"In the past, large volumes of data made us sweat". So said Pari Bajpay, vice president of Next Generation Enablement at CenturyLink, during a presentation titled "Can AI deliver its promise of a cost-effective, improved experience in telecom?" at the TM Forum's recent Digital Transformation World event in Nice.
"We didn't have the networking, compute and storage capacity to cope. A lot of the data would be turned off and you would only work on the critical aspects of the data because what you had on the other end of it was humans that could not process such large volumes," noted Bajpay.
However, as big data technology has matured, Bajpay and his team at CenturyLink have grappled with the issue and are now leveraging AI to extract more value from their data.
"With AI we are able to get value out of humungous volumes of data and make data-driven decisions with a broad scope of data points. We can correlate across unstructured and structured data to predict future events. The level of efficiency that we get now was not possible before."
During his presentation, Bajpay discussed the challenges of implementing AI, including siloed data, poor data quality and the need to build trust. We spoke with him after the event to dig a little deeper on some of the issues he raised.
Fragmented data and data integrity
Bajpay noted that siloed data, and poor data quality, are some of his major bugbears. "Fragmented data, and data integrity, is a problem -- it's going to be a challenge for some time. There is a lot of important data outside of our data lakes and data warehouses, and that data needs to be harnessed to tap the full potential of AI/ML."
Like most telcos, CenturyLink has accumulated multiple IT systems over the years and is on an ongoing mission to consolidate them. "We have many different data warehouses inherited through acquisitions. We cannot retire or consolidate them fast enough. Data is sitting in multiple different places and therefore data integrity becomes an issue. If you do not have good quality data, then the algorithms that you run will not work so well."
CenturyLink has a target data lake architecture that is a hybrid of on-premises servers, internal cloud and external (public) cloud. Bajpay says he's in no rush to move "data at rest" (legacy systems) to the cloud unless there is a compelling business need. However, "data in motion" needs scalability and he doesn't want the overhead of managing it in-house. "We like the flexibility that public cloud providers give to take advantage of best-of-breed technology. They are also constantly introducing new AI/ML technologies that we would like to leverage."
The need to build trust was another challenge for AI that Bajpay highlighted in his presentation. "It's getting better, but we have to work on building trust in our customer base and employee base that these algorithms are actually good. We have made headway, but trust is going to be a challenge going forward."
Part of the challenge is that there's a learning curve with AI; not everything works perfectly immediately. "We can predict failures in the network, but we get a lot of false positives. Operations won't trust AI if it is only 65% accurate. We use reinforcement learning to improve the accuracy. We won't get to 100% but if we can get to 85-90% that should be adequate."
Enhancing RPA with AI
Robotic process automation (RPA) enables the automation of repetitive rule-based processes by mimicking the interactions of users with multiple systems. CenturyLink started its RPA program last year with systems integrator Prodapt, using UIPath software. Since then, 300 bots have been deployed that are "delivering significant value for the company," according to Bajpay.
Now CenturyLink is increasingly looking to combine AI with RPA. "A lot of the use cases that we are picking up now have AI such as NLP [natural language processing] baked into them. By the third or fourth quarter of this year we aim to implement machine learning in conjunction with RPA."
CenturyLink is using RPA across multiple areas including service assurance (customer and network facing operations), service delivery, network management, billing, HR, legal, sales and marketing. "With RPA we are seeing 300-400% RoI," noted Bajpay. "There is a lot of low-hanging fruit out there and with ML the value goes up even more. But the KPIs are not just cost savings -- it's about what it does for the end customer. Imagine if you have fewer human touchpoints during a transaction -- cycle times improve significantly. We are saving the company money but we are also hugely improving customer experience as we shift customer support from reactive to proactive and predictive."
Data science Center of Excellence
CenturyLink has set up a Center of Excellence (CoE) for RPA and data science, including AI. The CoE takes care of analytics infrastructure, governance, testing and production support. The CoE team includes a business unit lead, who identifies and prioritizes opportunities; a delivery lead, who tracks deliverables; solution engineers; and process consultants. The CoE is then leveraged by federated teams across the business in areas such as IT, finance, network planning, product, pricing, marketing, margin assurance and HR.
The federation concept is key to the success of the program, according to Bajpay. "Think about it. Where does the business logic sit? It doesn't sit with the IT folks. They need to understand it and then program it in. But as the tools get simpler, now we can get a wider team to participate. In RPA we have nine development teams -- only two of these are within IT, the other seven are outside IT. Similarly, we have 250 data scientists of which 200 are outside of IT, and that number is growing."
From visualization to predictive action
Bajpay notes that analytics used to be largely about visualizing data so that humans could spot patterns. "Now we are taking action directly on the analytics and algorithms. The insights we get are driving action, not just visualizing it with business intelligence tools."
With AI, CenturyLink aims to move from a reactive mode of automation to proactive and the predictive. For example, while today they have structured data and rules for outage identification and troubleshooting, they are rapidly moving towards an AI-enabled, self-healing network. "We used to have customers call and tell us when there was a failure. With AI and ML, we can know when there is a problem before the customer calls us -- we tell them about it. We can predict failure. It's not easy -- ML models are not 100% accurate, but they tend to improve over time."
Bajpay sees further opportunities for CenturyLink to leverage AI to automate and improve its business. "We are embedding AI into all of our processes from quote to cash. Technology has gone from the point where it asks the business 'How can we support you?' to the point where technology is the business, and that's the way it's going to be in the future."
— James Crawshaw, Senior Analyst, Heavy Reading