The network is the product: How AI can put telco customer experience in focus

Telecom operators have long managed their networks as a technical asset more than a true product, maximizing network performance without drawing a clear link to customer outcomes. Recent advances in AI and telcos’ data maturity, however, open the door to a new, more sophisticated approach that treats the network as a core product to be managed with the type of sharp focus on customer experience (CX) that is the hallmark of software as a service (SaaS) and other tech providers. As a result, telcos can meet customers’ needs before they are even aware of them. By using AI to derive valuable insights about how customers interact with this product—the network—telcos can conceptualize the specific customer needs that the product fulfills, articulate product success, and translate the resulting business strategies into technical requirements.

These advanced CX measurements can help fuel a series of successes that can combine to build even more wins. For example, AI-driven CX intelligence can help identify customers who are up to five times more likely to churn in response to a poor network experience; it can identify opportunities to reduce capital expenditure by 5 percent to 10 percent; and it can also help generate a 10 percent to 15 percent increase in sales conversion, all of which can further drive up customer experience.

The limitations of past practices for measuring experience are rooted in the fact that they have been based largely on customer surveys and/or technical KPIs such as capacity or signal strength that engineers believed were most relevant to customers. While effective to a certain extent, these have fallen short. Engineers’ KPIs are a subset of what matters to customers and treat all customers the same regardless of the differences in their needs, while customer surveys rely on subjective memory and sampling rather than objective individual customer insights. These methods also overlook customers’ actual experience, which involves emotional and technical elements that customers may not be aware of, and are disconnected from commercial outcomes.

Until recently, telcos didn’t have much choice, since it has been difficult if not impossible to understand precisely how changes in network performance affect individual customers. Without the ability to correlate network actions to customer behavior, they had no way to answer questions such as: If we add a new band in a cell site, would customers appreciate the difference? Will affected customers purchase more or churn less because of that move, or will everything remain the same? These types of questions are very common and answerable in other industries, particularly in SaaS- and app-based businesses where the relationship between the product, its use, and the customer is closely monitored.

Data maturity and advances in AI, however, can now be leveraged to bridge that gap. While staying within the bounds of data and privacy regulations, a telco can begin to understand how specific actions on the network actually affect individual end customers, whether or not customers have perceived those effects, and ultimately to engage in more efficient network management. Some pioneering operators are already moving in that direction. They have developed the ability to discern this relationship and become more efficient in their capital investments while looking for every opportunity to increase both the top and bottom lines.

In this article, we’ll examine why network customer experience is increasingly important for telcos and why we believe a new approach is necessary; the key elements of such an approach and how multiple telcos are already using it to achieve encouraging results; and what it takes for a telco to begin to use AI in this way.

The growing relevance of CX for telcos

Today, CX is the key differentiating factor for creating value in telecommunications. Our research shows that 73 percent of senior telco executives see it as a top priority. In our surveys, the same senior telco executives told us that current methods do not provide specific information about what part of the customer’s experience influences their perspective, how to link tactics to variations in CX, or how to associate satisfaction scores with business outcomes. The result is that operators lack a granular view of customer experience, to the point that an operator can have the best overall network performance in a given market but not understand that certain customers in that same area are actually having a negative experience.

Existing scores that track network customer satisfaction have several shortcomings. Survey-based scores tend to be overly subjective, relying on customer memory and perceptions of what they experienced, and they tend to skew toward negative experiences while overlooking important factors such as the customer’s choice of device or Wi-Fi setup. The data they produce include lagging indicators that cannot offer real-time insights and are based on sampling. Internally developed metrics (such as time above a certain throughput) can also fall short, in that they can be biased to what is easily measured, rather than what actually affects customer decisions. And last, the correlation between those metrics and business outcomes such as churn and growth is limited. All those factors limit the range of applications that a more complete score could offer, particularly in the commercial realm.

By contrast, a robust, AI-enabled CX score explains commercial outcomes. In software and commercial contexts, CX takes into account emotional, cognitive, sensory, and behavioral factors, as well as brand perceptions, and the effect each of these has on customer behavior and the likelihood that customers return to use a product. In the network context, CX refers to the network performance factors that affect customer behavior. By leveraging detailed network performance information and AI, it is possible to produce highly granular and specific CX measurements, for example, for every single line every day, or even more detailed, every 30 minutes. This allows telcos to move from “macro” to “micro” analysis, and tie network performance to specific customer behaviors. For example, rather than assessing the average latency for all customers in a New York City zip code last week, they could determine the latency for an individual customer last Tuesday between 10 and 10:30 a.m., their likely perception of that latency rate, and precisely which interventions make sense from an ROI perspective.

