Strategies to Improve Customer Experience as You Roll Out 5G —
An experienced engineer once told me that there are consequences to every change made in the network. That means as a network engineer you must understand the intended effects and the side effects of a change you are making. Unfortunately, as the network densifies and virtualizes, understanding the side effects becomes increasingly difficult.
Communication service providers (CSPs) face a significant challenge maintaining and improving network service quality and availability as the environment gets more complex in a number of dimensions and as 5G networks continue to roll out.
Network Function Virtualization (NFV) is transforming dedicated network elements into stateless network functions that run as multiple containerized software instances. Cloudification of the network infrastructure is migrating into public and private cloud deployments. And an explosion of services and devices ranging from smart phones, and infotainment systems, to factory robots and environmental sensors with widely varying throughput, latency, security, and mobility, requirements are impacting their networks.
With the diversity of services and devices, understanding each subscriber’s network experience cannot easily be done directly from network key quality indicator (KQI) values, which capture a view of network performance from the perspective of the network. As a simple example, does an increase in packet jitter value impact a subscriber’s experience? It depends. It depends on the magnitude of the increase and the service that the subscriber is interacting with. If the subscriber is consuming a recorded video stream, the packet jitter will be hidden by the fetching of video chunks. However, if the subscriber is participating in an interactive audio and video conference such as Zoom, WebEx, or Google Hangouts, packet jitter can have a substantial impact on the user experience. This simple example is for just a single quality indicator. It typically requires a combination of measures to derive a true indication of a subscriber experience.
Scale this across hundreds of KQIs and services, thousands of locations, millions of devices, and tens of millions of connections, and you can see that understanding subscriber experience from the inside-out quickly grows complex. This complexity makes it difficult and expensive to scale as the network and services simultaneously evolve.
Using AI and Analytics to Really Understand Each Subscriber’s Experience
A best practice is to take an outside-in approach, bringing the outside perspective of network and service performance from the view of the subscriber in, to inform how the network is impacting subscriber experience. This subscriber experience measure is used to train algorithms, allowing the use of artificial intelligence (AI) and machine learning (ML) techniques to identify the complex relationships of KQIs along with other attributes such as location, device manufacturer, software version and service type. These relationships and values will evolve over time as the network, devices, and services change with new hardware and software constantly being rolled out. These trained algorithms can now be applied from the inside-out, providing the CSP with insights to understand how network operations are impacting subscriber
experience on a micro-segment basis.
The services being used are real-time. When insights are available only after a problem occurs, all the network operations team can do is apologize. These insights, including information about what is happening, where it is occurring, who is impacted, and actions that can be taken by the CSP need to be developed as the issue is occurring.
The insights needed sound very much like those provided by business intelligence and big data analytics. However, the real-time needs of the operations teams drive very important differences in the tools, rules, timing, data hygiene, and risk tolerance. A network operations team cannot wait for delayed data to arrive in order to determine the health of a cell site, while a churn report being shared with a regulator or the financial markets requires validation that data has not been omitted and the reported values are accurate, even if it is delivered a bit later than desired.
These insights for the operations teams require operational intelligence tools such as real-time stream processing, edge analytics, and online machine learning. Real-time stream processing acquires, enriches, and aggregates, massive volumes of data, such as generated by 5G networks, while using half of the compute/processing-related hardware required by traditional analytic and big data solutions. Stream processing, combined with edge analytics based on standard data manipulation language (DML) such as SQL, provides the flexibility needed for CSPs to generate insights to meet customers’ always-on, on-demand service experience expectations in the rapidly evolving 5G network space. Locating insight creation close to the data generation saves time and cost, allowing data wrangling, aggregation and enrichment as well as scoring to be done before data has to be collected and stored in a traditional processing/storage cluster.
• For example, a leading MNO we’ve worked with has used this type of real-time stream processing combined with anomaly detection, fault correlation, and root cause analysis, to improve customer experience by reducing the time it takes technicians in a network operations center to restore service when a problem occurs. Specifically, at a fixed staffing level, their MTTR was reduced by approximately 18%, while the number of true issues handled by their NOC increased by 40%. All this was done without the software historically provided by their internal DevOps team that builds and manages an ever-increasing rule-based alarm reduction system.
• Another large MNO in Asia found traditional survey-based subscriber Net Promoter Scores (NPSs) costly and slow to produce. While they provided a reasonable representation of customer sentiment, they were not producing actionable insights that could be used to drive engagement with dissatisfied customers. This MNO is now able to combine handset data, service data, care events, and billing events, to synthesize actionable insights using the NPS survey data to train analytic models. And they can now visualize this data on management dashboards, and engage with dissatisfied customers experiencing poor QoE in near-real time to reduce churn and maintain growth momentum.
7 Tips for Creating Outside-in/Inside-Out QoE Visibility as You Move to 5G
What are the best next steps for CSPs that want to move to this new approach? Here are 7 key considerations:
Tip #1. Build for real-time consumption. Understand if the analytics tooling you’re looking to use is built to inform real-time decisions or built to improve strategic decisioning and reporting. These domains, while using many of the same data sources and similar computing and analytic techniques, diverge significantly in the trade-offs made when faced with delayed, partial, or inaccurate, data.
Tip #2. Provide access to data for those who know it and understand what it means. Make it easy to connect with new, innovative systems. It is with evolved understanding of the generated data through rediscovery that insights and breakthroughs are developed.
Tip #3. Have an outside-in strategy. Ensure that you are able to take your subscribers’ experience as input for network operations. It does not matter if there are no network alarms and all KQIs are green if the subscriber is not having a good experience.
Tip #4. Plan for the rate of change to increase. The virtualization of network functions, the proliferation of devices and services, and the customers’ always-on, on-demand service experience expectations are all accelerating this effect. Build processes and systems that can survive and scale in this environment.
Tip #5. Automation and orchestration is critical in managing costs for 5G network operations. CSPs will not be able to continue applying the same methodologies and processes used to operate 4G networks today and scale up to 5G density and complexity. Plan for analytic insights to feed automation and orchestration systems.
Tip #6. Performance and scale really matter. CSPs deal with some of the largest scale systems on the planet. It is unreasonable to expect any system to scale up or down 20x without substantial changes. If needed to gain mindshare (and budget), plan first for a smaller scale proof of concept or proof of value, and then to rebuild for a production deployment at scale.
Tip #7. Leverage standards where they are available. Every network is a multi-vendor network. Even if it does not start out that way, as networks and business relationships evolve, it inevitably gets there. The 3rd Generation Partnership Project (3GPP) has defined interfaces for network data analytics function (NWDAF) and management data analytics function (MDAF) services to provide and use analytics information in 5G networks.
A new outside-in perspective of customer experience applied to the information available from 5G networks, combined with advances in computer science, data science, and computing scale, can inform network operations to improve customer experience while increasing network operations effectiveness.