Striking the right balance on promotions

Mobile CSPs in highly competitive markets often look to maximize ARPU of their existing customers because most people who want a mobile device already have one. Because the lifetime value (LTV) of postpaid customers far exceeds that of prepaid, or pay-as-you-go (PAYG) customers, it is in the CSP’s interest to convert as many prepaid customers as possible into postpaid customers.

Amdocs recently worked with a tier 1 EU CSP who in previous campaigns had sent monthly communications to its entire prepaid customer base. In the end, it successfully converted its target number of subscribers. However, it later realized that targeting people who were not likely to convert was not only a waste of effort but had the potential to negatively impact its Net Promoter Score (NPS). Accordingly, the communication policy has shifted so that only those PAYG customers most likely to convert would receive personalized migration offers and the remaining customers would receive more relevant contextual campaign offers.

Amdocs was tasked with helping the CSPs analyze and categorize all subscribers; create and execute real time personalized campaigns; and effectively fulfill the generated demand.

The challenge, then, was how to determine which customers were most likely to convert, free the rest of them to relevant campaigns and still achieve the desired number of converted customers.

AI maximizes likelihood of conversion

Amdocs began by working directly with the business team to identify the factors that correlate with increasing the likelihood of converting from PAYG to postpaid.

The Amdocs Machine Learning (ML) model was conceptualized, designed, and tested on real data prior to engaging actively with the customer. The model scored each customers’ probability to migrate, and demonstrable results were available from day one.

One of the most powerful features of the model is that it gave the customer clear sight of the optimum point of minimal targeting and maximum response. To enable this, it was necessary to evaluate the key migration drivers and identify, at a customer level, the behaviors that have the highest impact on the decision to migrate.

Once operationalized it was possible to pinpoint the optimal minimal number (of specific) subscribers to target and still meet the migration KPI. Unlike most, this tier 1 EU CSP recognized that there is a large portion of PAYG customers who are not interested in migration to postpaid.

Thanks to a detailed understanding of the customer, their data, and their business, as well as the end-to-end capability that the Amdocs system delivers, it was possible to embark on a project with an ambitious timeline.

During the development stage the model was designed and tested on real CSP data. New data aggregations were required, and once refined and validated system attributes were created which scored each subscriber’s probability to migrate to a postpaid solution. In parallel, new procedures were defined and implemented which automated monthly scoring of the base, and the periodic retraining of the model to ensure ongoing accuracy and long-term consistent results.

Before implementation the model was trained twice (to optimize its ability to learn and evolve). Amdocs then developed and QA’ed the new campaign logic leveraging fully the new ML attributes. End to end, the entire process took less than three months.