With this level of precision and objective metrics, telcos can understand better which actions can lead to more positive business outcomes, and thus how to improve customer value management (churn prevention, for example), strengthen network operations (including issue detection and prioritization), and refine capital allocation. At scale, it can also be used more broadly to inform most decision making and transform the network into a key driver of value creation (Exhibit 1).

Exhibit 1

Building effective telecom network CX scores requires addressing five different dimensions.

DimensionA good score should be able to…Example
Level of granularity…provide CX measurements at the level required by the use cases to be addressed. Typically two aspects are considered: unit (score per customer, per site, per zip code, etc) and frequency (daily, weekly, monthly). The finer the granularity, the better.CX measurements for every customer line andevery site for every day.
Correlation with business outcomes…differentiate those customers who are getting what they expect from the network from those who are not, and connect the negative experience to a specific business outcome (eg, increased customer churn, increased site alarms). Customers with the lowest scores will have different commercial outcomes from those with the highest scores.30% of clients with the worst experience churn ~3x more than the average, and account for 60+% of all churners.
Transparency and interpretability…provide a clear explanation of why a customer’s score is what it is, usually leveraging subscores, and what needs to change to trigger improvement.Scores are composed of ~5–7 subscores such as reliability, accessibility, capacity, voice, etc, each with clear values and weights.
Sources of insights…leverage a broad range of information sources related exclusively to network performance and combine them to get a holistic picture of the customer interaction with the network. Only KPIs that the operator can modify should be included. For example, capacity is a good score input, while “time spent consuming video” is not, since the operator cannot modify it.A best-in-class example would leverage customer-level metrics (aggregated from session-level data) with site-level KPIs and handset-generated data. RAN, backhaul, and core information should also be considered.
Customer personalization…tailor the relevance of the different KPIs to the preferences of the customer and the way they use the services.Scores for customers who do not have a 5G device should only leverage previous generations of KPIs; similarly, voice KPIs should not be used for customers who only use data.

One telecommunications company was able to use this method of measuring network CX to identify customers 2.2 times more likely to churn than peers with better network experience (Exhibit 2). By combining those network CX measurements with the company’s existing churn model, it was able to improve the overall identification of potential churners and differentiate between those customers with both network issues and commercial problems (such as pricing issues) and those only dealing with major network problems (low churn risk in the churn model, but low network CX measurements) (Exhibit 3). This was extremely relevant because those two profiles call for different actions to prevent future churn. If network performance is good while CX is still poor, then the response needs to be commercial—for example, with discounts or a free 5G mobile phone. If the customer’s commercial experience is good, but the network performance is poor, then a free phone won’t help, and without upgrades to the network, the customer will likely leave. Using the defined, AI-based CX approach, the company succeeded in dropping its network-driven churn rate significantly. In the specific case, the network CX experience was being measured daily, for 100 percent of lines, and an average of 600 RAN sessions per customer line were analyzed.

Telco customers with lower network customer experience (CX) scores have a higher probability of dropping their provider
AI-enabled network customer experience (CX) data can help enhance telco churn models and reduce overall churn.

Value creation and enhanced decision making

The many opportunities to leverage improved network CX scores across the telecommunications organization fall into two main categories: network management (including network planning) and commercial management. Within these, the CX scores can be applied to multiple use cases such as churn modeling, capital allocation, energy optimization, or cross-sell and upsell campaign fine-tuning, among others.

Network management. In the network management space, the sophisticated scoring system can help identify not only where CX can be improved to continually create business value but also what precisely needs to be done to make those improvements. Some examples of specific actions include:

  • Smart capital expenditures and planning optimization. To increase the ROI of investments in the network, smart capital planning, or smart capital expenditures, can be implemented by leveraging customer experience scores. With a clear understanding of the relationships between network performance and commercial outcomes, allocating capital budgets can advance from educated guesstimates to precise targeting. The results from potential interventions in different areas (with different competitor dynamics) can be estimated more accurately and translate the effect on commercial outcomes. Following this approach, operators can better understand what interventions to take in every site for the capital plan (for example, a MIMO, or multiple-input, multiple-output, upgrade to enable more reliable transmission of data, versus a sector addition) and which sites to prioritize based on the customer impact.

