Personalized Collection Strategies Powered by Data Science
Six years ago, we launched a data science initiative that reshaped how Proximus approaches customer debt collection. The goal was to move beyond binary collection flows and toward a nuanced understanding of customer behaviour. The result? A daily-running algorithm that segments customers based on their propensity to bad debt—driving smarter, more empathetic collection strategies.
The Challenge: Moving Beyond Binary Collection Streams
Before this initiative, Proximus already differentiated between customers in its collection strategy—offering leniency to most and applying stricter measures to known defaulters. While this approach acknowledged that not all late payers are the same, it lacked the granularity needed to tailor actions based on individual behaviour and context.
Understanding why a customer is late was key to improving both recovery rates and customer relationships.
The Data Science Journey: From Clustering to Deterministic Rules
We began with unsupervised learning, using clustering algorithms to uncover natural groupings in customer payment behaviour. This initial phase revealed four key segments:
- Good
- Confused
- Forgetful
- Bad
These clusters provided valuable insights, but they weren’t enough. Business stakeholders needed transparency: Why is a customer in this segment? Can we define it in clear, operational terms?
We transitioned to a deterministic, rule-based system that closely mirrored the clustering output. Through multiple rounds of experimentation and close collaboration with business experts, we refined the rule set to accurately capture operational realities and align with business logic.
The Final Segments: Seven Profiles, One Purpose
After multiple validation rounds and business reviews, we evolved the initial clusters into seven actionable customer profiles. These segments combine data science insights with operational expertise, and are designed to reflect real-world customer behaviour in a way that is both actionable and explainable:
- New Mobile Only – Recently joined, mobile-only customers.
- New Other – New customers with other types of subscriptions.
- Recent – Customers who are not brand new but still early in their lifecycle.
- Good – Reliable payers with a strong track record.
- Confused – Customers experiencing disruptions that affect payment behaviour.
- Forgetful – Those who miss payments unintentionally.
- Bad – Customers with a high propensity for bad debt.
These profiles are the engine behind smarter collection flows – helping Proximus make timely, targeted decisions that protect revenue while strengthening customer relationships.
Strategic Outcomes: Personalization That Drives Performance
The segmentation algorithm runs daily, enabling tailored actions that reflect each individual’s current situation.
The segmentation directly powers:
- Automated decision-making within the collection tool,
- Context-aware support in agent interfaces,
- Personalized outbound care calls, tailored to each customer segment.
This approach empowers Proximus to balance empathy with financial discipline—treating customers fairly while protecting business outcomes.
As a result, the initiative has helped Proximus:
- Improve recovery rates,
- Reduce unnecessary friction with customers,
- Foster trust through personalized communication,
- Align operational actions with strategic goals.
Privacy and Compliance by Design
This project was developed in close alignment with Proximus’s Privacy Review Process, ensuring full compliance with GDPR and internal governance standards. From segmentation logic to operational deployment, every step was guided by principles of transparency, data minimization, and responsible use—embedding privacy into the core of the solution.
Looking Ahead
This project is a testament to what’s possible when data science meets deep business expertise. It’s not just about algorithms – it’s about understanding people and acting with purpose.
At Proximus-ADA, we specialize in turning data into actionable insights. If you’re facing similar challenges or want to explore how data science and AI can enhance your customer strategies, we’d love to talk.
Luminita
Senior Data Scientist
Proximus Ada