PLATFORMS & SOFTWARE STRATEGY
ARTIFICIAL INTELLIGENCE: SEPARATING REALITY FROM HYPE
ARTIFICIAL INTELLIGENCE: LIVING IN A POST-MOBILE TELCO WORLD
As Communications Service Providers‘ (CSPs) business models continue to mature, the ever-present need to keep existing customers, attract new customers onto a CSP’s network and/or increase a subscriber’s Average Revenue per User (ARPU) continues to be the focus of attention for operators worldwide. In our view, in most developed markets today, the mobile services market has already peaked (in terms of subscriber penetration and service revenue growth) and any new revenue growth that may be experienced by service providers must come from new service delivery models and optimizing the customer experience.
In this post-mobile world, mobile has become a platform upon which future innovations are being built. The
Internet of Things (IoT) is one such innovation. Autonomous vehicles, Virtual Reality, Blockchain, Industrial IoT, the sharing economy, are all innovations that would not exist without the mobile platform.
What drives these innovations is the exponential generation and usage of data. Indeed, in our view, what will distinguish the CSP leaders of tomorrow from everyone else is how CSPs can leverage the data that passes through their networks to create economic value for their customers.
Today, Artificial Intelligence (AI) processes are perhaps the only way in which such large amounts of data can be
harnessed and efficiently utilized by CSPs. As we discovered during the course of our work, as of 2017, all the leading vendors in this space have a very rudimentary view on what AI means to them. To us, it means deep learning algorithms and cognitive computing.
Deep Learning involves automatic feature detection from data. AI techniques can be applied to a range of data types, including: network usage data, voice biometrics, images and sound, transactional data, sequences (LSTMs), text (natural language processing), and learning new behaviours and learning to behave autonomously (reinforcement learning).
THE IMPACT ON THE OPERATOR: THE NETWORK SIDE
Our research shows that, AI and Machine Learning (ML) will have a tremendous impact on CSPs. On the network side, AT&T‘s Domain 2.0 provides a vision of a network whose major functions operate through autonomous actions. With carriers already upgrading their network elements to software defined networks (SDNs) leading to increasing use of virtualized network functions and cloud-based services, we think that AI has great potential on the network side of a carrier’s business. AI can be used to predict when a system may fail, identify a network problem, take action autonomously within a certain policy framework.
Here, IoT and other transactional technologies (such as Blockchain) will create a very large number of end points on a carrier’s network. With 5G, the nature of the network will change, creating a large number of patterns which will not be easy to handle by even existing big data and high performance computing clusters. AI will be the only way in which a network operator will be able to manage its network functions efficiently.
The use of AI on carrier networks will impact operators in many ways: from traffic classification (flow identification, security, DDoS detection/mitigation, QoE) resource uitilization, fault management, network capacity
planning, to network orchestration, etc. – are all elements that will be impacted by the sheer volume of data passing through a network and carrier’s ability to manage and use this data.
THE IMPACT ON THE OPERATOR: THE CUSTOMER SIDE
Much of the analyst community has focused attention of AI/ML use cases in CSPs on the customer experience.
Here, so-called "AI" use cases can broadly be classified into the following categories:
Marketing / Customer Care: Pre-paid top-up optimization, bundle plan optimization, social network analytics for marketing, the 'one-customer' segmentation, proactive customer care for churn reduction, cross-sell / upsell products and services are all use cases that carriers are applying currently across the world.
Paid Media: Location-based advertising, mobile display advertising, direct marketing campaigns, large event
management are some of the use cases deployed by carriers globally that utilize big data analytics to increase
But are these true AI/ML-driven use cases in the telco or simply big data-driven statistical modelling exercises that
have been rebranded to be called AI?
During the course of our research, we asked this fundamental question to some of the leading vendors in the AI / ML space. It is clear that while there are some use cases that can be classified as true AI, that is machine-learnt autonomous actions undertaken by software code; in most instances, vendors‘ AI products are simply re-branded old big data analytics platforms with very little in terms of autonomous action.
To be fair, players in the industry have provided factorization machines, random forest techniques,
radiant boosting, regression analysis, and bayesian approaches to adaptive learning for some time now. However, in our view, this is not true AI that results in a certain action taken autonomously.
ARE CSPs READY?
Our research shows that we are at least 10 years away from CSPs implementing an autonomous network and autonomous customer relationship management business process. If analytics-driven dynamic pricing, CRM and network management with no manual intervention is the end goal, implementing such a scenario will involve changing or replacing both north-bound and south-bound systems already in place at telcos and a transformation of how telcos consume and use their customer data.
In our view, this will be a difficult proposition to undertake, even for CSPs that have full-stack vendors. Integration with third-party CRM, billing, order management, order capture, data analytics solutions can be a whole other complex set of implementations that CSPs will have to deal with if they go down this route of analytics / ML-driven dynamic pricing, customer management, and network management.
Another important issue for CSPs to consider here is that there are no (or very few) standards in this domain. Given CSPs’ tendencies to not veer too far away from established frameworks / models, this adds another layer of constraint on CSPs, should they want to move in this direction.