Contributed by Patrick Kelly, Founder & Principal Analyst, Appledore Research.
Applying AI for improving prediction and business outcomes will transform decision making in most job functions of telecom operators over the next decade. AI tools used in the right context will improve customer care, network operations and planning, fraud detection, and personalized marketing. But CSPs must still start with a clear view of the business outcomes they want to achieve and use that to drive their AI strategies.
AI Becomes Mainstream
The abundance of high-quality data, advances in computational processing, and sophisticated machine learning models, available in the opensource community, are reducing the cost and improving the accuracy of AI compared to conventional hard-coded methods. Attempting to deploy AI even 5 years ago was not economically feasible because of limited data sets, higher cost computing, and inferior machine learning models to conventional statistical regression techniques. In short, it was difficult to justify the economic benefits.
All that has now changed, and the technology is actively being deployed in many sectors outside of telecom for natural language translation, facial recognition, advance driver assistance systems, and industrial automation.
The value of AI is that it uses data to discover patterns, and then predict outcomes more reliably than current methods. The power of AI is that it is constantly improving its learning algorithm, using a technique called back propagation that changes weights in the hidden layer, to achieve higher levels of accuracy.
AI is not the only way to solve many problems telecom faces today. However, in many cases, over time it will become a much, much better way to solve those problems. Figure 1 illustrates the potential economic value that CSPs can achieve applying AI over traditional methods used today.
More economic value is realized as more data helps to continuously learn and improve the business outcomes. The value accrued continues to grow, mostly as a result of improvements in predicting outcomes. The breakeven point will take longer to occur due to the development cycles required to achieve the desired results in training the models.
Drivers of AI in Telecom
Investment in AI within telecom are largely as a result of four main drivers:
- Technology advances in the 5G era and cloudification of the network.
- The emergence of new business models in the edge cloud.
- Downward pressure on capital investments.
- Workforce retraining and OPEX efficiency.
The enabling technology for 5G is much more than the radio interface and includes virtualization which will lead to a massive increase in technological complexity. If 5G is deployed with existing operational processes and systems environment, operational costs will balloon and on a long-term basis be unsustainable.
In a 5G network, a provider may want to support a low latency application in a robotic manufacturing facility which demands recurring and rapid changes to factory production demands. The desired network state and associated KPIs must be captured in near real time to initiate new services, move workloads, or re-route network connections if service is impacted. AI is dependent on the ability to collect and process large amounts of telemetry data to identify patterns and determine if the service is operationally “green” or an event trigger should be generated to the network orchestrator or service controller.
The ability of the RAN-cloud to self-manage will be necessitated by the application of intelligent algorithms and AI. Over the past decade search algorithms have proven their value in SON.
Edge cloud represents a potentially large opportunity for CSPs in the next decade. CSPs have network connectivity at the edge as well as physical presence in the form of central offices and base stations. However, to make edge an opportunity, telco operators must focus on minimizing the cost of ownership for edge datacenters and the cost to run an application on their edge cloud. AI will be used in edge to the same extent as the virtualized core for predicting capacity and potential faults in the infrastructure. AI-based security and threat analysis will be utilized to predict critical IoT devices in the public sector and enterprise markets.
Today, most CSPs are cutting CAPEX budgets as a result of excessive debt on their balance sheet and management’s focus on improved earnings. To drive a much more efficient capital allocation plan, AI can be utilized to predict capacity demand, understand customer consumption patterns.
Operational expenses are 3X most Capex budgets for CSPs, with the bulk of those costs in staffing and network operations. A NOC driven predominantly by a human-only workforce is not only expensive, but often cannot scale to meet the future consumer demand. The automation of high-volume tasks will require CSPs to think about alternative workflow processes. AI applied correctly in the service lifecycle will outperform and dramatically change the OPEX curve.
Targeting AI Investment
It is important to look at AI in the context of the business problem and the results that you want to achieve. That is: where is the greatest business value potential from faster, cheaper and more accurate results than current workflow tasks or traditional IT tools?
- A 20-fold improvement in isolating faults and service impacting events.
- More accuracy in predicting demand of network capacity.
- Better asset allocation for capital investments.
- Increases in customer retention.
- Reducing fraudulent transactions.
- Improvements in personalized marketing campaigns.
- Predictive maintenance and avoidance of unnecessary truck rolls.
- Improvement in energy management.
Nonetheless, telecom executives and implementers of AI must also balance the potential benefits against the long-term strategy, organizational impact of AI on existing jobs, data privacy laws, and the economics of AI in improving decision-making for all aspects of its business.