Simply deploying a CVM solution for an operator isn't enough. It's also crucial to cultivate specialists' skills to create impactful campaigns. Eastwind supports its customers by organizing regular training sessions for the telco staff on different modules of the Mobile Marketing Platform (MMP).

In December 2023, we visited Kazakhstan to train the Kcell team on working with Big Data and ML models in MMP. Let’s see how it was and explore the key data analysis areas that telecom operators can work with.
  Kcell, a mobile operator in Kazakhstan, servesabout 8 million subscribers. Since 2014, Eastwind has equipped the operator with a robust CVM solution, the EW Mobile Marketing Platform. This platform empowers Kcell to harness big data and craft impactful marketing communications.

Developing ML Models in MMP: Basics and Nuances

Eastwind's data scientists presented the MMP capabilities, demonstrated model development processes, and answered the Kcell team's questions. Key discussion topics included:
 

• Feature construction and selection

Features are variables an ML model uses to make predictions. Significant features for a specific model could include external data about subscribers: gender, age, or the amount of traffic they spend on various services. For example, if a subscriber consumes a lot of data on YouTube, we can offer them an additional package or an unlimited plan for this channel.
 

• Iterative machine learning

This approach to model development enhances accuracy and performance in stages. Firstly, we test the model on a specific dataset and compare the predicted and the actual values. Then the model digests the feedback to refine the algorithm. The cycle repeats until the ML model produces the best outcome.

Iterative Approach Stages 

Correct feature selection and iterative model development allow for the most accurate predictions. This means subscribers will receive more relevant recommendations and purchase more of the operator's services.

Use Cases: Next Best Offer and Churn Prevention

In addition to theory, we discussed specific Kcell campaigns.

Next Best Offer (NBO) is one of the most in-demand cases, where the model recommends the next most suitable service based on the subscriber activity and profile characteristics. For instance, if a user frequently makes international calls, offering them a special package for calls rather than for SMS is better.


Models Used for Predicting the Next Purchase

Churn prevention is another popular case. An ML model predicts the likelihood that a particular user will stop using the services. If such risk is significant, the operator's team can try to prevent the churn by offering bonus packages or another plan.

In each a case, we examined the various approaches to model building, strategies for enhancing prediction accuracy, key features to rely on, and how to leverage the outcomes in advertising campaigns. In the long run, increasing the number of effective recommendations and dealing with churn increases ARPU, and the operator's revenues grow.

Learn More About MMP's Capabilities

 

Brainstorm: Integrating Technology and Business

At the end of the training, we organized a brainstorming session to exchange ideas for improving models. The engineering and commercial departments of Kcell participated in this discussion. Participants compared their visions of CVM from both technological and business perspectives to discover new insights.

"We discussed how operator data can assist in credit scoring. For example, you can analyze the number and frequency of SMS from banks and microfinance organizations, the change of subscriber devices and their use for casino or pawnshop services."

Natalia Bronina, Head of Data Analysis Group, Eastwind

Kcell Team at the Eastwind Big Data Training

The Kcell team has had prior training sessions arranged by Eastwind. Moving forward, we aim to consistently offer training and consultations to help the operator fully leverage the benefits of the EW Mobile Marketing Platform.