_pivot.divide(cohort_size, axis=0) * 100

print("\nCohort Retention Rate (%):")
print(retention_table.round(1))

# 3. Visualize Retention Matrix
fig, axes = plt.subplots(2, 1, figsize=(14, 8))

# Heatmap of raw counts
sns.heatmap(cohort_pivot, annot=True, fmt='g', cmap='YlOrRd', ax=axes[0],
            cbar_kws={'label': 'User Count'})
axes[0].set_title('Cohort Sizes - User Counts')
axes[0].set_xlabel('Cohort Age (Months)')
axes[0].set_ylabel('Cohort Month')

# Heatmap of retention rates
sns.heatmap(retention_table, annot=True, fmt='.0f', cmap='RdYlGn', vmin=0, vmax=100,
            ax=axes[1], cbar_kws={'label': 'Retention %'})
axes[1].set_title('Cohort Retention Rates (%)')
axes[1].set_xlabel('Cohort Age (Months)')
axes[1].set_ylabel('Cohort Month')

plt.tight_layout()
plt.show()

# 4. Retention Curve
fig, ax = plt.subplots(figsize=(12, 6))

# Plot retention curves for each cohort
for cohort_month in cohort_pivot.index[:8]:  # First 8 cohorts
    cohort_retention = retention_table.loc[cohort_month]
    ax.plot(cohort_retention.index, cohort_retention.values, marker='o', label=f'Cohort {cohort_month}')

ax.set_xlabel('Cohort Age (Months)')
ax.set_ylabel('Retention Rate (%)')
ax.set_title('Retention Curves by Cohort')
ax.legend(bbox_to_anchor=(1.05, 1), loc='upper left')
ax.grid(True, alpha=0.3)
ax.set_ylim([0, 105])

plt.tight_layout()
plt.show()

# 5. Average Retention Curve
fig, ax = plt.subplots(figsize=(10, 6))

# Calculate average retention at each age
avg_retention = retention_table.mean()
ax.plot(avg_retention.index, avg_retention.values, marker='o', linewidth=2, markersize