Experiment - Collection Process

Experimentation in Debt Collection: A Case Study

Experiment Design

A bank could experiment with different collection strategies for customers who have not paid their credit card fees. For instance, the bank might test different call scripts, timing of calls (day of the week, time of day), or the use of different channels (phone, email, SMS).

Machine Learning:

Machine learning can be used to identify customer segments with different response rates to various collection strategies. For example, the bank could use customer data (payment history, demographics, credit score, etc.) to build a model that predicts the likelihood of a customer responding positively to a specific collection approach. This allows for targeted and personalized collection efforts.

Experimentation:

The bank can conduct A/B tests to compare the effectiveness of different collection strategies. For instance:

Attrition and Spillover

Attrition: In the context of debt collection, attrition can occur when customers become unreachable or unresponsive. This can bias results if certain customer segments are more likely to become unreachable under specific collection strategies. For example, if customers who are more likely to pay are also more likely to answer calls, the results of the experiment might be skewed.

Spillover: Spillover effects can occur in debt collection, although they might be less pronounced compared to other industries. For instance, if customers who receive a particularly effective call script share information with other customers, it could impact the results of the experiment. However, given the sensitive nature of debt collection, this type of spillover is less likely to happen.

Mitigating Attrition and Spillover

To mitigate attrition, the bank can implement strategies to increase call connect rates, such as multiple call attempts, call-back options, and SMS reminders. To address potential spillover effects, the bank can carefully design the experiment to minimize information sharing between treatment and control groups. For example, customers in different treatment groups can be assigned to different call centers or collection agents.

By carefully designing experiments and utilizing machine learning to identify optimal strategies, banks can improve collection efficiency, reduce costs, and enhance customer satisfaction.


Key Metrics for Debt Collection Experiments

Measuring the success of a debt collection experiment requires a careful selection of metrics. These metrics should align with the overall goals of the experiment, such as increasing collection rates, reducing delinquency, or improving customer satisfaction.

Here are some key metrics to consider:

Core Metrics

Customer-Centric Metrics

Cost-Efficiency Metrics

Operational Metrics

Additional Considerations