Understanding the Predictive Relationship Between Wireless Network Experience and Customer Churn

Wireless networks have transformed society, and their evolving capabilities support numerous economic and social goals. Continued investments in wireless technologies and network coverage by mobile network operators (MNOs) depends on their achieving a favorable return on investment. Retaining existing customers by reducing churn is the most financially efficient means to improve economic returns, however the causes of customer churn are not fully understood due to the lack of studies on the relationship between wireless network quality and customer churn. This quantitative study leveraged existing secondary data from one MNO containing nine network quality factors, which formed three network quality constructs in the model, measured in 646 counties for one month in combination with a partial least squares structural equation modeling methodology to evaluate the predictive value of the network quality factors on customer churn. The study further assessed the degree to which the predictive value varies based on the population density of each county. The surprising findings suggest that network quality does not predict a meaningful amount of customer churn (p < 0.001), but that the relationship between the network quality factors and customer churn differs meaningfully depending on population density. These findings suggest ways that MNOs may fine-tune their network investments to achieve improvements in customer churn and financial results.

Introduction

Wireless communication services are an important part of the modern digital-enabled society, powering economic growth through increased productivity, connecting individuals and communities to each other, and enabling education and the delivery of information. The Cellular Telecommunications Industry Association (CTIA) [1] noted that wireless services are responsible for generating enormous consumer benefits and contributions to the economy. The range of benefits provided by wireless communication services make it clear that they are a critical part of the modern economy. These services are enabled by modern mobile networks built and operated by telecommunications companies who invest heavily to deploy wireless networks, requiring a return on their investment to repay debt and deliver ever evolving services. The networks themselves are composed of complex computational systems designed to form networks of access points, aggregation points, and interconnection points, organized in multi-layer hierarchies to deliver scalable service capacity concurrent with high service reliability. This complex system is shown in Fig. 1.

Fig. 1. Wireless network architecture.

The return on investment for mobile network operators (MNOs) is most often in the form of monthly consumer subscriptions for wireless service, however based on market competition providers often pay customers to sign up for service, in the form of device subsidies or other costly incentives. The economics of this provider-customer relationship depend heavily on retaining customers for as long as possible by avoiding customer turnover, or “churn” [2]. Customer churn depends on many factors including pricing, brand positioning, service quality and social network effects, among others. As consumers evaluate wireless services and select from competitive offerings in their local marketplace, they consider different criteria to identify which services and operators will be best for them [3], [4]. While there are several aspects of service to consider, one important consideration for consumers is the quality of the network provided by the wireless operator [5].

This research work is concerned with quantifying the relationship between wireless service quality, measured through network experience key performance indicators (KPIs), and customer churn. MNO economics are particularly sensitive to the percentage of customers leaving the carrier, or churning, each month because the cost of promotional and other incentives to gain new customers is often measured in the thousands of dollars per new customer depending on the season and market dynamics [6]. Retaining active customers with service relationships is an important part of MNO economics, thus MNOs must understand the causes of customer churn. Several studies have sought to identify the causes of wireless customer churn, which include factors such as price, coverage, customer service, social network effects, and network service quality [7], [8]. The existing literature shows hypothesized relationships between measures of network performance, which form the network quality constructs, and customer satisfaction which suggest that the consumer experience of the network may inform customer decisions to remain with or churn from their MNO [9], [10].

A quantitative non-experimental study using a partial least squares structural equation modeling methodology is employed to model and analyze the relationship between three specific wireless network quality constructs, formed by a total of nine network parameters, and the rate of customer churn. The three network quality constructs, shown in Fig. 2, collectively represent the foundational aspects of a customer’s experience of a wireless network. Coverage density reflects the amount of wireless network infrastructure in an area, network strength represents the ability of customers to access the most widely deployed fourth generation (4G) network technology, and network experience captures key elements of service quality including data download speed and whether voice calls are reliable. The original contribution of this research is to quantify the degree to which differentiated wireless network experience contributes to customer churn, thus enabling MNOs to invest more efficiently for customer retention.

Fig. 2. Conceptual framework.

