AI Transformation for Telecoms

The Scale of Global Telecoms

The global telecommunications industry is immense – the 54 telecommunications companies on the Forbes Global 2000 list accounted for more than $3.4 trillion in assets and totaled nearly $1.5 trillion in revenues last year. The industry has reached this scale through constant adaptation to new technologies and competition. Today, As telecoms look ahead, it is becoming increasingly accepted that there are two major technologies that they must embrace in order to succeed: 5G and AI. 

5G alone is expected to contribute $13 trillion to global output by 2035, with most of these gains occurring outside the US. According to PwC, AI could contribute up to $15.7 trillion to the global economy by 2030 – more than the output of China and India combined.  In order to maintain competitiveness, it will be incumbent on all players in the telecommunications industry to incorporate these technologies into their strategic growth plans.

5G’s moment may have already arrived with Apple’s recent announcement of the iPhone 12 with 5G. AI, for its part, is currently less readily associated with mobile and business telecommunications services, but we will show why these two technologies go hand in hand – a double wave of digital transformation that has the potential to unlock growth and cement market positioning for the most agile telecoms and enterprises able to take advantage of it.

Waves of Transformation

Global telecommunications leaders are no strangers to transformation. The industry began to change dramatically in the 1980s with the break-up of regional telecom monopolies under market liberalization in the US, Japan and the UK; and this transformation continued with the dawn of the Internet age in the nineties when the technologies of packet-switching, IP (Internet Protocol) and the world wide web came to the fore. Over this period, telecoms grew from legacy providers of circuit-switched voice communications services into the modern high bandwidth mobile and data services providers of today.

The process of transformation in the telecommunications industry has only accelerated in recent years as internet giants such as Google and Facebook and cloud providers such as Amazon and Microsoft now operate global services using their own hyperscale data centers built using commodity hardware and homegrown software. As telecommunications companies are forced to compete with these newer market entrants, their capital-intensive, technology-focused model has given way to a user-centric service delivery model enabled by a next gen infrastructure.

Modern telecoms today primarily focus on growing wireless revenue by increasing their subscriber base and ARPU and by retaining high-value subscribers. They furthermore seek to grow strategic revenue with various pay-as-you-go service offerings, and finally to reduce their overall cost structures. Once implemented, 5G and AI will provide telecoms with powerful new capabilities for accomplishing all of these goals.

In order to attract and retain high-value corporate accounts and increase strategic revenue, modern telecoms must offer their corporate customers a variety of innovative services. British Telecom, to use one example with particularly well developed enterprise IT offerings, offers its customers a fully-fledged enterprise communications and collaboration platform, vertical- focused customizations for the public sector and retail and utilities industries, and cloud and data security services. They also offer a platform for managing distributed workforces (based on software they developed and dogfooded themselves to manage their own tower maintenance operations).

But service offerings must continuously evolve in order to retain subscribers. As communication and networking technologies inexorably improve, telecoms must adapt and upgrade their infrastructure and procedures in order to provide customers with the latest services and unlock the additional value those technologies provide. Machine Learning Operations (MLOps) is a key elements of these upgrades.

Transformation in the telecommunications industry is never complete, but comes in well defined waves – some of which are well understood and expected and others which are only captured by the most agile and forward thinking. The next wave of transformation for the industry is 5G and we would argue that a no less important and closely-related upcoming transformation wave for the telecom industry will be AI.

From Swells to Waves – 5G + AI

5G will undoubtedly be a major step forward – up to 100x faster data speeds and network latency lowered by a factor of five will provide instantaneous access to services and applications. 5G envisions a heterogeneous and software-defined network that can integrate massive numbers of lightweight sensor nodes and a diversity of transmitting and receiving devices such as macro and small cells to provide pervasive connectivity indoors and outdoors, leading to data volumes many thousands of times higher than today.

Proposed use cases for 5G include industrial and consumer IOT, gaming, pervasive 2D and AR/VR media and social experiences, intelligent logistics, autonomous and smart vehicles and many more.

As the CEO of Ericsson Börje Ekholm recently said, it is likely futile to try to identify the killer app for 5G at this early stage. “By the time we identify the killer app”, said Elkholm, “it will actually be too late. Many looked for the killer app in 4G, however we were not able to envisage a world where a ride-hailing app would be the normal way of ordering transportation.” Shopping and watching movies on mobile devices was also hard to imagine when 4G LTE development was underway, he added.

The second major transformation wave confronting telecoms today is AI. AI represents a major opportunity for telecoms to improve their internal efficiency and overall competitiveness, to provide compelling value-added external service offerings as well as to facilitate their clients’ own AI transformation efforts. 

Internally, telecoms can and are using AI to optimize their networks, provide improved customer service, conduct maintenance operations and prevent fraud. Externally, telecoms are uniquely positioned to facilitate and extend the AI transformation efforts of their clients.

