AI in telecoms: a game changer? Mark Newman, Chief Analyst mnewman@tmforum.org 2017 TM Forum 1
AI is the next big thing in digital Artificial intelligence would be the ultimate version of Google. The ultimate search engine that would understand everything on the web. It would understand exactly what you wanted, and it would give you the right thing. We're nowhere near doing that now. However, we can get incrementally closer to that, and that is basically what we work on. Larry Page, CEO, Alphabet 2017 TM Forum 2
And the FANGs are building up their armouries The cost of signing up a top AI researcher is comparable to signing up a quarterback in the NFL, Peter Lee, VP Microsoft Research IBM Here BAE Systems Rethink Robotics Magic leap Robot Accenture Adobe Huawei Booz Allen Oculus Cylance GE Intel Facebook Nvidia Microsoft Google Amazon Annual number of AI recruits Source: Paysa 0 200 400 600 800 1000 1200 1400 2017 TM Forum 3
So how should the telecoms industry respond? Based on a TM Forum survey of 187 executives from 76 communications service providers (CSPs) operating in 51 countries We re taking a wait-and-see approach 18% 29% We have started proofs of concept 22% We are working with suppliers who are building AI into some products and services 30% We have already built some internal AI expertise and are incorporating it into our product and service roadmaps 2017 TM Forum 4
How should a CSP organize itself for AI? Do we have centralized AI, one brain that controls all customer experiences and addresses internal operational needs? Chief Architect, European CSP Group. We need to look at bringing in an AI platform that you build on top of, rather than looking at point solutions. Because at the end of the day, you will have so many AI solutions in your environment, how are you going to control that? Head of Enterprise Architecture, south-east Asian CSP. We are trying to pull initiatives together but individual needs are very strong. Business needs often trump corporate efforts to centralize. Senior Consultant, central European CSP. My boss owns AI but in truth it s all over the place. We have a single data lake but our approach is to let ideas incubate across the organisation. CTO, North American converged CSP. 2017 TM Forum 5
How should a CSP organize itself for AI? Data Science and Cognitive Intelligence (DSCI) Data Global Supply Chain Strategy and analytics Big data and artificial intelligence systems Verizon Enterprise Data Analytics [ data management, data governance, data warehousing and data lakes, common analytical and AI technologies ] [ cross functional, cross organisational projects ] 2017 TM Forum 6
Customer experience is the main driver What is the main driver for AI in your company today? 48% 15% 37% 30% of CSP respondents said they have rolled out chatbots 71% of them have retained human agents who are ready to step in to deal with complex interactions 1 in 7 have redeployed some customer service agents to complete higher-value tasks Source: TM Forum, 2017 Reducing OpEx in the organization by driving automation and closed loop systems Delivering a better customer experience Upselling new products to existing customers 2017 TM Forum 7
But network is the bigger prize Does it make sense for customer solutions to be first users of AI? No. Why do customers call us? Not to tell us they are happy. They have an issue. It may be billing but is more likely to be a problem with the network, with an incident which is a result of a product or process failure. We need to offer hard service availability. We need self-healing networks. Maybe we are looking at customer experience first because if you have 100s of agents in call centers it costs a lot of money. It s a pure cost gain. How are you using AI (machine learning) combined with analytics for network management? 12.99% 33.33% 51.41% 23.16% 14.69% To predict problems To self-heal To automate changes in policy We will use AI in the next two years We have no plans 2017 TM Forum 8
But network is the bigger prize Telecom Advanced Next Generation Operational Supporting System (TANGO) uses big data analytics and machine learning to improve traffic management and network operations more broadly. Telefónica has launched AI-enabled Service Operations Centres (SOC) in Argentina, Chile and Germany that will allow it to capture, in real time, the quality of customer service experience. AT&T Labs video analytics team is developing a deep learning-based algorithm that will allow it to use fullyautomated drones to inspect and repair its 65,000 cell towers. 2017 TM Forum 9
