Increasing the number of loyal customers and maximizing profit per user are some of the key goals businesses aim for. This is the reason why telecommunication companies are trying to convert users from pre-paid to post-paid payment methods.
Companies invest significant resources into promoting their offers to a wide audience with lower ROI when the offers are not targeted or adapted to suit different types of users. Our client wanted to reduce the cost of call center agents while increasing, at the same time, the take rate of promoted offers.
Knowing our customers, and how and when to approach them is key to creating a stable, long-term partnership, and increasing campaign take rate and profit while optimizing invested resources in terms of reduced call center expenses as well as promotional costs. Creating a shortlisted customer list with different customer behavioral profiles and tailor-made offers for each user profile could help maximize the value of cross-selling (prepaid to postpaid).
The cross-sell (Pre2Post) model gives insight into the behavior of every user of a telecommunication company based on a series of parameters and data sources including the communication type and its intensity. The model identifies users with the highest probability to switch from a prepaid to a postpaid subscription. It also provides insight into unexploited past opportunities to allow the company to reach out to those customers as well.
Clustering of users who are similar or related to each other (but are different from users in other clusters) allows telecom companies to understand user behavior. Prediction of cross-selling probability (Pre2Post), supported by the clustering algorithm, leads to optimal decision-making.
The Pre2Post model presents the data and most relevant parameters which define user behavior in a visual and intuitive way.
Data from different sources was integrated and stored in a Data Lake. Python was used as the main programming language for the development and deployment of machine learning models.
Fig.1. Cluster examples
Fig. 2. Lift curve example
“ The Pre2Post candidate detection model provided more precise targeting of users whose behavior is most similar to postpaid users so that on the one hand the number of users for contact on a monthly level is optimized, and on the other hand, the result of Pre2Post campaigns was improved multiple times (4 to 6 times, depending on the month and current campaigns on the postpaid service), which consequently led to an increase in revenue per user and a rise in growth on the postpaid service. Within the process of analyzing the business problem and developing two machine learning models, a very important aspect of the implementation was the cooperation with the Comtrade data science team, who led the Telekom data science team with an adequate level of quality, to produce results which exceeded expectations.”
Dr. Siniša Arsić
Head of Service for Intelligent Automation of Business Processes at Telekom
