Portfolio Overview

Key Assumptions

• Portfolio: synthetic 5,000-customer portfolio
• Time horizon: 3 years
• Discount rate: 10%
• Retention probability: derived from churn model
• Segment method: K-means clustering

Enhancement: - 1. Extend the mock data generation to include retention uplift. Show not just “high-value customers at risk of churn”, but “high-value customers likely to respond to retention”.

• Rather than simply predicting churn, they use uplift models to identify customers who will respond to retention action — not just customers likely to churn.

The key distinction:

Churn prediction tells you who might leave.

Uplift modelling tells you who will stay because of your intervention.

This is more relevant for CLV optimisation. The goal is often to move from static CLV to dynamic, action-aware CLV. with a A. Profit-Centric CLV, Instead of only using net margin, they use profit-aware modelling that accounts for:

Acquisition cost

Retention cost

Cross-sell and up-sell opportunities

Risk-adjusted revenue

Capital allocation cost

Operational cost per segment

• 2. Add an optimisation table or chart: Customer CLV, Churn Probability, Expected Retention Uplift, Expected Profit from Intervention.
• 3. Create an R function that encapsulates the Baesens-style profit-aware targeting logic. For this, replace simple net margin with a more detailed profit function.
• 4. Add a short note in the dashboard: “Method inspired by profit-centric customer analytics research.”
• the research strand with a sequential approach:

Define churn

Predict churn probability

Estimate CLV

Estimate uplift or treatment effect

Optimise targeting based on expected profit

Verbeke, Dejaeger, Martens, Baesens — New Insights into Churn Prediction in the Telecommunication Industry Focus: profit-centric churn prediction.

Baesens et al. — Uplift Modelling / Persoanlised Treatment Learning Focus: identifying customers worth targeting.

Óskarsdóttir, Bravo, Sarraute, Baesens — Customer Lifetime Value in Financial Services Focus: CLV modelling for credit card and lending portfolios, sometimes using network data and deep learning.

Customer Value by Segment

Churn Risk by Segment

Top Value Customers

Model Approach

Customer Lifetime Value is calculated as:

\[ CLV_i = \sum_{t=1}^{T} \frac{M_i \cdot r_i^t}{(1+d)^t} \]

Where:

• \(M_i\) = net margin after risk and operating costs
• \(r_i\) = retention probability from churn model
• \(d\) = discount rate
• \(T\) = time horizon

The central insight: a customer is not a transaction, but a stream of risk-adjusted cash flows.

Value Composition

• Revenue: interest margin + fee income
• Cost: capital cost + risk cost + operating cost
• Retention: derived from churn prediction

CLV Distribution

CLV versus Churn Risk

High-Value / High-Risk Customers

Clustering Approach

Customer segments are identified using K-means clustering on:

• Income
• Transaction frequency
• Digital engagement

The method groups customers by similarity, not by prior business rules. This allows the data to reveal structure that may not be visible in traditional segmentations.

Segment Profiles

Segment Map

Segment Value Concentration

Segment Validation

Churn Prediction Model

Churn is predicted using a logistic regression model with behavioural features:

• Product tenure
• Number of products
• Transaction frequency
• Recent decline in usage
• Complaints or service interactions
• Digital engagement

The output is a 12-month churn probability for each customer.

Top Churn Drivers

Predicted Churn Distribution

Churn by Segment

High-Value Customers at Risk of Churn