Technology, in general, is continuously evolving, creating a challenge for CSPs to cope with the versatility demand of the market without exploring the latest technologies from both technical realization and business perspectives. Any new product, package, or service relies on an underhood technology that ultimately forms a base for CSP revenues.

Artificial Intelligence is not an exception. The shift of ISPs & CSPs to DSPs has opened doors for a wider portfolio that has been always limited to voice, Data, Connectivity services, basic VASs. Now, Business Departments (B2B, B2C, Wholesale, etc) strive to get new revenue streams with the aging of legacy services and the tough competition with OTTs, Startups, & Public Clouds.

During my class at MIT for "Artificial Intelligence: Implications for Business Strategy", I worked on the Mobile Operator case study and I assessed the areas where the AI can be deployed fulfilling the main business strategies. I am listing below some processes that could benefit from the deployment of Machine Learning as part of my assessment.

Machine Learning is one of the main streams of Artifical Intelligence in addition to Natural Language Processing (NLP), Robotics & other streams.

1.    Mobile Traffic Forecast

Traffic forecast in Mobile Operators is a process that defines the expected growth of traffic and its projection over 2-3 years. ML can study the data and inputs from previous years and build a model that predicts the traffic forecast helping the company to set the exact budget for expansions.

2.    Customer Support/Preventive Maintenance

Managing Customer complaints is the most hectic task in an operator organization. ML can learn the common traffic patterns and customers' behaviors to be able to dynamically identify any unusual pattern that can be considered as a fault or a potential problem and trigger other systems to perform preventive maintenance actions.

3.    Market Offerings

Marketing teams define what Marketing campaigns and what services to launch. ML can categorize the users based on their services and traffic trends and recommend the right marketing offers for each market segment. It is about predicting what kind of services these customers may like to have.

4.    User Experience Measurement.

User Experience is the main KPI of any offered service. ML can learn the traffic patterns of various streams such as Browsing, VoIP, YouTube, etc. building a model for each traffic stream identifying patterns that map to degradation in user experience such as the silent period in a VoIP call.

This is in addition to a wide variety of use cases in orchestration, automation, & Service assurance.

Touching the main business strategies, Marketing differentiation is where ML enables the DSPs to offer unique innovative services especially in the B2B domain where marketplaces and upsell recommendations are governed by ML.

The second common direction is the Focus on new segments and verticals and this is enabled by the wide variety of use cases that ML/AI brings to many verticals such as Medicine, Agriculture, education, etc. This will allow DSPs to have a broader B2B portfolio with a focus on specific verticals and more business opportunities as a whole.

The third aspect is the Cost leadership and that’s the promise of building a low-cost price model for the services offered with no profitability impact. ML is used to control the OpEx spending by enabling the preventive maintenance, Chatbot Support, etc. and the CapEx by properly predicting and defining the traffic forecast avoiding the over-dimensioning.

Also, understanding the market trends via ML helps to stop spending money on unsuccessful products thus focusing on the exact market requirements and trends.

when it comes to technology adoption, there are always waves, motivations, and enablers.

The 5G wave is thought to be the perfect timing to explore the ML Capability building the Future Networks including 5G.

While operators are focusing on 5G realization, more and more business cases are being projected and it is not a surprise, AI is one of the catalysts.

The ITU FG-ML5G, Focus Group on Machine Learning for Future Networks including 5G has started doing good effort in this direction drafting what can be considered as a logical architecture of Machine learning in 5G Network. The imposing of AI technology has been crafted in a way to be an overlay for existing technologies provided by 3GPP for example. This means that there should be a minimum impact on the underlying technologies.

Let's review the proposed Logical Architecture and the terms & definitions.

• ML Pipeline: A set of logical entities (each with specific functionalities) that can be combined to form an analytics function.

Each functionality in the ML pipeline is defined as a node as per the below definitions

• src (source): This node is the source of data that can be used as input for the ML function.
• C (collector): This node is responsible for collecting data from the src.
• PP (pre-processor): This node is responsible for cleaning data, aggregating data or performing any other pre-processing needed for the data so that it is in a suitable form for the ML model to consume it.
• M (model): This is an ML model. An example could be a prediction function.
• P (policy): This node provides control for an operator to put a mechanism to minimize impacts into place on a live network so that operation is not impacted.
• D (distributor): This node is responsible for identifying the sinks and distributing the ML output to the corresponding sinks.
• Sink: This node is the target of the ML output, on which it takes action, e.g., a UE adjusting the measurement periodicity based on ML output. 
• MLFO, Machine Learning Function Orchestrator: A logical orchestrator that can monitor and manage the ML pipeline nodes in the system. 
• Interface 8: A multi-level, multi-domain collaboration interface between nodes of an ML pipeline that allows the ML pipeline to be disaggregated and distributed across domains, e.g., edge and core cloud.  
• Sandbox domain: This is a domain internal to the network operator (NOP) in which machine learning (ML) models can be trained, verified and their effects on the network studied. A sandbox domain can host the monitor-optimize loop, also called the closed-loop, and can use a simulator to generate data needed for training or testing, in addition to utilizing data derived from the network. 
   

In the next article, I will start to put more practical use cases and define potential relations between the new logical functions and 3GPP/ETSI Standard Functional Blocks.

Please stay tuned!