    One telco was able to identify 15 percent of investments within its three-year build-out plan that would achieve a better ROI with retargeting either because its initial plans addressed areas with already high scores, because its proposed solution did not address the driver of the low score, or because the number of customers and amount of revenue affected was too low to warrant the planned investment.
  • Network performance management. Having a strong understanding of the customer experience that is being delivered and the drivers behind it can be an unprecedented opportunity for network organizations to speak the same language. This would hopefully put a stop to long-running debates over whether poor commercial outcomes are driven by commercial considerations or network problems. With a single network CX score and clear linkages to commercial outcomes in their tool kit, telcos can now assign employees to own the tracking of overall scores and subscores, allowing for improved performance management. For example, operators can understand which areas are more affected by network capacity and which are more sensitive to network reliability (based on customer preferences), or where a voice experience is more relevant than a data experience, and how well they are doing on those dimensions. This knowledge can help inform the actions required to upgrade the network performance of those areas.
  • Anomaly detection. Just as telcos routinely monitor alarms and KPIs in the network operations center, they can monitor CX to identify strange events or patterns that could signal critical network faults. Monitoring network CX this way brings two benefits over traditional network monitoring, which consists of alarms for individual network elements and reliance on customer inquiries. First, it helps identify complex, or hidden, faults. Since a network CX score is composed of multiple performance metrics simultaneously (voice and data service, capacity, reliability, et cetera), it is affected by network faults that touch multiple KPIs simultaneously. This enables another layer of refinement to identifying issues. For example, a 10 percent drop in capacity at one site might not appear relevant, but a simultaneous 10 percent drop in all network performance indicators can indicate a major problem.

    The second benefit CX-driven anomaly detection has over traditional network monitoring is that it is capable of pinpointing business-relevant faults. Because CX scores are correlated with business outcomes, a score drop indicates that customers may have cause to churn, contact the call center, or complain on other channels. This can redefine how operators view “network problems” and the criteria to trigger alarms. For instance, a 15 percent throughput drop might not even be considered an event to resolve, but if an operator knows that the drop will cause churn probability to double in the area, the need to respond takes on much greater urgency.
  • Targeted outage repair. The impact a network outage has on CX can be used to evaluate the severity of business outcomes stemming from an already identified fault. Such repairs can be integrated into existing processes to change the prioritization of how outages are resolved to minimize the effect on customers’ perceptions of the network. One telecom operator was able to identify a backhaul problem affecting multiple sites by monitoring CX scores that showed trouble before traditional alarms were triggered or specialist teams began to investigate the problem. In this case, because the backhaul was not working properly, customers were not able to connect to the network, but capacity KPIs appeared to be okay, no RAN alarm was triggered, and phones were still connecting. Customers called about the problem, but engineers didn’t know the root cause. While some degradation in a fiber link was evident, this was not initially viewed as the problem. Only an AI-enabled CX score could synthesize the sheer volume of metrics required to pinpoint such an issue. When the CX score signaled an issue, engineers were able to drill down into subscores and individual KPIs to identify the specific backhaul problem, as well as its severity (including the number of customers affected). Manually reviewing the vast number of metrics for the various network elements involved in this identification process would have been practically impossible.
  • Energy optimization. By basing energy optimization efforts in part on CX scores, telcos can maintain or improve individual customers’ experience of the network while also improving the operator’s overall energy usage. For example, power to a site can be reduced if doing so would not degrade CX, or increased if CX scores indicate a measurable improvement would result. This level of optimization can be augmented further based on the types of customers using the site at a given time and their churn likelihood.
Abstract 3D representation of artificial intelligence: a stylized silhouette of a head with a pixelated brain placed atop a cell phone, surrounded by a network emanating from the head.

Why AI-enabled customer service is key to scaling telco personalization

Commercial management. Opportunities for using customer experience scores to improve commercial management include knowing whether and how to invest in improving the network, deciding which customers should be offered a discount or a new phone to mitigate the effects of poor experience, increasing average revenue per unit (ARPU) at the individual level, growing specific markets, and optimizing commercial operations. Some examples of specific actions are as follows:

