SD-WAN has played a key role in helping manage this complexity by providing some element of visibility and automation over enterprise networks. But SD-WAN is still bound by the same underlay options – Multiprotocol Label Switching (MPLS) or the internet – and this constrains the technology.
However, as artificial intelligence (AI) and machine learning (ML) continue to mature, they’re becoming invaluable in helping managed network service providers enhance their customer experience and further automate network operations.
Below I’ll share some details around how we at Tata Communications are adopting applied AI/ML to deliver ubiquitously seamless customer services and improve operational efficiency. Exploring the three main areas we’re focusing our efforts: customer service, capital efficiency, and improving productivity and operational efficiency with end-to-end network management.
1. Automating and personalising customer service
We’ve started using AI/ML to automatically diagnose faults in our network services, irrespective of whether those issues happen on our network, a third party’s, or if it’s at the user’s end.
All faults ranked as ‘severity one’ are taken through the auto-diagnosis engine, which performs a service validation and alarm analysis on it. The diagnosis engine’s ML algorithms analyse all of the active alarms on the service and look for correlation to help engineers better and faster understand faults in the future.
So, both the engineer and the customer can enjoy faster and more consistent troubleshooting, and thus, earlier resolutions. Currently, 85% of severity one faults are successfully diagnosed by the ML-powered engine, reducing the Mean Time To Recovery (MTTR) while providing a seamless experience to customers.
Aside from faults on our network, we’re also using AI/ML to automate our response to customer queries. For example, our contact centre receives more than 12,000 calls every month on average.
“By leveraging AI, we’ve been able to create more personalised, intelligent interactions with those customers when they call to report faults.”
This comes in the form of a CLI (Calling Line Identification Number) associated with each customer registered on our CRM system. When a customer calls, the CLI instantly provides the customer service rep on the line with their account details, still open or new faults that’ve recently been recorded, along with previous diagnoses and segmentations linked to that caller’s account.
Armed with this information, the caller can be greeted by name and provided with an update on all their existing faults, as well as a first level diagnosis of any new ones – all before a new ticket is opened, making for a much more personalised experience for the customer.
Furthermore, we process all interactions through speech analytics to better understand customer sentiment. These analytics provide call handlers with near real-time insights on customer tone, so they can identify potentially dissatisfied callers and attempt customer experience recovery while they’re still on the line.
2. Improving capital efficiency
As an organisation, we remain invested in enhancing and expanding our network coverage. In this, we’re continuing to use AI and ML to analyse multiple factors and predict where future connectivity demand may come from. With the technology’s help, we can provide on-net connectivity for more customer sites with better service experience.
For Customer Premises Equipment (CPE) management, we’re exploring the use of AI and ML to sharpen our CPE stocking algorithm. That way, we can stock the right models and quantities of products, speed up service delivery and turn up to customer sites faster.
3. Productivity enhancement and better operational efficiency
On an average, we receive around 50,000 emails every month. It’s a very popular channel for us and in the past our average response time was about 25 minutes. However, as a global provider of business-critical digital services to a plethora of industries, we understood that our customers’ queries need to be heard as soon as possible. We also felt it important to make sure our customers had the ability to contact our support groups through any channel they so choose.
To aid this, we started using AI and ML in the routing of customer tickets and calls.
“Innovations in ML and Natural Learning Processing (NLP) now help the automation of ticketing and customer responses by analysing the keywords of incoming correspondence.”
The NLP engine helps automate our customer support workflows, which includes a chat-bot solution that fully automates third-party ticket resolution.
The solution can also understand and respond to supplier updates, escalate when needed, share timely updates with customers and provide RFOs (Reason for Outage) in standardised templates. All of which is complemented by a seamless handover process between human and bot agents in the case of exceptions.
Since the shift, 75% of our customer complaint emails are now processed via automation, with no need for human interference. This has significantly reduced our response time from 25 minutes to almost instantaneously – much faster than the industry standard of 15 minutes.
There have also been numerous positive productivity benefits as a result, such as lower turnaround time, reduced time to rectify faults and better customer experiences.
4. End-to-end network event management
Because our service offering is so tightly integrated with our partners and suppliers, we’re able to provide many additional end-to-end managed services for our customers.
However, because of this diverse, third-party network ecosystem, the ability to detect faults in our network (which involves visible alarms and events from our network elements) is sometimes limited by our inability to also monitor all our partners’ networks for faults.
To address this issue, our network operations teams deployed AI/ML solutions to observe and learn our ecosystem’s network traffic behaviours instead.
“For instance, by deploying ML-based models to our Call Detail Records (CDR), we can detect multiple unusual patterns in support call volume which helps us identify network faults in partner networks faster.”
Similarly, major events in the supplier network usually trigger multiple service faults in our systems. So, by comparing advanced fault and ticket patterns with ML-based algorithms, we can detect common and major faults in partner networks and quickly initiate responses.
The path to a fully automated and intelligent network
Through these solutions, our network is constantly learning and adapting to securely meet our ever-growing business requirements. We are also using these advanced technologies to detect time series anomalies, correlate events to perform root cause analysis and improve customer experience.
And as we continue to make good use of AI and ML-driven services to optimise our network architecture and give us better control and management of our services infrastructure, we’ll be able to continue to deliver higher quality networks, voice, and other innovative services.