Learning to Take Advantage of Machine Learning

The unfolding advancements in big data technologies of the recent few years have boosted telco OSS/BSS systems especially for revenue assurance, fraud management, marketing, mobile money and compliance as true real-time data summarization and analysis becomes possible. Combine this with massive historical data processing, faster integration of new data sources (both structured and unstructured) and minimal dependency on hardware (thanks to the cloud) and we find that operators are finally getting what they have been wanting for a long time.

The ability to process data at these volumes and pace has led to the growth of new domains like data science and lately, machine learning (ML) and artificial intelligence (AI). ML as a theoretical concept has been around for some time, but it is only recently that practical technology has caught up with the concept and promise. All the major players – Google, Facebook, Microsoft, Amazon, and IBM – have released open source versions of ML engines, with Google and its Tensor Flow getting the largest mind share and media.

At the basic level ML can be described as using historical data to identify patterns that can be modeled, which then can be applied to new data to generate results. This is a process where the system continuously ‘learns’ from new data and updates the model to fine tune the results for increasing levels of accuracy. A common example is the kind of recommendation engine that retailers use to suggest products for customers based on their shopping history.

As the term describes it, ML is a learning process and is only as good as the training data that is used to teach the system. A primary requirement for ML is availability of very large volumes of data. Here telecoms has an advantage as operators generate large data on a regular basis and has significant amounts of historical data available.

Telcos have been using ML techniques for more than a decade using conventional technology including manual analyses, primarily focused on predicting churn and measuring customer experience. Churn prediction is a well suited problem for ML as historical data can provide information on an event that has happened and data can be analyzed to find patterns or identify relevant factors. The challenge was that by the time it took conventional technology to identify the patterns of churn newer patterns had emerged and predictions were off. But with mass processing capability that has changed as newer scenarios could now be included.

Researchers and data science practitioners have broadly classified ML into “supervised learning” and “unsupervised learning”. There are also models combining both these models. Within this broad categorization there are approaches independently developed for “categorical data” and “numerical data” and a combination of both. Churn prediction involves combination of unsupervised and supervised learning with both categorical and numerical data. But ML and AI has more to offer for telcos.

We can consider predicting churn as artificial narrow intelligence (ANI). Although it’s called narrow within the AI field, ANI can provide a tremendous boost for decision-making. My Commsrisk article entitled “Harnessing The Power Of Accurate Prediction” talked about the importance of descriptive analytics and explorative analytics in creating predictive analytics. Now I would like to talk about the benefits that telcos can obtain from ANI.

ANI has the potential to be a game changer. The telco mantra has been that they don’t want to be a dumb pipe; they want to add value added services and move up the value chain. ML and ANI provides that opportunity. Telcos have information on their subscriber behavior through calling patterns and social graphs. Operators can now leverage that information to provide more customized services as well as additional services to their subscribers. In the USA Verizon has acquired AOL, TechCrunch, The Huffington Post and recently Yahoo in addition to services for streaming video and music. Diversification of revenue is also happening in Europe, Asia and Africa as well.

Artificial general intelligence (AGI) and artificial super intelligence (ASI) are more complex and applicable for robotics, image processing and advanced natural language processing (NLP). An article by Eric Priezkalns in Commrisk titled “WeDo Uses AI To Judge Your Personality” was very interesting, showing how AGI has been put to use to judge personality. The results are not perfect, but there is huge potential for fraud management, customer experience management and for the delivery of new products and services. For example, imagine augmented reality maps with points of interest specifically chosen for you or your peer group.

MI and AI evolved over the past few decades and the sudden burst of progress made over the past few years predicts new and better solutions for known challenges. The message to telcos is to start thinking of how to exploit ML and ANI. These technologies are not sci-fi any more. They offer clear advantages both from the perspective of cost and functionality. Examples of the benefits include technology upgrades (VoLTE, B/OSS upgrade, change of vendor), capacity expansion (network capacity, distribution channels), price revisions and promotions. Also setting up data stores to process network and operational data with anonymized records will be a great source for ML solutions. I am currently working with operators to enhance their ETL capability to process years of data stored in their cloud archives in order to support ML initiatives.

To illustrate the value of ML, The Telecom Regulatory Authority of India (TRAI) has developed an Android app for users to test the network speed, packet loss, network delay information and provider ID. The user can also share the data with TRAI, and this would be a good candidate for ML and AI activities. I am not sure how TRAI is going to utilize such data but if TRAI systematically collects such information from the operator, the accuracy will be phenomenal.

I have only scratched the surface on this topic, as I wanted to start with a more general view of the technology. Operators should include these new advances into their technology mix and use any opportunity they have to introduce them into their business.

Daniel Peter
Daniel Peter
Daniel Peter is Vice President of Analytics at Gamma Analytics. He heads Gamma’s Data Science group working with customers in advanced predictive model development, business data analytics, data science, and product strategy. He also has significant expertise working with Fortune 500 companies for Connectiva Systems and Hewlett Packard.

Daniel has a Business Analytics degree from IIM Calcutta, Masters in International Business from Kedge Business School, France, and MBA from Loyola Institute of Business Administration, India. He is the author of: “Corporate Response to Recession (2008-09)”. He speaks and writes on telecom topics and can be reached at daniel@gammanalytics.com