Best practices and recommendations for more efficient, personalized collections strategies

Collections activities enormously impact the financial performance of U.S.-based wireless carriers. There are 1%-5% of all U.S. subscriber accounts in delinquency at any given time. And with over 450 million post-paid wireless accounts active in the U.S. and an average past due balance between $200 and $300, that means there are over $3 billion dollars that are past due and at risk. To combat these startling stats, wireless carriers need to take advantage of new innovations in advanced analytics and holistic, cloud-native risk decisioning solutions to execute best-practice treatments before consumers go past due. Telcos that deploy advanced analytics to get ahead of payment risks see up to a 10% improvement in recovery rates when compared to those who use legacy processes and static scorecard methods.

Adopting these newer innovations and best practices can drastically reduce operating costs within your collections functions and also increase returns on collections activities. The ease with which internal and third-party data sources can be integrated and orchestrated, and the ease with which advanced analytics can be set up, tested and promoted to production, are primary drivers of these returns on investment.

So we’re looking at exactly what these best practices are for pre-collections and collections decisioning, and what has worked for large telco organizations around the globe.

Looking at best practices from telco companies around the world reveals that a collections risk decisioning strategy for wireless carriers should consist of at least seven key components. And the platform upon which these are configured and executed must allow simple, self-service access for business users to set up, test, and deploy each component without added burden on tech teams or IT.

Modern, cloud-native risk decisioning solutions allow business users to administer the creation and testing of individual decisioning objects or nodes. These nodes interact with each other either concurrently or sequentially and range in complexity from simple business rules to advanced analytics, which users can then create and manage through a low-code interface to improve returns on collections activities. Additionally, decisioning software that is user friendly reduce the technical burden and operating costs of the collections function. What does this mean? In short: business users must be able to manage the end-to-end flow in both test and production environments without having to involve IT.

Here’s an example of a best-practice collections decisioning workflow, which comes from dozens of large-scale implementations thanks to the subject matter expertise of risk and collections professionals. They created this end-to-end sequence for wireless carriers to use, and it can be modified as necessary to adapt to different requirements in order to efficiently execute next-best treatments.

The workflow pictured above uses a combination of on-us behavior data, off-us behavior data from 3rd parties such as credit bureau and speciality telco data, previous contact history data, and socio-demographic data. All of these combine to build a holistic, comprehensive view of a delinquent customer, as outlined in the seven components we described.

Leveraging these various data sources and applying advanced analytics such as random forest or XGBoost machine learning techniques to predict behavior, propose settlement amounts, and to gauge time and channel preferences allows collection teams to build a more targeted, personalized approach to collections, based on customer preferences and circumstances.

Making a significant departure from more traditional, legacy processes (which often rely on core static classifications such as days past due or single risk scores), this new approach highlights a more modern, individualized way of ensuring efficient, effective collections strategies. By evolving beyond logistic regression and decision trees to next-generation collections models that lean on machine learning (which learns from previous nodes within its model construct), the final customer treatment is much more personalized, focused on outcomes and response propensity.