When speaking about his predictions for 2021, Erik Brenneis, Vodafone’s CEO, pointed out that telecoms no longer need just “connected” systems, they need “performant” systems, especially in the context of the Internet of Things. In a world where the number of systems, vendors, and technologies that telcos need to have to keep up with the times, convergence becomes paramount for business. But doing business is no longer about being the fastest. It’s about being reliable, consistent, and predictable to ensure smooth operations and customer satisfaction. The way to do this? Artificial intelligence.
This is very close to the mindset we have adopted at AVSystem. Looking at the market, it’s hard not to see AI and data analytics as the drivers for better, more performant systems that will help our customers – not just in the IoT space, but also in the telco industry – deliver even more advanced services. That’s exactly what our Broadband Service Assurance Platform was created for and why the team responsible for its development is applying machine learning to real-life client data to devise new solutions for better network monitoring, issue prediction, and automated maintenance.
In our next post, you’ll learn everything about how they attempted to correlate network and business data to provide better customer service. That is a truly technical dive into the deep waters of machine learning. But first… let us start with the why.
Let's get in touch!
Consistency is key...
The Broadband Service Assurance Platform uses data from access network devices (OLT/CMTS/DSLAM/MSAN/eNodeB) for network monitoring and fault prediction to improve telecoms operations, in particular the work of customer care (CC) and network operation centers (NOC). These two elements: monitoring and prediction are of key importance for telecoms. Only they can ensure stable operations and consistent service by preventing downtimes caused by unidentified issues.
Identifying the nature and the source of the issue may seem like a trivial problem unless you consider the complexity and fragmentation of telco networks. If you actually take into account how many end customers a telco company has and how many technologies it operates on, finding an issue and its cause is like finding a needle in a haystack. In fact, proper issue classification is the main obstacle on the road towards achieving the ultimate goal: auto-healing.
An ideal scenario would look something like this: an intelligent tool identifies an anomaly that forecasts an upcoming downtime; it then properly classifies the source of the issue and takes a reparative action to restore normal operations before the downtime actually happens, securing stable Internet access for the end customer.
This obviously hinges on artificial intelligence and machine learning to become the reality. And machine learning needs data. Therein all the problems lie. First, unsurprisingly, if you lack historical data (for example because a service is new), anomaly detection simply becomes impossible for lack of reference. Moreover, not all issues happen on the network. Take misconfiguration issues as an example. Network monitoring is helpless against an end user who meddled with their router settings. What is more, not all networks were created equal: what constitutes “normal” parameters on one network may be completely out of the ordinary on another. This is why these kinds of tools require a lot of tinkering and fine-tuning before they’re deployed and usually aren’t available out-of-the-shelf.
Not all is lost, though. On our road towards fully automated service assurance,there’s a lot that we can do with AI to create more performant networks. First and foremost, AI facilitates proactive maintenance. It anticipates issues and alerts network operation centers to take preventive measures.
On the telco side, proactive maintenance leads to both less and shorter downtimes than if the NOC were to act reactively. This translates to tangible savings as network downtimes are known to generate a lot of costs. It also leads to time savings for network operators and customer care engineers, letting them focus on other things.
More importantly, though, proactive maintenance minimizes the impact that slowdowns have on customers, as they don’t experience an actual shutdown and the psychical effect it might have on customer satisfaction. Even if their service slows down for a time, the end customers perceive it as more stable and infallible if there aren’t any actual outages.
...Convergence is another
No less important than consistency is convergence. Considering the fragmentation of telco networks and the plethora of operations support, business support, and network management systems (OSS/BSS/NMS), there is no other way to ensure smooth operations than through convergence. Operational data needs to be correlated with business insights, not just to ensure the proper functioning of the company, but also to improve strategy and drive revenue.
Take a simple example of upsell reports: if you take operational data (the information on maximal possible connection speed for an end customer) and put it against business data (the actual plan they have at the moment), you can identify opportunities for service upsell. Moreover, with group reporting (data on groups of users, not just individual customers), you can find places where there aren’t any possibilities for upselling, because the customers already have maximal speeds. Perhaps that means the company should invest in new, faster infrastructure?
Build your solution with us!
What did we want to do?
Having performance and convergence in mind, our team decided to analyze the data on network performance and the business data about customer care calls to see if they could accurately predict customer problems. This would be a step towards fully automated network maintenance. More immediately, it would be helpful for customer care engineers who’d have a known source of the issue even before the customer calls.
How did they approach that problem and what were their results? Read our next post, “Improving customer care through machine learning,” to find out!