But network (automation) is the bigger prize "By introducing technologies such as SDN, NFV or network slicing, the network becomes more flexible and powerful. Nevertheless, the complexity of the future network is not reduced, but transferred from hardware to software, from the network itself to management and operation, from equipment to people. Experiential Networked Intelligence is expected to help operators to solve these problems". In order to keep pace [with the continuous increase in network size, traffic volume, service complexity, and customer expectations], network operators must supplement today s human centric trouble-shooting and manual remediation methods with machine-based decision making and autoremediation approaches to enable the accelerated deployment of new services while supporting hyper traffic growth at a lower cost structure. 2017 TM Forum 10
But network (automation) is the bigger prize A new Open Digital Architecture Five core principles 1. AI capable and autonomous 2. Data-centric (rather than process centric) 3. Microservice-based using open APIs 4. Real time 5. Supports cloud business models and cloud-native capabilities Source: 2017 TM Forum 11
But network (automation) is the bigger prize Simple automation is not enough given the need for superhuman speed/complexity of decision-making. Therefore, our aim will be for an event-driven model governed by a holistic knowledge-defined artificial intelligence. Brian Levy, Senior Advisor, TM Forum AI layer operates with common data from the ODA in a closed loop decision process with applications to each ODA platform layer (customer experience, product and service development, network management. https://www.tmforum.org/wp-content/uploads/2017/10/tm-forum-open-digital-architecture-final.pdf 2017 TM Forum 12
Five sets of use cases Customer Experience (including fraud and security) Service Management Network New products and services General business functions HR Legal Finance Partner management To create an industry standard for making AI apps reusable and easily accessible to any developer by building a common framework and platform for exchanging machine learning solutions. 2017 TM Forum 13
How big an impact could AI make on opex and EBITDA? 2016 Financial results Revenues: NOK 131bn Opex: NOK 51bn Personnel costs NOK 13bn EBITDA: NOK 46bn EBITDA Margin: 35.4% Total employees 37,000 Cost/employee NOK361,000 ($44,600) If, as a result of the use of AI and ML, Telenor was to reduce its personnel costs by 10%, it would result in an increase in its EBITDA margin to 36.5%. If, as a result of the use of AI and ML, Telenor was to reduce its personnel costs by 25%%, it would result in an increase in its EBITDA margin to 38%. 21% 5% 13% 13% Personnel Operations & maintenance Energy 25% 23% Sales & Marketing Regulatory Other 2017 TM Forum 14
What are the key challenges? Overall challenges No data, no AI Finding the right (technology) partner(s) Matching the business (use case) with the technology Recruiting the right skills Taking the plunge! Working in the cloud The impact of AI on people and jobs. Network challenges 1. Lack of maturity of network components and support systems. 2. Lack of software expertise within the organisation. 3. Lack of data analytics expertise within the organisation. 4. Overcoming fear that automation will limit control and result in outages. 5. Lack of standards for E2E management. TM Forum expert challenges Technical: ML needs large training sets. Can the fragmented telco industry generate scale? What is the best practice for introducing AI/ML progressively? If it is introduced into specific domains, how can these be stitched up retrospectively? Regulatory: who is financially, legally, ethically and morally responsible within a CSP when a machine makes decisions? 2017 TM Forum 15
What CSPs need to do to make it happen Learn about AI technology and what is required in terms of skills and resources to leverage its potential. Integrate with analytics function. AI and machine learning needs data. Build a data lake to facilitate the deployment of AI use cases. Standardise. If CSPs are to attract the interest of technology firms that are developing AI solutions and capabilities they will need to make it easy for these companies to do business with them. Be realistic. The market for AI talent is red hot and CSPs are not the obvious career choice for a computer scientist that wants to build a career in AI. Think automation. Unless networks can be automated, the deployment of SDN and NFV may actually increase the costs of network operation. AI holds the key to advanced automation. Do not over-commit. It is important for CSPs to incubate AI projects across the business without necessarily pre-judging how and when they should come together. Be a guinea pig. Many technology suppliers are seeking to gain visibility to showcase their AI capabilities. Collaborate. AI is an area where there is everything to gain and very little to lose from collaborating with other CSPs. 2017 TM Forum 16
Thank you! A new TM Forum report, AI: The Time is Now, will be published later this month www.inform.tmforum.org 2017 TM Forum 17