Heavy Reading Analyst: Autonomous Network Operations Demand Agility & Adaptability
DALLAS -- Automation Everywhere -- If telcos are to make significant strides in automation in 2018, reduce costs and increase efficiencies in their network operations, carriers must overcome barriers like engineers' skills gaps, and develop agile company cultures that are "adaptable to change," says Heavy Reading Analyst James Crawshaw.
Crawshaw kicked off Light Reading's Automation Everywhere event this morning with a look back at the evolution of automation, which began by automating exchange equipment to electromechanical switching, to electronic switching of software management systems, and finally to the actualization of SDN and NFV.
Crawshaw defines automation according to McKinsey's framework of intelligent process automation which includes three categories -- robotic process automation (RPA), machine learning (ML) and natural language processing. Mastering both RPA and ML is key for telcos to achieve success on their automation journeys.
Robotic process automation
Robotic process automation has garnered the most traction in the financial services industry, but is gaining momentum in the telecom industry as a way to automate mundane tasks, "eliminating the need for manual, error-prone, duplication of data entry across different systems," explains Crawshaw. RPA ties together different software systems to perform tasks such as back office activities, customer support or IT support.
"It's a way of trying to automate a lot of disparate processes without having to do some big software rollout or software development -- you can use these relatively inexpensive tools to automate," he says. "AT&T has adopted RPA quite extensively, Deutsche Telecom is using it, Telefónica has a good case study. It's a useful tool for telco's broad back-office processes -- not so much on network automation per se, but automating things like customer contact centers."
AT&T Inc. (NYSE: T), for example, has trained over 2,000 employees to write bots that have been used in the migration of customer accounts from DirectTV to AT&T, completing requirement documents for Ethernet services, automating sales order entries and reconciling revenue against assets and inventory, says Crawshaw.*
Automation will not only be central to reducing manual tasks such as customer support, but also in managing 5G networks where machine learning will also play an important role to addressing the influx of data on operators' networks, explains Crawshaw. A subset of artificial intelligence (AI), machine learning is the aspect of AI drawing the most interest from both academia and industry.
"One of drawbacks of applying ML to networking … is there isn't this theory of networking in the way we can theorize how the brain or eye works -- we don't know what is the correct way a network should work, there's nothing you can model per say."
While machine learning isn't a new concept, Crawshaw says it only made significant strides about ten years ago due to academic breakthroughs, and with ongoing improvements in compute capacity. Machine learning has also become a more cost-efficient endeavor: the telecom industry has access to more massive data sets, and data analysis can be performed using cloud-based systems like AWS.
Next page: Use case for machine learning
Use case for machine learning
The use cases for ML are endless, from customer-facing opportunities such as chatbots, contact center optimization, fraud detection and prevention to networking use cases including security, prediction of network faults, and SLA monitoring and enforcement.
"There are lots of applications [for machine learning] in operations, so fraud detection for example, security, eventually service assurance," says Crawshaw.
Developing machine learning code is only one small part of building ML systems, says Crawshaw -- service providers still need IT infrastructure, process management tools, data collection tools and process management systems in place. Several standards bodies are working on addressing these challenges to applying ML to network automation. ETSI, for example, established the Experiential Network Intelligence working group, which is striving to create a system "that will sit alongside existing systems like the management and orchestration system for NFV," collect data and deliver suggestions back to the system, says Crawshaw.
The Telecom Infra Project (TIP) also has an ML working group focusing on machine-learning based network operations, customer behavior driven service optimization and Multi-Vendor ML-AI Data Exchange Formats to develop generalized ML models applicable across the telecom industry.
Telcos still have quite a few barriers to fully realizing the benefits of machine learning, Crawshaw explains: "Network engineers don't necessarily have all the mathematical tools they need and don't necessarily have good enough data sets -- a lot of the data they have has a lot of errors in it. A lot of the work involved in any ML activity is about cleaning up the data sets."
While robotic process automation and machine learning can eliminate the need for manual processes in the network, these tools are only as effective as the engineers developing and managing them. Herein lies the paradox of automation -- the more advanced the control system is, the more crucial the contribution of the human operator, explains Crawshaw. When an automated system fails, it still falls back on human oversight to intervene.
*(Editor's note: An earlier version of this article incorrectly stated the number of employees writing bots for AT&T. The number is 2,000. In addition, the company has not officially stated the savings that it collects from its use of bots.)— Kelsey Kusterer Ziser, Senior Editor, Light Reading