  • Enhanced churn models. As much as 38 percent of churn is driven by network issues, with customers seeking better data speeds and voice reception.1 Research has found that in the top half of markets analyzed, where network quality is highest, the lead service provider enjoys a 31 percent higher ARPU and 27 percent lower average churn than its peers.2 Churn models also can be enhanced with AI-driven CX insights, with targeted churn reduction actions such as promotions, plan changes, and moving customers from prepaid to postpaid.
  • Cross-sell and upsell advances. It stands to reason that customers who enjoy better network experience are more likely to purchase additional products and services. But the network is typically not considered when launching cross-sell and upsell efforts. The failure to take this critical factor into account when launching those efforts not only reduces the effectiveness of those campaigns but might also create a risk: launching a cross-sell or upsell effort on customers with poor network experience might backfire and trigger a negative reaction. By integrating CX scores in those marketing processes, these sales initiatives are likely to be more successful.
  • Early intervention on connectivity issues. Individual customers dealing with a poor network experience could be offered femto cells (small, low-power cellular base stations) or mesh units to improve their connectivity. This can achieve three to four times as much churn reduction over more reactive measures based on customer complaints alone. Focusing solely on customer complaints can lead to driving discounts and giveaways while overlooking a full understanding of the root cause of the complaint and long-term resolution.
  • Proactive client communication and call routing and prioritization. Building on the anomaly detection use case explained earlier, once a customer’s experience drops significantly, the telco has the ability to proactively reach out to the affected users. For example, if a tower is knocked out by an unavoidable event such as an accident or weather, customers could receive a message informing them that service will be diminished temporarily. Alternatively, call center agents can use the same information to acknowledge the degradation in service when a customer calls to complain, and thereby save both the customer and the representative the time and frustration of figuring out what the problem is.
  • Device or plan upgrades. Customers could also be proactively offered upgrades if they are still using fourth-generation (4G) LTE devices in areas where fifth-generation (5G) is prevalent. Customers who have 5G devices but 4G LTE plans could be upgraded to a 5G plan.

Getting started

As with most data-related endeavors, building a road map for developing improved network CX scores is critical to success. Four key elements are involved:

Decide use cases. Knowing what you want to do with CX scores will determine what data is needed, so deciding the key use cases where the score will be used is step one. Churn reduction, optimizing capital budget allocation, or optimizing call rerouting are good places to start. The data required will change slightly depending on the use case, as will some modeling design choices (for example, churn-related use cases benefit from a score that helps identify customers experiencing very poor CX, rather than a score that differentiates between middle and high CX, such as might be required by service-level agreements for enterprise or government contracts).

Assess existing performance data. Understanding what network performance data is currently available is also important. The more sophisticated approach to CX scoring has demanding data requirements involving highly granular information that, in some cases, is not readily available. For most telcos, some amount of new data preparation (including creation and collection) will be necessary, and this can easily become overwhelming. As digital transformation proliferates, telcos’ data maturity has improved markedly. Still, this remains a massive challenge for telcos, whose core product—the network—generates staggeringly vast amounts of data every day. And every operator and its OEMs collect data in different ways, which becomes even more complex if they acquire or merge with another operator.

At a minimum, building an AI to use CX in this new way will require network performance data at a customer level—either per session or by combining meters and customer connections to sites, at least every hour, for all customers and sites as described in the defined journey (4G phone lines, for example). A minimum of 60 days’ worth of data stored is also needed. This data will be combined with commercial information, which might include churn, calls to customer service, or subscriber plan.

Build on any past efforts. Many operators have tried to develop similar metrics in the past, prior to today’s AI advances and data maturity achievements. Those previous efforts can still be helpful, as the data investments involved further the goal of deriving clear insights from vast data collections and can be combined as inputs alongside additional data points to produce these AI-driven CX measurements.

Define the journey. As with any other data and analytics effort, the journey is more likely to succeed if it is done progressively. Starting small to achieve initial successes and then building on those with increased complexity is usually better than tackling the most complex and granular ambitions first. Once an initial CX-based score is developed for a use case, more and increasingly complex scores and use cases can be developed, in a process of successive releases, akin to software development. For example, operators can start their journey with a focus on understanding the daily experience of 4G phone lines before moving to hourly IoT CX measurements. This approach can reduce the number of data sets that need to be combined at first, and the complexity involved.

The opportunity to adopt an objective, AI-enabled method for tracking and optimizing customer experience as a key driver of decision making is a ray of light during the prolonged era of challenging economics for telecom operators. Anticipating and preventing the problems that diminish customers’ experience of the network, and linking business outcomes to that experience, can be a source of renewed growth and innovation for the industry. This is only the beginning of how network operators can use AI to their benefit, however. Using AI to develop a deeper and more precise understanding of network CX can serve as a first milestone in a broader AI network transformation.

Explore a career with us