Customer Lifetime Value Modeling

Customer Lifetime Value Modeling

CLV is the North Star metric of customer economics. Knowing what a customer is worth — not just today but over their full relationship — determines how much to spend acquiring them, how much to invest in retaining them, and which segments deserve the most attention.

CLV Frameworks

Simple CLV (for initial estimates):

CLV = (Average Order Value × Purchase Frequency × Gross Margin) / Churn Rate

Limitations: assumes constant purchase rate and margin, ignores the time value of money.

Discounted CLV (for more accurate models):

CLV = Σ [(Gross Margin × Retention Rate^t) / (1 + Discount Rate)^t] for t = 0 to N

This accounts for the fact that $1 of revenue in year 3 is worth less than $1 today.

Example: Gross margin = $60/year, Retention rate = 75%, Discount rate = 10%:

• Year 0: $60 / 1.0 = $60.00
• Year 1: $60 × 0.75 / 1.1 = $40.91
• Year 2: $60 × 0.5625 / 1.21 = $27.89
• Year 3: $60 × 0.4219 / 1.331 = $19.01
• CLV ≈ $147.81 (first 3 years) vs. the simple model: $60/0.25 = $240

The simple model overestimates by 62% in this case because it ignores time value.

BG/NBD and Pareto/NBD Models

For e-commerce and subscription-adjacent businesses, probabilistic models predict CLV from transaction history.

BG/NBD (Beta Geometric / Negative Binomial Distribution):

• Models two processes simultaneously: transaction process (how often do they buy when active?) and dropout process (when do they go dormant?)
• Input: recency, frequency, monetary value, and total observation period
• Output: probability that customer is still active, expected transactions in next N periods
• Used by Amazon, Spotify, and many e-commerce companies for CLV prediction

Practical implementation: the "lifetimes" Python library provides BG/NBD and gamma-gamma CLV models. For a SaaS business, churn-based CLV is simpler and more accurate.

RFM Segmentation

RFM scores each customer on three dimensions (typically 1-5 scale) and segments them into groups that predict future behavior:

Champions (5-5-5): bought recently, buy often, spend the most. These are your best customers — protect them.

Loyal customers (4-4-4): high frequency, moderate recency. Likely to respond to loyalty programs and referral asks.

At-risk (2-5-4): high frequency and value historically, but haven't purchased recently. High priority for win-back campaigns.

Lost (1-1-1): haven't purchased in a long time, haven't bought frequently, low monetary value. Usually not worth re-acquisition effort.

RFM-driven marketing:

• Champions → exclusive access, referral program, product beta testing
• At-risk → win-back campaign with strong offer ("we miss you — here's 20% off")
• Lost → either do not contact (cost of outreach > expected value) or test re-engagement with a very low-cost channel

Cohort Analysis

Cohort analysis tracks customer groups over time to understand whether business quality is improving or declining.

Revenue cohort chart: visualize monthly revenue from each acquisition cohort over 12+ months.

• Steeper initial drop = higher early churn
• Flat line over time = retained customers stabilized
• Upward curve = expansion revenue (customers are buying more over time)

Benchmark questions from cohort analysis:

• What % of month-1 cohort is still active at month 12? (12-month retention rate)
• Is the month-12 revenue from each cohort trending up or down? (cohort quality trend)
• Do customers acquired through specific channels have better retention? (channel quality analysis)

The importance of retention curves: two businesses can have identical month-1 revenue but dramatically different CLV if their retention curves diverge. A business with 90% month-2 retention vs. 60% month-2 retention will have 3-4× higher CLV after 24 months.

Churn Prediction and Intervention

Proactive churn prevention is more efficient than re-acquisition. Predicting who will churn before they do enables targeted intervention.

Churn indicators (behavioral signals):

• Declining login frequency (SaaS)
• Reducing purchase frequency (e-commerce)
• Decreasing session length
• Increasing support ticket volume (often precedes churn)
• Feature adoption decline

Churn prediction model outputs:

• Probability of churn in next 30/60/90 days per customer
• Predicted revenue at risk from high-churn-risk customers

Intervention targeting: not all churn is worth preventing. Focus intervention on:

High CLV × High churn probability = highest priority for personal outreach

High CLV × Low churn probability = monitor and maintain

Low CLV × High churn probability = automated discount or win-back email only

Low CLV × Low churn probability = no intervention needed

CLV-Based Budget Allocation

CLV fundamentally changes how to think about marketing investment.

CLV-informed CAC targets: if CLV = $500 and target CLV:CAC = 3:1, acceptable CAC ≤ $167.

CLV by channel: customers acquired through different channels often have dramatically different CLV. Organic search customers may have 2× the CLV of paid social customers because they had higher purchase intent. This changes channel economics: a channel with higher CAC but higher CLV can be more profitable than a cheaper channel with low CLV.

Segment-specific investment: high-CLV customer segments justify higher CAC (more targeted, more personalized outreach) because the return per customer is higher. Segment-blind marketing averages this out and under-invests in high-value segment acquisition.