Machine learning has seen rapid adoption in industries across the world. In our latest expert interview, we discuss with Dr George Bolt, Head of Analytics and Product Manager for Optimus RP (Revenue Protection) and Optimus DI (Data Integration) at Neural Technologies, the unique opportunity machine learning in telecommunication offers to reduce risk, improve financial resilience, and boost market competitiveness for telecoms operators.
Q1: Can you explain in simple terms how you would define machine learning for a business audience?
Machine learning is a range of modelling techniques that run on standard computers to solve business problems based on example data. The example data provides insight into actual cases that represent the business problem, and the solution is achieved exclusively through automated data-driven analysis.
This contrasts to software which is programmed with a specific coded computer program that implements a set of pre-specified instructions corresponding to how a business solution is achieved.
The data-driven problem-solving approach provided by machine learning technology enables business decision making to be performed on complex issues which are not easily solved using conventional software techniques. If a human software expert cannot define the sequence of steps/operations that solve the business problem, then such data-driven techniques are invaluable.
At Neural Technologies we’ve applied a wide range of machine learning technologies to effectively deliver business solutions in telecommunications, finance, insurance, banking, marketing, and even mineral exploration!
Q2: Why does this technology offer a particularly powerful opportunity for telecoms operators, and what areas of operations is it seeing greatest use?
Telecoms operators generate huge amounts of data in their internal systems and interaction with subscribers and other devices using their networks. As such, manual analysis of data is not feasible, neither is implementing standard requirements-driven software solutions—especially where a dynamic human interaction occurs.
As such, billing, customer relationship management (CRM), and point-of-sale (POS) are the most obvious areas of utilization. Equally however, network usage where behavior is often complex, and can even be designed to purposefully deceive, means that conventional software solutions can find it hard or even impossible to deliver effective business decision making.
Q3: How widely can machine learning technologies be applied in telecoms, and what are examples of some of the rewards that enterprises could unlock?
In examples such as CRM and billing, the benefit of understanding the context of subscribers in terms of factors including likelihood of churn or payment default are key to the bottom line of a telecoms operator. Equally, during application stages, accurate up-front identification of credit risk enables a telecoms operator to offer appropriate services to a new subscriber.
In terms of revenue generation, machine learning can be used to up-sell, cross-sell, or even offer services sold to businesses providing targeted advertising. These all areas where machine learning can significantly improve what might have been implemented based on a set of expert rules.
Q4: What are some of the challenges that must be overcome in embracing an effective machine learning solution?
Usable data! Although telecom operators produce huge amounts of data, the recognition of how information needs to be captured rather than just gathering raw data is not understood.
If the end objective of wanting to produce increasingly more effective business solutions driven by quality data are set clearly, then this can be the driving factor for how sources of data should be managed.
This goes beyond just business intelligence—where processed data is summarized to aid decision-making—to actively implementing machine learning solutions that constantly monitor, identify, and extract structures within data, and drive predictions to automate business operations.
Q5: How does Neural Technologies utilize machine learning in its own solutions to empower communication service providers?
Neural Technologies’ Optimus DI product has ActivML—our sophisticated machine learning technology—embedded at its heart. Optimus DI is our streaming data integration, transformation, and business orchestration solution platform, and offers the ideal point in which to embed machine learning and AI tools.
For fraud, revenue assurance, credit risk management, our Optimus RP (revenue protection) platform utilizes ActivML for a range of business problems. These include, but are not limited to, assessing credit risk of new applications for services, identifying fraudsters committing SIM box fraud but camouflaging their activity to avoid rules-based detection, categorizing dealer sales profiles to identify those who are not following required business processes, and identifying changes in traffic trends based on multiple characteristics. This increases the capability of revenue assurance to look across multiple data sources rather than one at time using traditional techniques.
Q6: How does Neural Technologies work to help integrate machine learning solutions for telecoms providers, and do customers need specific machine learning experts to deploy these solutions?
We have always placed great focus on providing products that are business-solution enabling platforms—Optimus DI and Optimus RP are both great examples. These adaptable platforms enable our customers to not only deploy solutions, but build their own effective implementation and continually enhance and widen their scope.
We have taken the same approach with ActivML, and it provides the same type of platform to enable our customers to employ machine learning without requiring dedicated machine learning experts—after all this type of skill is hard to resource worldwide!
ActivML is highly automated and requires no particular expertise in machine learning, just standard business analyst capability enabling good understanding of how data sources and content map to business processes.
Q7: How widespread do you expect machine learning solutions in telecoms to become over the next decade?
Machine learning and AI techniques are commonplace in other business verticals in comparison to their current deployment in telecoms. Banking, insurance, marketing, and other industries already have fairly wide-scale adoption.
Given the abundance of data in telecoms, and the increasing challenge of gaining market margin in a competitive business environment, we can expect the utilization of machine learning echo the growth patterns seen in other industries in coming years.
If your in-country competitor is able to better align their risk to bad debt or individual subscriber activity in a more optimal manner, then without these same capabilities you are going to find it harder to compete financially.