To meet ever-rising customer expectations, operators need to increase the intelligence of their network operations, planning and optimization. Machine learning (ML) and artificial intelligence (AI) will be key to automating network operations, optimizing the customer experience and innovating new revenue streams.

With 5.5 million new ‘things’ connected to the Internet every day contributing to anywhere between 50-200 billion connected devices by 2020, the data output has already passed the point where it can processed manually.

“This data must be qualified and processed into usable information,” says Patrice Slupowski, Orange’s Director of Digital Innovation. “To achieve this, the use of AI seems crucial. Tools such as algorithms, or even learning mechanisms, like deep learning, will allow us to process, analyse and ultimately make the best use of this bulk of data.”

Telcos can use ML/AI in many different areas, ranging from customer service to the planning for fibre-optic rollout, network planning and traffic management. AI can be used to ensure customers are receiving consistent quality of service by allocating resources where needed, and fixing problems. It can also be used to identify malware, hacks and suspicious user behaviour.

Operators are evolving towards future AI-driven network operations. Here are some applications.

With the rise and rise of messaging apps and AI digital assistants, the way we interact with media, machines and each other is changing. Consumers are increasingly open to interaction with Siri, Alexa or Google Assistant. It’s something operators are making a major area of focus in the quest to serve customers better.

A consistent customer experience means being always being up to date with the latest services and offers, always resolving similar queries using proven resolutions. In the past, this has been a bit of a pipe dream, as bots simply did not have the capability to effectively engage with humans. With advances in AI technologies and industry specialisation, these challenges have been overcome.

For example, the Amdocs SmartBot, in partnership with Microsoft, can engage customers over multiple channels (Messenger, Kik, Skype, Slack, SMS, IVR, Amazon Alexa, etc.) and is pre-trained on telcos’ business processes and telecom-specific intents. Microsoft says this “radically shifts” the customer contact experience, “bringing consistency to customers and reducing the huge burden on telcos to staff and manage Call Centres.”

Vodafone has been trialling voice authentication services for over a year, expanding the capabilities of its chatbot TOBi. Voice authentication is intended to make it easier for its customers to verify their identity, access their details quickly and receive an account update through Amazon’s Alexa.

Vodafone is developing TOBi to be able to handle more account transactions by the end of the year. The chatbot, which is a blend of AI tech from IBM Watson and LivePerson, can already offer advice on SIM-only price plans and answer account-specific questions such as those to do with roaming.

“From an internal standpoint the move to voice is to do with fixing problems. It wasn’t a tech question it was a business question – what can we do to help our customers?” explains Neil Blagden, COO/director of digital & commercial operations at Vodafone UK.

Deutsche Telekom’s chatbot Tinka is used on T-Mobile Austria’s website, principally to aid consumer search. On a monthly average [at May 2018], ‘she’ chats with about 60,000 callers, answering 120,000 questions in the process. According to DT, Tinka is able to handle about 80% of all questions put to it including supporting customers in setting up LTE-based home Wi-Fi networks and explaining how to insert SIM cards into phones.

Tinka will soon be able to recognise customers and greet them by name, remember previous conversations and be able to refer to them. It will also be able to know – subject to customer consent – what a customer has already viewed on the Austrian T-Mobile website. This will make interactions with her much more specific and more focused on customers' individual concerns.

“From an internal standpoint the move to voice is to do with fixing problems. It wasn’t a tech question it was a business question – what can we do to help our customers?”
- Neil Blagden, COO/Director of Digital & Commercial Operations at Vodafone UK.

Deutsche Telekom's operations in Germany have been using a digital assistant to answer caller questions on selected topics since 2016. Customers interact about 50,000 times per month with ‘him’ – regarding such matters as SIM cards, smartphones or Wi-Fi networks. Upgraded with an AI system based on IBM Watson to automatically learn and refine its ability to handle communications, Deutsche Telekom plans to introduce a smart speaker and digital assistant software. Activated with the command ‘Hallo Magenta’ it’s designed to serve as a hub to connect with its smart home, TV and other online services. DT has also integrated the Amazon Alexa digital assistant with the device, so users can opt for that system rather than DT’s software.

While the aim of AI-powered voice assistants is to understand natural speech patterns to simulate human conversation and responds to customer queries just like a regular advisors, the key is to retain the human element.

