AI in Telecoms: From Self Driving Cars to Self Driving Networks

Until and unless you have been hiding under a rock, odds are that you have heard about the fascinating world of AI and the transformational impact it is going to have on our lives. Certain parts of the euphoria remind me of the dot-com bubble from the 90’s – with the technology hype hitting it’s peak. There seems to be intense excitement at all levels from CEO’s to analysts to grass roots developers; expecting AI to dramatically change our lives and the world, so to speak.  Well, I really wouldn’t blame them given the hype created by the billions of dollars being invested in this domain. All the stars seem to be perfectly aligned – but let’s peel the onion a little bit to better understand what really is AI, why the hype, why now and how can it dramatically change the Telecom Industry landscape, in particular.

What is Artificial Intelligence? Why the hype now?

AI is essentially humans teaching machines how to learn & mimic their intellect.

Quite like a toddler who learns by experience- constantly absorbing information, decoding it and understanding patterns; we feed machines huge amounts of data that create their own algorithms, and constantly tweak them to meet their objectives. Over time the program gets smarter and very human like i.e. less artificial and more intelligent.

During the past few years, a few factors have led to AI coming closer to unleashing its true potential: mainly a ton of data becoming available along with faster GPU’s for processing it.

As billions of things get connected to the web, there is a tsunami of information being generated. From the dawn of time until 2004 – 5 exabytes of information was generated and captured. If you ask how much is that data- it essentially is books stacked from earth all the way to Pluto and back 80 times over. But over the last decade the amount of data being captured has exploded with 90% of the world’s data reportedly being generated in the past two years alone. And now thanks to advances in processing speeds, computers can make sense of all this information a whole lot more quickly.

Because of this, tech giants and VC’s are pumping the market with cash and new applications. China alone has already invested $109BN in this space, and the race between countries is heating up. The reason being that 15-20% of the global GDP is forecasted to be linked to AI-related industries by 2025! Mind numbing!

Source: CBInsight

How is AI different from Predictive Analytics and Big Data? Are there financial benefits of investing in all 3?

Often, people enquire what the difference in Big Data, Predictive analytics and Machine learning is. They want to have the ability to differentiate between the three to ensure adequate allocation of resources, and to set right expectations with all stakeholders.

Big Data: Collection of “data sets” which are large & complex and difficult to process using on-hand database management tools

Predictive Analytics: Practice of extracting information from existing data sets to determine patterns and predict future outcomes & trends

Machine Learning: Machines performing tasks that normally require human intelligence (visual perception, decision-making, learn etc.). This takes into account the entire context of the situation

In a nutshell, creating Big Data Lakes for gleaming insights and learnings is the foundation, which all corporations must invest in. Using Machine Learning, the self learning engine to unlock value from the data, organizations can extract insights, predict outcomes or perform human like tasks. Investment in all three i.e. Big Data, Predictive Analytics and Machine learning is soon becoming table stakes. Business & IT teams need to work “hand in glove” to prioritize the output needed. Often working in isolation leads to converting the haystack into a bunch of needles rather than finding the one insight which is critical to business

Should we bet on AI and Robotic Process Automation (RPA)? Everyone seems to be deploying them, should we invest in them now?

 These my dear Watson are the wrong questions. The proverbial cart coming before the horse. The correct question should be, “How do I use technology to solve my business problem quickly & efficiently?”.

However, lets dwell into both RPA and AI to distinguish the two. The quest is to bridge the journey from Human to Digital labor in the most efficient form. Given the pressures on the bottom line, “Digital Labor” is no a longer a consideration but now a mandate, for most companies across the world. RPA and AI are the two unique sets of tools used to traverse the journey from one end of the spectrum to the other i.e. Digital Labor = Cognitive Computing (AI) + RPA (or Digital work)

Basic RPA consists of automating entry-level transactional, rule based and repeatable processes. It involves working with structured data and within well defined parameters, for the bots to complete the tasks autonomously. However, digital labor is automation driven by self learning and adaptive technologies. It involves self learning, digestion of super data sets, hypothesis generation and evidence-based learning. All fun stuff!

Most companies and business units, I speak to, love the concept of AI and want to jump right into it. However, it is the journey of realization i.e. what really are the use cases, articulating the pain points, benefits ascertaining metrics that would quantify whether AI and/or automation is necessary. It is then that collectively we decide on the course of action and implementation.

What are the telecom specific use cases? What are the financial benefits?

 The telecom industry is at a crossroads: no other industry is driven as much by bits & bytes and at the same time faces disinter-mediation challenges for the same reasons. Business models are changing, customer expectations & behaviour are constantly evolving, legacy economics & paradigms are becoming untenable; and that really is the reason why 2014 was the first year when the total global revenue of telco’s had marginally negative growth. Every Investor relations call is bombarded with questions on increasing revenue, reducing Opex & Capex. So here are some ways AI, could alleviate the pain

 Capital reduction ($10-12 BN savings over a decade) :

Between the 3 Telco’s in Canada, the collectively annual capital spend ~$10BN to grow and sustain their respective networks.  Using this run rate, would peg the capital spend over the next decade to be at ~$100BN, even if we do not take into consideration a spike of an estimated 40% due to increased spend on 5G and fibre deployments.

