AI in telecoms: categorizing AI applications in network operations
The categorization of artificial intelligence (AI) in telecoms can be cut any number of ways. Spanish telecoms group Telefónica identifies four business needs: in business optimization, market insights, customer engagement, and business innovation. Francisco José Montalvo, director of its ‘fourth platform’ business, says these combine as a “critical pillar to build a smarter and more cost-effective network” and to “change the relationship with our customers.”
Positive Technologies, which uses machine learning (ML) techniques in telecoms security, also describes four types of usage: for consumer analysis, networks management, diagnostics and troubleshooting, and information security. “It can be applied for many other business areas as well,” remarks Dmitry Kurbatov, the company’s head of telecommunications security. “The number of activities that can be optimized is enormous and promising.”
Aria Networks, a specialist in network optimization, describes three: in customer-facing applications, for churn and fraud prediction; in network-facing applications, for predicting faults, failures and breaches; and in business-focused applications, for margin analysis and demand forecasting.
What is clear is the majority of applications that have already been deployed in carrier businesses are customer-facing, rather than network-facing, invariably linked with care, marketing, and sales functions. “It is easier in these customer-facing functions to close the loop – to let a machine handle everything,” comments Dimitris Mavrakis, research directir at ABI Research. Critical infrastructure is not in play in these instances – just easily categorized lines of communications with customers.
In particular, Mavrakis points to the use of natural language processing (NLP) in customer-facing chat-bots, where auditory and textual synthesis gives the impression human representatives are online. “That is the most popular use case for AI.” Mavrakis lists the usual suspects among tier-one brands, in no particular order: “Orange, China Mobile, SK Telecom, Deutsche Telekom, Vodafone, Telefónica – they’re all using chat-bots,” he says.
Some claim to be solving up to 60 percent of customer enquiries without human intervention, he suggests. But there is little in the way of recorded uplift beyond the resolution of calls to automated care lines. “We can’t yet say what effect there has been,” he says.
Orange is more explicit about the improvements it has achieved. The marriage of big data and AI is a “huge” affair, it declares.
“The gains from mixing these are significant at a group level,” says Luc Bretones, executive vice president of the France-based group’s Technocentre and Orange Fab businesses, established as pipelines for new services and start-ups, respectively, and both focused on self-fulfilling innovation in fields such as AI. Its net-promoter score (NPS) for apps integrated with some self-service functionality, even where final actions are taken by humans, is up by “10-20 points”, it reckons. “The back office is becoming the front office,” says Bretones.
AI-led voice assistants are also readily available in the consumer market. Both Orange and Telefónica have launched services, Djingo and Aura respectively, to go up against the likes of Amazon’s Echo and Google’s Home devices. But these NLP techniques are relatively easy, and rather commonplace. “The way the industry is using AI from a front-office point of view is completely the same as finance or insurance,” remarks Robert Curran, marketing director at Aria Networks. “That’s not interesting from a telecoms point of view – it’s only when you step into the way this business works that it gets more specific.”
Besides the application of NLP and automation in care scenarios and gadgetry, the telecoms version of AI is also deeply ingrained in sales and retention, designed to analyze usage patterns and customer segments to direct promotional activity and head-off technical issues. “We have been able to improve the performance of traditional business intelligence models to predict churn, generate product recommendations, and reduce our risk in commercial operations,” comments Montalvo at Telefónica.
There is a clear logic to the industry’s pattern of deployment with AI. “First you try to sell more, then you try to optimize your costs,” says Kurbatov. “The most common usage lies in the areas of sales and marketing – to analyze and understand customers to predict their requirements and pre-empt difficulties.” Network-focused tasks like diagnostics and troubleshooting, management and optimization, and information security are like “supporting activities”, he says.
But these parallel tasks get to the heart of telecoms. Operators have two choices when applying AI in their systems: to deploy machine-learning models at the core of their networks, invariably in the data center, or at the edge, in the operations center or the base station. Most customer-facing use cases retain AI at the core, in order to compile a centralized view of the data; applications that seek to bring intelligence to network operations rather deploy AI at the edge of the network.
This last part is linked at the deepest level with the nitty-gritty of essential infrastructure, and promises to deliver the most dramatic transformation, impacting every department. Network maintenance and design is getting smarter, notes Telefónica. “We are completely reviewing our network planning model to optimize our cap-ex efforts through AI and ML processes, being able to achieve significant cost efficiencies in one of our most relevant budget items.”
Click here to register for the Enterprise IoT Insights webinar on March 21st on the developing role of AI in telecoms. Look out for the editorial report, ‘Artificial intelligence and machine learning: making IoT work for telecoms’, also published March 21st.