“TOBi’s great for dealing with simple transactions, but if the bot gets stuck halfway through for any reason, that seamless transition to a human advisor is vital,” says Blagden.

To avoid frustration, the technology uses a ‘sentiment’ function to pass users onto a real advisor if the bot can’t help or if they are not satisfied.

Agents, bots or other self-service capabilities are still far from being able to replace real human service agents in addressing complex concerns because they are incapable of empathy, a key ingredient in any top-quality service.

“The trick for us is that the assistant does not claim to be something it is not,” says Susanne Lebkücher who tests AI developments for Deutsche Telekom. “It should show understanding without pretending to display empathy. ‘I can understand your frustration’ is not acceptable – whereas ‘I can help you with this’ is. One customer requirement which we repeatedly hear in our conversations is particularly important to us: Users should be clear from the outset who they are dealing with.”

Advocating the role of humans in digital development is central to efforts at Orange. “The anticipated progress around AI is generating many expectations and promises, but also a few delusions,” says Slupowski.

“Today, we’re dealing with so-called ‘weak’ AI. It’s able to problem-solve in four main areas: understanding language, advanced data analysis to identify links, real-time data classification and winning games with defined rules, like chess. Humans are still relevant in many ways, and will be for a long time. This is particularly true in terms of our ability to express and recognise emotions.”

“The trick for us is that the assistant does not claim to be something it is not”
- Susanne Lebkücher, Deutsche Telekom

Any process today that is dependent on human intelligence to execute, represents an obstacle to scalable growth. Perhaps few parts of a telecom operator’s business are more dependent on human expertise than the planning, design, and day-to-day management of the network.

Increased complexity in networking and networked applications is driving the need for increased network automation and agility. AI and ML approaches are beginning to emerge in the networking domain to address these challenges. Network automation platforms such as Open Network Automation Platform (ONAP) should incorporate AI techniques to deliver efficient, timely, and reliable management operations.

“ML and AI promise to reveal new insights from network telemetry and flow data, enabling operators to predict capacity demands and scale their networks appropriately,” observes James Crawshaw, senior analyst, Heavy Reading. “These new techniques will add a layer of ‘intelligence’ to today's state-of-the-art network management and automation toolsets.”

For every telco, network agility is a prerequisite for sustainable success. Operators able to instantly scale capacity and performance to demand will take the lead. The promise of virtualization is the ability to make rapid, instantaneous change to networks and the freedom to expand required resources without limitation.

AT&T, for example, began automating its network five years ago since deploying hundreds of virtual network functions running on its network cloud and managed and orchestrated by its own software. It aims to virtualize 75% of its target network by 2020.

“For most operators, network planning and design remains a largely manual process, run by smart, experienced engineers using spreadsheets and whiteboards,” says Robert Curran, CMO, Aria Networks. “It’s complicated and difficult work. But with networks undergoing constant change, it’s a risky strategy to rely on human design alone to ensure no such single points of failure arise.

“Automation is very much on the agenda for telcos but it must come with the sort of intelligence that prevents hidden vulnerabilities from creeping into the network. That’s an even more complex computational problem – which makes it a great use case for AI techniques.”

“It’s complicated and difficult work. But with networks undergoing constant change, it’s a risky strategy to rely on human design alone to ensure no such single points of failure arise.”
- Robert Curran, CMO, Aria Networks.

The financial benefits to the operator of Self Optimizing Networks (SON) can be substantial. Automating network operational tasks can reduce running costs while capex reductions are achieved by increasing the efficiency of network infrastructure assets.

“SON can predict when a site is likely to go down, can drive savings in infrastructure capacity and prevent revenue losses, which ultimately result in improved return on investment for the large capital investments telcos need to make in their networks and data centres,” says Microsoft digital strategist, Andre Truter.

Now SON is being linked with AI to create cellular networks that know the user, perform brilliantly and defy complexity.

“The race to build a self-driving cellular network is critical to the survival of existing mobile network operators (MNOs),” suggest Rethink Technology Research. “The rise in self-optimising networks, along with a huge increase in the total number of cells, will lead to a radical uptake of virtualised Software Defined Networks (SDN). This will make it possible to dial network resources up and down on-demand, but only if MNOs can cope with another level of network complexity.”

McLaughlin explains that the objective is to use big data, artificial intelligence and machine learning, to automate how Sky makes product recommendations for its customers. With the right approach, customers are happy and profits at the firm increase. These two outcomes can only be achieved if Sky makes effective use of the insight it generates.

“In the instance that we’re lucky enough the customer wants to share their data with us, we have the opportunity to understand them and think about what they need,” he says. “That doesn’t mean that we pre-determine the products they must have – it means we simply present them with content and ideas, and we’re constantly looking to increase the relevance of those offers. I refer to this approach as maximising serendipity – we’re looking to increase the chance that the things our customers see are going to be the things that they’re interested in.”

The financial benefits to the operator of Self Optimizing Networks (SON) can be substantial. Automating network operational tasks can reduce running costs while capex reductions are achieved by increasing the efficiency of network infrastructure assets.

“SON can predict when a site is likely to go down, can drive savings in infrastructure capacity and prevent revenue losses, which ultimately result in improved return on investment for the large capital investments telcos need to make in their networks and data centres,” says Microsoft digital strategist, Andre Truter.

Now SON is being linked with AI to create cellular networks that know the user, perform brilliantly and defy complexity.

“The race to build a self-driving cellular network is critical to the survival of existing mobile network operators (MNOs),” suggest Rethink Technology Research. “The rise in self-optimising networks, along with a huge increase in the total number of cells, will lead to a radical uptake of virtualised Software Defined Networks (SDN). This will make it possible to dial network resources up and down on-demand, but only if MNOs can cope with another level of network complexity.”

For example, says Diomedes Kastanis at Ericsson’s Innovation Office, “network ‘autopilot’ could detect the slightest predicted deviations from the optimal path and issue warnings to human operators long before actual problems emerge. Continuously collecting data and comparing predictions against reality will enhance accuracy, leading to better next-gen models.”

Telefónica teamed with Juniper Networks to develop such a predictive autonomous network for its customers across Spain. The goal is to evolve Telefónica's infrastructure into one that can self-analyse, self-discover, self-configure and, ultimately, self-correct.

This means that the detection and correction of network faults and anomalies will occur before they impact services and the customer experience. Similarly, the approach will help mitigate cyberattacks. It is expected to accelerate the speed of business and result in lower operating costs alongside improvements in security, reliability and resilience.

“The adoption of ML, AI and control systems will act to guarantee latency, speed and other relevant parameters are monitored, analyzed and addressed before they impact the performance of our network,” explains Joaquin Mata, CTO, Telefonica Espana.

Telcos can derive optimal maintenance schedules by comparing real-time information with historical data. Machine learning algorithms can reduce both maintenance costs and service disruptions by fixing equipment before it breaks. ML can, for example, already predict service degradations on cell sites up to seven days in advance. Nokia’s uses deep-learning to find patterns in unstructured data to guide engineers to the best solutions to network issues, leading to what it claims is a 20-40 percent improvement in first time resolution.

Rethink expects the rollout of 5G mobile networks to lead to huge shipments of automated SON tools. “The introduction of AI/ML will help operators automate their networks in a highly responsive way, to create what we are calling ‘the self-driving telco,’” the analyst says.

Standards body ETSI is developing use cases and requirements for telco network AI, which it will test in 2019-2020 ready for the first 5G self-driving networks.

“While we can see this is not an overnight process, because the technologies and processes are immature, the operators who start early will get the greatest benefit,” states Rethink. “For many MNOs, this will be part of a wider progress towards SDN and 5G, which will take place over as many as ten years.”

Barriers to integrating AI/ML into a SON platform to improve intelligent automation include a lack of internal skills, the perceived cost of building the AI engine, and the uncertain ROI.

When it comes to end-to-end automation, including core, backhaul, RAN and other elements, such as edge compute servers, and even some devices, there is even greater caution. In a Rethink survey of MNOs of January 2018, only 14% expect to have automated 60% of planning and optimisation processes by the end of 2020, and in 2025, 39% will still have automated fewer than 40%.

Planning for expansion of the fibre-optic networks will soon be carried out with the help of a special vehicle that, via various sensors and laser-scanning technology, gathers precise data about the environment. The AI/ML system, being piloted by Deutsche Telekom and research institute Fraunhofer learns to recognise landscape features, such as houses, grass, trees, etc., in terms of their planning relevance, incorporates available reference data such as street maps, and uses this capability to rapidly propose ideal routes for subterranean cables.

“While we can see this is not an overnight process, because the technologies and processes are immature, the operators who start early will get the greatest benefit”
- Rethink