Given that the lifecycle of technologies has reduced considerably, CDMA lasted 25 years, HSPA possibly 10 and LTE even lesser, time to recuperate investment cycles are reducing. In the mean while, cost of electronics has gone down and most of the capital effort is spent in optimization of networks and installation.

At last count 70% of capital costs to upgrade from one technology to another involved was spent on labor & infrastructure and only 30% on the actual OEM related electronics. That would technically peg the Telco’s as infrastructure players vis a vis a pure play telco, so to speak. However, AI and other combinatorial techs can be leveraged to reduce that labor cost for optimization significantly. A lot of the OEM’s have started providing their own software for SON (Self Optimizing networks) but that would really be akin to asking the wolves guard the sheep. Telcos need to have their POV and build on it over time, to ensure basic hygiene and best practices be kept in-house. Secondly, predicting usage of subscribers by geography and ensuring the network dimensioned correctly- should and will not be done on the side of the desk of engineers- but through intense scenario analysis and planning, considering all parameters. Lastly, optimization of spectrum auction pricing can be done using game theory and decision trees. Given the appetite for all the operators to spend billions on it, we expect a healthy dose of AI to be used to save thousands of consultant hours and saving millions (if not billions) in the bargain.

We estimate at least 10-15% ($10BN) of the Capital envelope can be reduced using Cognitive technologies holistically across the process.

Opex Reduction ($3-5BN savings over a decade) :

• Predictive Maintenance: The ability to fix problems with telecom hardware (such as cell towers, power lines, etc) before they happen, by detecting signals that usually lead to failure. This significantly reduces down time and dramatically improves CX and churn metrics.
• Video Analysis: Instead of sending field workers to check on site hardware periodically, AT&T has invested heavily in drones and in AI to analyse video data captured, which proactively raises flags for the field team to go in and fix issues.
• Self Driving Networks: In the future its likely that governments might ban car driving by humans as a reckless or unnecessarily risky activity- quite similarly, there is no reason why we should be done away with the Network engineers! Utopia really would be where networks would run themselves, away from our grubby, inefficient hands. This truly is a huge opportunity for cost cutting and efficiency gains. In an ideal world, we should not need so many staff to manage operator networks i.e. A lightly staffed network powered by an economical but sophisticated AI will look more desirable than ever. That is truly aspirational today, but something to work towards.
• Customer service chat bots– Every telco customer calls in 4 times a year at an average cost of $15/call for password resets, billing enquiries or changing plan details. This Opex of ~$2BN/year across the 3 operators can be dramatically slashed by using chat bots which are equipped to do the same. Automating 10-15% of the customer service inquiries, routing customers to the proper agent, and routing prospects with buying intent directly to sales people, will make a material impact to the bottom line! Further, leveraging NLP, sentiment analysis & text to speech- virtual customer service agents could be used to replace humans, with the customer not being able to distinguish one from the other. That could be a tangible reality in 5-10 years time; literally taking the cost per down from $15 à 15 cents/call.
• Speech and voice services for customers– Allowing customers to browse, explore and buy content by spoken word rather than remote control (e.g. DISH network and Amazon’s Alexa partnership)’
• Security & Fraud: Detect fraudulent activity in credit-card transactions or identifying web traffic to the website and customers, looking to exploit network and infrastructure vulnerabilities.

Revenue Enhancement:

• AI algorithms can help combine historic patterns, Psychographic analysis and behavior (plus “look alike” patterns) with ongoing real-time engagement to provide relevant, targeted, contextual experiences for consumers. The outcome will be upsell recommendations and offers, helping improve the conversion rate of offers, and enabling incremental wallet share.
• The same algorithms could be potentially leveraged to predict subscriber’s willingness to pay a particular price for a product and estimate product price elasticities
• Quantify sale’s leads likelihood to close

 How do we tackle AI, as an organization? How much investment is required?

 It is best to use a good framework to get the ball rolling organization wide. Some frameworks that we have used in the past include:

Please feel free to engage us to better understand the foundational capabilities needed to get the AI program up and running. From building use cases, decision making paradigms, economical impact analysis, choosing governance model and platforms to be deployed, we have gone through the process a few times. Every business requirement is unique, and we look to hearing from you on your needs.

Will we be all out of work if AI unleashes its true potential? When will that happen?

The notion of Singularity is based on the hypothesis that the invention of artificial superintelligence will abruptly trigger runaway technological growth, resulting in self-improvement cycles causing an “intelligence explosion” and self creation of powerful machines that would, far surpass all human intelligence. Scientists however have concluded that Singularity might be a tangible reality in the 2040-50 timeframe, but people will evolve and learn, they will retrain and redeploy in different parts of the value chain.

In the short term this is absolutely not a problem as companies are focusing their attention on solving very specific AI problems in their respective niches. Trying to solve the general AI issue would be boiling the ocean, so to speak.  However, today AI is a huge opportunity- with an average developer making north of $300K/year. Got a teenage kid? Get them coding!