According to Boston Consulting Group, AI should be central to telcos’ transformation because it will enable them to better cope with fluctuating demand levels, adjust to supply chain disruptions and adapt to sharp shifts in consumer confidence and priorities. They recommend that fast moving telecoms should reinvent themselves by embedding AI at the very heart of not just their products, but also their key processes by using the technology to reimagine their existing operating procedures throughout the organization.

AI will further assist telcos to drive growth by providing digital services in areas such as healthcare, media, entertainment and third-party analytics – but they must be careful to avoid potential risks to customer privacy that may wipe out years of building customer trust.

Overcoming Data Privacy Issues + Unlocking IoT

To solve these potential data privacy issues of AI on mobile networks, operators may opt to implement a new framework called ‘federated learning’. Federated Learning allows for neural networks to be trained locally on the same device where the data is generated/collected. Once trained, the locally trained weights of the neural network are transported to the telecommunication company’s cloud data center where federated averaging takes place using techniques such as secure aggregation and differential privacy to further ensure the privacy and anonymity of the data, and a new model is produced and communicated back to the remote device. This technique will be particularly useful with health-related data to improve diagnostics without breaching the privacy of patients, but many categories of sensitive consumer preference and behavioral data will be relevant.

For less sensitive data, however, 5G could actually make edge computing largely irrelevant, as training AI systems in the cloud would be almost the same as doing the processing on-device. Centralized processing could even be the preferred option in most cases because of higher processing capabilities and less restrictive power budgets.

AI has further synergy with 5G in the context of IoT. The increased bandwidth and volume capacity of 5G has been hailed as a potential pivotal moment for IoT, supporting millions of low-power sensors on the network in both indoor and outdoor environments. These IoT data sources will feed into AI systems, generating future services and efficiency gains – and both centralized and edge AI processing will be employed to support that vision.

An additional option for AI and IoT applications that 5G will present to telecommunications companies is to do processing at the base station or edge router, so called small and large cells – essentially creating a middle processing layer between the cloud and the device known as fog computing. This architecture, while advantageous in terms of privacy and latency, will require significant specialized MLOps capabilities to manage orchestration of distributed computing resources and coordination of training and feedback loops on multiple networks.

AI for Internal Telecom Operations

AI will also be an essential technology for telecoms to manage the new 5G networks themselves. ML will enable seamless automation of network management to reduce operational costs and enhance user experience as traditional optimization techniques are not agile enough to handle the complex, real-time analysis required in 5G networks. ML at the edge can also be used to predict changes in demand and scale up network resources as needed. 

Already, telecoms globally are using AI technologies to optimize their networks, 

allowing for predictive maintenance, self healing and protecting networks from any fraudulent activities. AT&T is combining AI and drone technologies by testing computer vision analysis of automated drone visual data for cell tower maintenance. Customer services and support, agent-guided assistance, personalized marketing, and product bundling recommendations are also being driven by AI. Dutch telecom KPN is already using NLP deep learning systems to analyze notes produced by their call center agents, and using the insights generated to make changes to its interactive voice response system.

Supporting Customer AI Transformation

The final, and possibly most important aspect of AI Transformation for telecoms we will cover is the competitive advantage telecoms can achieve over other technology infrastructure providers by facilitating AI adoption for their customers.

Telecoms have several advantages over hyperscale cloud providers and other non-telco competitors. First, telecoms have large numbers of existing B2B client relationships. Second, they are already entrusted to secure sensitive client data. Third, they are frequently local and regional specialists – with closer client relationships and more years of experience in the markets where they operate than competitors. And finally, they possess large existing infrastructure networks with massive capacity, providing them with an extremely low marginal cost basis for providing new high-bandwidth high-volume data services to customers.

On the other hand, it is also fair to say that the hyperscale cloud companies realized the importance of AI workloads for their strategic growth very early, adding massive amounts of ML-accelerated hardware and investing heavily into their own home-grown MLOps software to ease AI transformation. 

We contend that telecoms (and the systems integrators and hardware manufacturers who supply them) have a massive opportunity to transform their infrastructure and turn the tables on their growing cloud competitors.

In order for customers to take full advantage of the capabilities of 5G and AI, they will require familiarity with agile development procedures as well as significant MLOps expertise to ensure that ML pipelines can be quickly created and maintained. They will also require access to large pools of specialized ML training hardware (most commonly GPUs) that integrate seamlessly with existing storage for efficient AI training and deployment. 

To this end, it is essential that telecoms provide their customers with a fully fledged MLOps Solution, such as Neu.ro, that integrates with their own offerings, combines flexible resource orchestration for ML workloads on their mobile networks and in their clouds with integrated pipeline creation and management tools for the entire ML lifecycle, including data collection and preparation, experiment tracking, hyperparameter tuning, remote debugging, distributed training and model deployment and monitoring.

Automating infrastructure management with an easy to use code-first development environment such as Neu.ro will allow customer AI teams to optimize their infrastructure costs, streamline operations management and freely integrate with their choice of the leading open source and proprietary tools. These capabilities will allow customers to fully take advantage of the combined potential of AI and 5G and guarantee telecom growth and market share for many more upgrade cycles to come.