Use-case Analysis to Mitigate Value at Risk in Telecom

Use-case Analysis to Mitigate Value at Risk in Telecom

Thursday, March 21
4:00 PM - 4:40 PM
Room 124-125

Main Contribution
With this talk, I intend to bridge the gap between the business case owners and the data scientists who eventually deliver those business cases. In my experience, there exists a noticeable gap between the two, mainly due to different set of priorities. I will indulge these disparate streams of audiences by sharing our two years of data science work at T-Mobile Austria (TMA) from the perspective of helping my fellow data scientists to formulate their use-cases better so that the underlying business value is emphasized. I will present various data science use-cases and how they come together to give a bigger picture of mitigating value at risk. Learning from experience, I will emphasize the use of applied research by the data scientists to help speed up the attribution of their use-cases. This will help the audience to understand the value of an effective data-driven culture and the practices it brings along. Since the entire focus is on data science that attributes the business value, this talk will chalk up the aforementioned objective.

I will summarize our 2 years’ of experience on big data related use-cases at TMA and the best practices in replicating such efforts. All the use-cases are carried out while keeping the single objective in mind, that is, “Mitigating Value at Risk (VaR)”. It comprises of both technology and commercial realms of telecom sector and are solely based on the Hadoop eco-system.

The entire practice of mitigating VaR comprises of set of actions directed to reduce customer churn. Within each set, there exists several use-cases for which the presence of a stable big data platform is vital. For example, some of our use-cases are built on several billion events generated daily by the network, hence the use of a Hadoop cluster with several data nodes came to serve the purpose.

In order to achieve the objective of mitigate VaR, we have executed the use-cases in four steps, some of which are done in parallel. These steps are as follows: (1) measure the monetary value of the customers namely, customer lifetime value (CLV), (2) then identify the factors effectuating customer churn such as bad network quality and customer service, (3) once the CLV and churn effectuating factors are identified, the ultimate VaR is measured and projected into future, (4) finally, the actions required to mitigate the projected VaR, either by improving network quality and coverage or by launching inbound campaigns, are specified.

This presentation would give a brief explanation of the use-case(s) in each of the aforementioned steps with sole focus on the business value though I intend to give some scientific details as well so as to keep the curious data scientists indulged. The aforementioned use-cases are briefly summarized in the steps below:

1. Customer Life-Time Value (CLV)
This section will give a brief explanation of how we have measured the lifetime value of our customers. This gives us the expected loss per each potential churn.

2. Quality of Experience of Customers (QoE) in Telecom
Measuring customer satisfaction by mapping it to network parameters (throughput, packet drop, etc.) is such a challenging task that a very few telecom operators are successfully managed to implement. I will present a brief overview of our customer management system which measures the customer satisfaction in terms of the network service delivered.

3. Customer Churn Modeling
A ML based model that incorporates QoE and several other factors to forecast potential churners and hence expected VaR.

4. Actions to mitigate VaR
Back tracking the factors identified by the churn model, actions are performed to restrain the potential churners, respectively. Out of several possible next-best-actions, I will present only the value-base-roll out which is about network upgrade strategy fueled by the bad QoE of the customers.

In conclusion, I will give a passing-by reference on the efficient use of ML and probabilistic algorithms and how should the results/accuracy/efficiency of such algorithms be presented to the senior management such that the underlying business value is marked.

Presentation Video


Dr. Wasif Masood
Data Scientist
T-Mobile Austria
I am an employee of Deutsche Telekom AG, working as a data scientist for both commercial and network related use cases. I have profound experience of designing and implementing both analytical and machine learning algorithms in Apache Hadoop ecosystem. My interest in data modeling started six years ago when I got the chance to work on experimental data during the pursuit of my PhD degree. Unlike structural modeling where the true nature of the data generating process can be modeled in close form, majority of the processes in real world are too complex to be understood in their entirety. Consequently, I gained expertise in several parametric models such as, dynamic stochastic models, time-series analysis and the state-space modeling. From the beginning of my career as a data scientist at T-Mobile Austria, I have made adequate use of machine learning and applied research in market science and mobile network, which led to several data science projects with attribution to high business value. Since industry demands the end-to-end working solution and not just a prototype, I have mastered several programming languages and have served as the data engineer for most of my use-cases as well. A brief list of my prowess and skills along with the projects and publications can be found on my linkedIn profile at: