Looking back at 2020: AI as a catalyst for transformation in TM&E
No one could have foreseen where we currently stand in 2020. But as Sun Tzu wrote in his ancient military treatise, “in the midst of chaos, there is an opportunity” (The Art of War). The Telecom, Media & Entertainment (TM&E) sector has proven its agility, resolve and flexibility by emerging stronger from its downturns in the COVID-19 pandemic. Artificial Intelligence (AI), complemented by big data and cloud-native technologies, has been at the forefront of this turnaround. A near-50% growth in AI investment among the US-based Telecom companies during the outbreak is a testament to its significance.
The industry has seen more innovation and digitalization in the past few quarters than perhaps in the last several years. With technologies like 5G networks, edge computing, IoT, augmented / virtual reality (AR / VR), eGaming etc. gaining prominence, TM&E is at the cusp of another tectonic shift in market dynamics and competitor landscape. The parallel advancements in open-source solutions (e.g. TensorFlow, FastAI, RAPIDS.ai etc.) and the range of services provided by major cloud providers like GCP, AWS, Azure etc., implies a pivotal role for AI in the wide-scale adoption and value realization of these offerings. As the share of these investments continues to rise, the TM&E companies that are quick to harness its growth potential will be better placed to cope with fluctuations in demand, variations in supply chain functions and shifts in consumption.
AI trends in 2021: Complementing 5G to drive acceleration at scale
In light of the above disruptions, it is imperative for business leaders and data science practitioners in TM&E space to understand how AI may influence fundamental changes and make better-informed decisions for the future. To view this in greater details, we shall focus on five key trends which are likely to evolve in the next 12-15 months:
- Technological innovation triggered by Edge AI & 5GWith North America, Europe and East Asia at the forefront, 5G adoption is expected to increase by nearly 200% in 2021, with a potential to unlock $4.8 trillion globally. Its combination with AI is likely to transform industry operations and usher in a new phase of the IoT revolution. According to leading telecom infrastructure provider, Ericsson, AI-powered RAN (mobile communication system) enables up to 25% more 5G coverage with advanced traffic management. With high-speed wireless connectivity becoming mainstream, intelligent devices / services driven by Edge AI technology, (or Intelligent IoT), shall gain prevalence. This means a large volume of data shall now be processed closer to the source, either locally through on-device processing or via cloud over low-latency 5G networks. For the AI industry, this has two major implications. One, complex Deep Learning models shall now be trained and deployed on small physical infrastructure via decoupling of hardware / software stacks. Second, an increase in the importance of automated scalable deployments through modularized AIOps / MLOps pipelines, inter-platform portability (containerization), version control and monitoring. These two factors are also likely to drive a wide adoption of latest hardware technologies like Google’s Edge TPU – a purpose-built ASIC designed to deliver high-performance AI at the edge.
- AI-enabled real-time hyper-personalization at scaleAs the 5G revolution gathers pace, matching the customers’ ever-increasing need for personalization in almost every dimension will be the difference between success and failure in the TM&E domain. The companies shall be pushed to deliver a unique immersive experience to each customer across every product, content, interaction and even advertisement. With petabytes of data generated via a multitude of interconnected platforms / devices / sensors (websites, social media, OTT platforms, IoT devices etc.), organizations will have the fuel to realize the true potential of AI and pro-actively engage with millions of customers at a one-to-one level. Powered by the development and adoption of interpretable deep learning models (e.g. TabNet) and cloud resources (serverless, streaming, parallel processing, TPUs etc.), businesses can be expected to gain 4-10X incremental engagement over prior solutions. A direct application of such solutions can be in taking AR / VR technology to the next level of customization and provide personalized experiences across gaming, content viewing (sports, movies), music, online shopping etc. with much lower 5G-latency.
- Customer service digitalization via Transformer AI’sAccording to a recent study in the UK by Salesforce Research, only 4% of emails and chats are currently handled completely by chatbots while 5% of emails and 20% of chats are dealt with by human-assisted AI. The opportunity cost is particularly massive for large TM&E companies, which manage 100k+ interactions every week. With COVID accelerating the transition to digital-first service and rapid breakthroughs being made in Transformer AI (e.g. BERT, GPT-3) – a self-attention based Deep Learning model in the NLP domain, the role of intelligent bots automating tasks like translations, sentiment analysis, search query contextualization etc. are projected to increase. Studies have shown that adopting AI solutions and pairing it with omnichannel cloud communication i.e. aggregation of voice, video, messaging etc. within one platform, can reduce costs by up to 20% along with a significant increase in ‘first contact’ preventive resolutions.
- AI-powered solution in cyber-security & anti-piracyCyber-security and content piracy have long been a pain-point for TM&E companies. The piracy rates in the US had gone up by more than 40% during the COVID pandemic while the cost of malware attacks among telecom companies increased by 42% to nearly $900k per attack in 2020. The latest research proposes applications of Deep Reinforcement Learning (DRL) security methods to improve cyber-physical systems, autonomous intrusion detections, and multi-agent game-theoretic simulations for defense strategies. On the other hand, anomaly detection algorithms can be applied in network traffic monitoring in combination with malicious website identification through web-scraping (taking legal considerations into account) and content watermark detection (using latest CNN architectures) to counter online piracy. Software tools powered by such AI solutions connected over low latency networks shall now be able to respond to critical events, security incidents etc. and act on the network changes, address patching, and predictive governance in real-time – all without any manual intervention. With the loss of revenue due to fraudulent activities on the rise, 2021 is expected to witness a significant boost in such AI-based preventive measures.
- Adopting Ethical AI principle in model developmentAt present, more than 80% of CEOs across industries, believe that AI needs to be explainable for it to be trusted. On the other hand, only about 35-40% of enterprises have a detailed understanding of how and why their systems produce outputs that they do. This is especially true for TM&E which gathers information on hundreds of millions of customers worldwide. Moreover, as the volume of data increases exponentially, regulatory oversight will become more critical. Businesses will need to assess the risks and challenges for every data science project including its social impact, privacy, bias, and transparency. In a significant step towards this direction, a recent paper by Begley et al., 2020, introduces a meta-algorithm for fairness-imposing perturbation on an unfair model. It provides a new perspective on model fairness, flexibility, and stability, without compromising the effects of training-time interventions. Development and adoption of such Ethical AI solutions / toolkits, in sync with guidelines published by international bodies like IEEE SA, OECD and EU Commission will be the key to maintain the trust of both business and its consumers.
How to get there: Identifying roadblocks & planning for the future
There is little doubt that navigating through such periods of rapid change will be challenging. This will be especially true for TM&E organizations that are still catching up in AI and infrastructural maturity. The need is to invest ahead of time and mount cross-functional collaborations to drive transformations at scale. It is, of course, easier said than done and calls for an in-depth evaluation of the challenges to define a clear roadmap leading up to 2021.
To overcome these roadblocks in AI adoption across the TM&E space, we have designed three key strategic recommendations which may be adopted at enterprise-level. While the list is not exhaustive, adopting these measures can go a long way in preparing businesses for the near future:
- Adopt AI across every core operationLarge TM&E organizations operate across many dimensions – Technology, Broadcasting, Engineering, Marketing, Supply Chain, Retail and even Financial Services. The need is to embed AI at the core of all its operations. A fundamental step would be to enable cross-collaborations across data science and business units – a shift away from in-silo developments. This can help reimagine and optimize key processes to deliver more value. E.g. the Broadcasting functions are typically managed disjointly through manual processes by business users, which often leads to suboptimal planning and inefficiency in terms of effort and cost. AI-led innovation can deliver significant uplift in content planning, ad optimization and investment decisions while enabling automation of critical manual tasks in real-time.
- Invest in data & cloud infrastructuresWith the explosion of IIoT data, companies need to have proper data / model governance in place along with scalable infrastructure and standardized pipelines (AIOps / MLOps). Cloud technology, specifically Hybrid architectures, can help organizations strike the right balance between on-Prem solutions and on-cloud flexibility. Our industry research indicates a large majority of TM&E players are either at the inception or in the middle of such tech-migrations. The right combination of skills, experience and know-how across on-Prem and cloud-native ecosystems will be critical to accelerate these journeys, understand / overcome the potential roadblocks and ensure high return on investments.
- Focus on actionable & explainable AIOne of the major concerns around AI has been its ‘black-box’ approach. It can often lead to lack of trust from business teams, users and end-consumers. This is especially important when it comes to diverse range of marketing activities and services undertaken by TM&E corporations. E.g. unexplained bias in offering loyalty discounts or TV-package upgrades to specific groups of customers. Adoption of explainability and business decisioning at every step can significantly boost productionalization effectiveness. It can also help in moving towards a more customer-centric approach. Some key steps in that direction may be setting up an internal Ethics committee headed by a Chief Ethics Officer to ensure fairness, explainability and governance and / or institutionalizing ‘design before develop’ mindset across any AI initiative.
- Nguyen, T & Reddi, V (Jul’20). Deep Reinforcement Learning for Cyber Security.
- Begley, T, Schwedes, T, Frye, C & Feige, I (Oct’20). Explainability for Fair Machine Learning.
- MIT Technology Review Insights (Sep’20)
- European Parliamentary Research Service, Scientific Foresight Unit (Mar’20)
- Salesforce Blog (Oct’20)
- KPMG India (Oct’20)