“If you don’t know what a customer is worth, every marketing campaign is just a blind date with your CFO.”

Welcome to CLV modeling, where we try to answer the question every CEO asks at least once a quarter:

“How much is this customer really worth… before they ghost us?”

🧩 The Core Idea¶

Customer Lifetime Value (CLV) estimates the total future profit you can expect from a customer over their “relationship lifespan.”

Mathematically speaking: [ \text{CLV} = \sum_{t=1}^{T} \frac{(R_t - C_t)}{(1 + d)^t} ] where:

• (R_t): revenue at time (t)

• (C_t): cost at time (t)

• (d): discount rate (because money tomorrow is worth less than money today)

• (T): customer lifespan

In human terms:

CLV = “How much cash can we milk from a customer before they break up with us.”

🧠 Why CLV Matters¶

📦 CLV ≠ Just Revenue¶

Remember: CLV ≠ Total Money Spent CLV = Profit-adjusted, time-discounted, churn-aware forecast of future value

Or as one analyst put it:

“Revenue is vanity, profit is sanity, and CLV is strategy.”

🧮 Simple Deterministic CLV Model¶

If you’re new to this, start small.

[ \text{CLV} = \frac{(m \times r)}{(1 + d - r)} ] Where:

• (m): margin per period

• (r): retention rate

• (d): discount rate

Example:

• Margin = $100/month

• Retention rate = 0.9

• Discount = 0.1

Then: [ CLV = \frac{100 \times 0.9}{1 + 0.1 - 0.9} = $450 ]

That means: on average, each customer contributes about $450 in profit before they churn — unless your support team “helps” them churn faster.

🧪 Python Example: Probabilistic CLV (BG/NBD + Gamma-Gamma)¶

If you’re feeling fancy — and love acronyms — use the lifetimes package.

from lifetimes import BetaGeoFitter, GammaGammaFitter
from lifetimes.datasets import load_cdnow_summary_data_with_monetary_value

data = load_cdnow_summary_data_with_monetary_value()

bgf = BetaGeoFitter(penalizer_coef=0.001)
bgf.fit(data['frequency'], data['recency'], data['T'])

# Predict future transactions
data['pred_purchases'] = bgf.predict(12, data['frequency'], data['recency'], data['T'])

ggf = GammaGammaFitter(penalizer_coef=0.01)
ggf.fit(data['frequency'], data['monetary_value'])

# Predict expected CLV
data['clv'] = ggf.customer_lifetime_value(
    bgf,
    data['frequency'],
    data['recency'],
    data['T'],
    data['monetary_value'],
    time=12,  # months
    discount_rate=0.01
)

data.head()

🎯 The result? A shiny DataFrame of who’s truly valuable, and who’s just here for the trial month and the free tote bag.

🕵️ CLV + Survival = Power Couple¶

• Survival analysis helps estimate how long a customer will stay.

• CLV modeling converts that duration into dollars.

Together, they turn:

“Who will churn?” → “How much will it cost us when they do?”

💡 Use Cases¶

🧠 Pro Tips¶

• Keep an eye on churn — CLV without churn modeling is just optimism.

• Don’t mix up ARPU (average revenue per user) with CLV — that’s like confusing salary with net worth.

• Visualize CLV distribution — it’s always a “power law.” A few customers make you rich, the rest keep your servers busy.

📊 Visual CLV Segmentation Example¶

import seaborn as sns
sns.histplot(data['clv'], bins=30)

You’ll usually find:

• 10% of customers contribute 60–70% of value

• The rest are “nice-to-have” (and nice to email less often)

🎭 TL;DR Summary¶

🧭 Next Stop¶

➡️ Lab – CLV Estimation Let’s get hands-on with data and calculate which customers are your real VIPs (and which ones are just here for the coupons).