Key Takeaways: Chapter 27 — Customer Analytics and Segmentation

The Core Idea

Not all customers are equal, and treating them as if they were means systematically under-investing in your most valuable relationships and missing early warning signs of churn. Customer analytics is the discipline of making those differences visible so you can act on them.

Customer Lifetime Value (CLV)

• CLV = Average Purchase Value × Purchase Frequency × Customer Lifespan (simple model)
• For richer analysis, compute CLV at the individual customer level from transaction history
• Margin-adjusted CLV (using gross margin, not just revenue) is more accurate
• CLV informs how much you can rationally spend to acquire and retain customers

RFM Analysis

• R = Recency: Days since last purchase. Lower days = better = higher score.
• F = Frequency: Total number of purchases. Higher = better = higher score.
• M = Monetary: Total spend. Higher = better = higher score.
• Quintile scoring (pd.qcut()) converts raw values to 1–5 scores with equal customer counts per bucket
• The combination of scores, not any single dimension, determines the segment

Key Segments and Their Actions

Cohort Analysis

• An acquisition cohort is all customers who made their first purchase in the same time period
• Cohort retention rates measure what percentage of an acquired cohort returns in each subsequent period
• The retention heatmap is one of the most powerful visualizations in customer analytics
• Key insight: a decline in cohort retention rates (especially month-1) indicates a structural problem with customer experience, not just a bad month
• Revenue retention can exceed customer retention — surviving customers tend to expand their spend over time

K-Means Clustering

• K-means finds natural groupings in customer data without requiring predefined rules
• Always scale features before clustering: use StandardScaler from scikit-learn
• The elbow method plots inertia (within-cluster sum of squares) vs. K to find the optimal number of clusters
• Cluster labels are assigned by you (not the algorithm) based on interpreting the centroid profiles
• Use K-means for discovery; use rule-based RFM for operational, explainable segmentation

Churn Indicators

• Recency is the earliest behavioral signal — customers go quiet before they officially leave
• Declining order frequency and shrinking product breadth are secondary signals
• The most useful churn analysis compares recent vs. prior periods to detect relative decline, not just absolute inactivity
• Early detection is far more cost-effective than win-back after the customer has left

Customer Health Score

• A 0–100 composite score that synthesizes recency, frequency, monetary, and trend signals
• Enables account management at scale: sort by score, focus on the bottom
• Health tiers (Critical / At Risk / Healthy / Thriving) give non-technical colleagues an instant signal
• Must be monitored over time — a single snapshot is far less useful than a trend

Technical Patterns to Remember

# Quintile scoring with inversion for recency
rfm["r_score"] = pd.qcut(
    rfm["recency_days"], q=5, labels=[5, 4, 3, 2, 1], duplicates="drop"
).astype(int)

# Building cohort period number
df["period_number"] = (
    df["order_period"] - df["cohort_period"]
).apply(lambda x: x.n)

# Scaling before K-means
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
features_scaled = scaler.fit_transform(features)

# K-means elbow
from sklearn.cluster import KMeans
inertia = [KMeans(k, random_state=42).fit(X).inertia_ for k in range(2, 11)]

Business Principles

1. Analytics without action is decoration. Every segment you identify should have an owner and a defined next step.

2. Recency predicts churn earlier than spend decline. By the time a customer's spend drops, the relationship may already be damaged. Watch for silence.

3. The "At Risk" and "Cannot Lose Them" segments deserve disproportionate attention. They represent customers who have already demonstrated willingness to spend — the relationship is proven. Recovery is easier than new acquisition.

4. Make it a habit, not a project. Customer analytics run quarterly is useful. Customer analytics run monthly with segment-over-segment comparison is transformative.

5. Cohort retention is a leading indicator. Monthly revenue is a lagging indicator. If your cohort retention is declining, your revenue will follow.

What Comes Next

Chapter 28 (Sales and Revenue Analytics) builds on this foundation by analyzing the sales pipeline, rep performance, and revenue decomposition — the mechanics by which your business converts customer relationships into revenue.