Telenor's lead data scientist, Simon Johannes de Wit, details the company's model approach to improving productivity in its analytical department and how it addressed the challenges it faced along the way.
Telenor takes a Model Factory approach to improving customer experience
Data science projects have the potential to bring huge improvements to network automation and customer experience as CSPs journey towards becoming Digital Service Providers. However, they are often hampered by data preparation (DataOps), which typically takes up 80% of the time, while only 20% is dedicated to modeling and implementation (AIOps). In a recent gathering of TM Forum's Global Architecture Forum (GAF), Telenor Norway described how it solved this challenge by implementing a “Model Factory”.
The idea behind the Model Factory is “to have control of the whole analytical lifecycle. So it’s a DataOps, ModelOps, DecisionOps approach, but a little bit extra," explained Simon Johannes de Wit, Lead Data Scientist at Telenor.
According to George Glass, CTO of TM Forum, the Model Factory is a concept that will enable CSPs to boost productivity in their analytical department. “It plays very much into the work of data science and offers a huge improvement around network automation and customer experience,” Glass said.
Telenor has been working with analytics and AI specialist SAS for a number of years, including on a strategy to change how it promotes, guides and sells its mobile services.
ModelOps provides a “framework for integrating different data sources and this needs to ensure that models are trained on high quality data … to generate accurate insights, and all of this then will result in enhanced customer experience," according to Sasa Crnojevic, EMEA Network Analytics AI&ML, Business Principal at SAS. "It will help CSPs to personalize the customer interactions, detect anomalies, fraud patterns, security trends, leading to improved time to market and overall customer experience, delivering greater value to the end users,” Crnojevic added.
In addition, it enables CSPs “to keep the consistency and continuity of the architecture approach” while also coping with rapid change of network demands, “because machine learning models can be used by CSPs to predict and manage network congestion …and improve network resiliency, but they can also be used on the marketing automation side,” explained Crnojevic. And its usefulness extends to business environments beyond customer experience.
“What I like about it,” said De Wit, “is that all the tools in the different phases of the analytical lifecycle are highly integrated to each other. So you have manage data, capture data, prepare data, then explore, and there you can also already make some models, but we have some specific tools for building models and then managing the models, and then make the right decisions. So, modeling and decisions are highly integrated.”
De Wit also noted that if you have a lot of models to build, a Model Factory facilitates the build, control and monitoring processes. Telenor currently has 69 models in production for consumers and about 12 for businesses. It has been helped by what De Wit describes as "a special approach to data preparation," which has allowed it to reduce the time spent on data preparation from 80% of a data science project to just 5%. The CSP achieved this by preparing the data in advance for all theoretically possibly models. “So that saves a lot of time. If marketing managers want to build a model, the data is already there,” said de Wit.