Introduction

This report is part of a series exploring how telecoms operators can leverage artificial intelligence (AI) to improve their business operations, from customer experience to new services. Previous reports on AI in telecoms include:

• AI in customer services: It’s not all about chatbots
• AI on the smartphone: What telcos should do
• AI: How telcos can profit from deep learning

This report explores the applications of AI for network operations, detailing the prerequisites and stages to implementing AI and automation in networks, real-world examples of what some telcos have done already, and their potential value across different application areas.

Enter your details below to request an extract of the report

We divide the applications for AI in telecoms networks into three main categories:

• Fault detection, prediction and resolution: speeding up the process of identifying and resolving network faults, including predictive maintenance. This also includes identifying and mitigating network security risks, although security is a highly specialised field that merits its own report, so we do not cover it in detail here.
• Network optimisation: optimising the use of network resources to mitigate the impact of network faults and adapt to or anticipate changes in demand. This is also the foundation for automated service provisioning in software defined networks, while insights on network usage and traffic could be valuable for new service creation.
• Network planning and upgrades: optimising new infrastructure planning as well as the transition from legacy to next generation network solutions.

The first area is critical for all telcos, since service impairments are an inevitable element of running a network. The second is of immediate value for telcos that are still in the process of expanding existing network coverage and density, since it can enable operators to use their existing resources more efficiently. However, it is also increasingly tied into the first area of fault detection, since a large part of the fault resolution process is finding ways to re-route traffic from underperforming to underused assets, a process that is made easier with the adoption of SDN and NFV – processes can only be automated if they are software-based.

Compared with the first two categories, using AI for smarter network planning and upgrades is a nascent field. This is partially because many Tier 1 operators, who are leading the charge in adoption of AI elsewhere in network and business operations, completed the bulk of 4G deployments and have not yet fully embarked on 5G deployments. However, this report also looks at some innovative applications of image recognition models for network expansion in emerging markets.

While most of the data used for training and informing AI systems across network operations comes from operators’ own networks, telcos are also beginning to tap into new data sources to further refine their decision-making, such as using drones and image recognition to inspect towers, weather patterns and social media data.

Laying the foundations for AI in telecoms networks

Before jumping into how telcos are implementing AI for fault detection and resolution and in network operations, it is important to clarify what we mean by AI, and lay out the pre-requisites for any meaningful use of the technology.

What counts as AI? From automation to advanced AI

The term AI is nebulous – everyone has a different definition for it. Is it when a computer can make a faster, more accurate decision than a human?  Is it when a process is fully automated? Is it when the computer learns and continuously improves its decisions in real-time?

Wherever people draw the line between manual processes, (big) data analytics, automation and machine learning (ML) / AI, no company goes directly from manual to AI in one go. The transition is gradual. In this report we therefore use a broad definition of AI in this report, as outlined in Figure 1.

Figure 1: Not all AI is equal

Source: STL Partners

Two transitions are happening in parallel as operators move from left to right on Figure 1. First, there is a shift towards increasingly intelligent analytics techniques, from rules-based automation, where policies outline if-then sequences of actions for the computer, to machine learning supported automation, where models are trained to fulfil an intent (a goal) based on guidelines from experts and historical data.

The second transition that occurs in the move towards more sophisticated AI systems relates to decision-making. In rules-based automation, computers don’t have any decision-making power, they can only take pre-defined actions in specific circumstances. Making the transition from telling computers how to do something to what you want them to do means giving computers decision-making power. Telcos can do this gradually, by requiring humans to verify and approve recommended decisions before they are implemented. But in the promised future 5G and ‘sliceable’ networks, human approval for routine decisions would require more network engineers than operators could profitably employ, or drastically slow down network operations. This is not just a technical issue for telcos but also a cultural one that demands clarity from management teams on the evolving role of network engineers.

Contents:

• Executive Summary
• Making the shift from manual operations to autonomous, intelligent networks
• Recommendations
• Introduction
• Laying the foundations for AI in telecoms networks
• What counts as AI? From automation to advanced AI
• AI works at two levels for network operations
• Data: The bridge between rules-based automation and ML
• Fault detection, prediction and resolution
• What is it worth?
• How does it work?
• Real-world example of a recommendation model: AT&T Tower Outage and Network Analyzer
• Next step: From fixed to self-learning policies
• Optimising network capacity
• What are self-optimising networks worth?
• Use case overview
• How to do it
• From self-optimising to knowledge-defined networks
• AI for network planning
• Telefónica case study
• Driving automation internally versus partnering with vendors
• Reasons for developing solutions internally
• Reasons for partnering with a vendor
• Vendor profiles
• How AI fits with SDN/NFV
• Conclusions and recommendations

Figures:

• Figure 1: Not all AI is equal
• Figure 2: Rules-based automation versus machine learning
• Figure 3: A snapshot of rules-based automation versus machine learning
• Figure 4: Overview of automation and AI in network operations
• Figure 5: Telemetry is faster and uses less compute power than SNMP
• Figure 6: Elisa growth of automated trouble ticket handling
• Figure 7: Tupl results for automatic customer complaints resolution AI platform
• Figure 8: Implementing fixed policies for fault detection and resolution
• Figure 9: Visualisation of network alert clustering tool
• Figure 10: A self-healing network
• Figure 11: Elisa self-optimising network results
• Figure 12: Elisa maintained flat capex intensity throughout 4G deployment
• Figure 13: Finland 4G network performance, August 2018
• Figure 14: Self-organising network example use cases
• Figure 15: Numerous applications of machine learning and AI for 5G networks
• Figure 16: Break self-optimising networks down into mini loops
• Figure 17: The knowledge-defined network
• Figure 18: Facebook TCO savings over traditional multilayer planning
• Figure 19: Telefónica image recognition for network planning
• Figure 20: Ciena Blue Planet overview
• Figure 21: Google SDN layers
• Figure 22: Overview of cross-industry initiatives relating to network AI and automation
• Figure 23: Telefónica network automation roadmap
• Figure 24: Overview of SK Telecom Advanced Next Generation OSS (TANGO)

Enter your details below to request an extract of the report