DIY #24 - Calculate Expected LTV from Churn and Revenue Models

2026-07-06 · Source: Machine Learning Pills · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Sales & Commercial Development · Depth: Intermediate, medium

Summary

This article details a method for calculating Expected Customer Lifetime Value (LTV) by integrating existing churn and revenue prediction models. It proposes the formula "Expected LTV = (1 - churn probability) × expected revenue if active", emphasizing that the revenue model must be trained exclusively on active customers to prevent double-counting churn risk. The workflow involves six steps: calculating churn probability, predicting revenue conditional on activity, multiplying these values, analyzing changes in customer rankings compared to revenue-only views, and evaluating the combined model using metrics like MAE (287.31), RMSE (398.45), and Spearman correlation (0.6412). This approach provides a more realistic customer valuation, as demonstrated by examples where Customer A (70% churn, £1,000 revenue) yields £300 Expected LTV, while Customer B (10% churn, £600 revenue) yields £540.

Key takeaway

For Data Scientists or ML Engineers building customer valuation systems, integrating churn probability into LTV predictions is crucial. You should implement the "Expected LTV = (1 - churn probability) × expected revenue if active" formula, ensuring your revenue model is trained only on active customers. This prevents over-discounting and provides a more accurate, risk-adjusted customer ranking, enabling better resource allocation for retention and marketing efforts. Prioritize models with strong Spearman correlation for effective business decisions.

Key insights

Combining churn probability with conditional revenue prediction yields a more accurate Expected LTV for customer valuation.

Principles

• Churn and revenue models are incomplete alone.
• Revenue models must be conditional on active customers.
• Relative ordering (Spearman) often matters most.

Method

Calculate Expected LTV by multiplying "(1 - churn probability)" from a churn model by "expected revenue if active" from a revenue model trained solely on active customers, reversing log transformations with np.expm1().

In practice

• Retrain LTV models on active customers only.
• Use np.expm1().clip(0) to convert log-scale predictions.
• Compare customer rankings by revenue vs. Expected LTV.

Topics

• Customer Lifetime Value
• Churn Prediction
• Predictive Analytics
• Machine Learning Models
• Model Evaluation
• Customer Segmentation

Best for: Data Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• I Built a Churn Prediction Model for a Telecom Company. Here’s What the Data Actually Revealed. — Effective churn prediction requires thorough EDA and model comparison to reveal non-linear customer behavior.
• DIY #23 - Predict Customer Lifetime Value — LTV prediction uses early customer data via supervised regression to guide acquisition strategy.
• When Customers Churn at Renewal: Was It the Price or the Project? — Accurate churn attribution at renewal requires distinguishing between price shock, value exhaustion, and their interaction.
• Your Churn Threshold Is a Pricing Decision — Churn models must optimize for dollar costs, not just accuracy, by integrating LTV and misclassification costs.
• The Math Behind the Bots Dominating Prediction Markets — Strict mathematical models, not sentiment, drive consistent profitability in prediction markets.
• Extra #12 - Turning Churn Predictions into Business Actions — Effective churn prediction requires economic threshold optimization and comprehensive model diagnostics beyond basic metrics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning Pills.