Related Works

Wireless Industry Economics

One important aspect of wireless connectivity is the connection to data services and the internet, enabling a range of benefits from the prosaic such as obtaining library books online to the meaningful such as obtaining remote telemedicine access to healthcare. The amount of wireless data that people use continues to grow rapidly at a pace that is accelerated for many reasons, including the pandemic-driven shift to remote workplaces [11]. Increasing internet usage is directly tied to the availability of high-speed wireless networks deployed by MNOs [12]. This increase in wireless usage creates challenges for MNOs who must determine how best to allocate their investments to serve rapidly growing customer needs in a manner that delivers the economic returns necessary to enable continued investments in new services, network coverage, and advanced technologies. Research and development investments in particular deliver long-term improvements to productivity and economic growth which are important to society, but these investments can be made only by companies which are themselves financially healthy [13].

MNOs are subscription-based businesses, with economic models that are dependent upon maintaining a base of customers who regularly pay for their services [14], [15]. Over time these subscription payments enable MNOs to recover the up-front costs of acquiring the customers, after which point the subscriptions yield operating margin and cash flow sufficient to enable repayment of debt incurred to purchase spectrum and deploy network services, and then to invest further in their networks and systems. In highly competitive saturated markets like wireless services, it is more economically efficient, and thus more important, for companies to retain existing customers than to invest in the acquisition of new customers, increasing the importance of understanding and reducing churn [16].

Customer Loyalty

Numerous factors contribute to customer churn, including many which are exogenous to MNO control, such as the behavior and decisions of the customer’s social network or the development of regulatory support programs which defray the cost of wireless service for low-income consumers [4]. Substantial uncertainty remains with respect to quantifying the factors which impact customer satisfaction with their MNO [7]. While service pricing demonstrates the clearest linkage to customer churn, other notable factors which affect loyalty include data throughput, voice service quality, and network coverage [3]. Network data speed and voice call quality are frequently identified as directly influencing perceived service quality [9].

Several studies have shown the importance of wireless network performance to measurements of service quality as perceived by customers, with factors such as reliability, network quality, and call quality all being identified as closely related to customer satisfaction [17]. To leverage service quality measures to improve customer retention companies must model the relationship between the most important measures and their impact on customer loyalty decisions, enabling them to more deeply understand the extent to which each measure predicts resulting customer behavior [18].

Measuring Wireless Service Quality

Wireless network factors, including coverage, call quality, and network quality, have been identified as having an impact on customer satisfaction with their wireless experience [7]. As researchers explore the nature of service quality and its impact on customer loyalty, they have found that halo effects exist, wherein improving customer perception of one aspect of service quality can lead to improvements in their perception of related services [19].

Several aspects of service can lead to competitive advantage, including service quality, value and satisfaction [5]. Customer demographic factors, subscription pricing, and the amount of mobile use are the largest contributors to customer satisfaction, however network measures have a direct relationship to customer perception of service quality [20].

Wireless network performance is measured by KPIs which are produced by network elements in abundance, with typical networks reporting on hundreds of KPIs per network node or cell site [21]. Network KPIs are the single best tool that MNOs must evaluate the quality of their network [22]. The KPIs most deserving of study, with respect to their impact on the customer experience, include measurements of mobility, such as dropped calls, measurements of integrity, including average download throughput, and measurements of availability, which reflects the accessibility of the MNO’s network [23]. Data connection download speed is the primary measurement of wireless quality for mobile broadband service and is an important factor in the degree to which the benefits of wireless service are realized by customers and communities [24]. Dropped calls represent active voice conversations which are terminated abnormally, without either participant intentionally ending the call, while integrity defines the quality of services for their intended use, and availability describes the ability of users to access services based on network coverage or reliability. Access to more recent generations of wireless technology has a direct benefit to the customer experience, though the most current fifth generation (5G) networks are available to less than half of consumers [25].

Customer Churn Analysis

The availability of big data collected by MNOs in their normal operations is enabling experimentation with several approaches to churn prediction leveraging statistical analysis and machine learning tools [26]. Big data collected by MNOs during operations can contribute to customer churn analysis and prediction [27]. MNOs are increasingly building complex models with their existing data to deeply understand the causes of customer churn and important levers to reduce churn, however predicting customer decisions is very challenging [16]. Statistical models are evaluated according to performance measures, such as accuracy, precision, and recall, which describe whether the model in combination with the selected factors yield a useful means for predicting future outcomes [28].

Data mining is an effective approach to predicting churn which is increasingly being adopted by MNOs, but this methodology requires significant analysis to determine the appropriate predictor variables which are associated most strongly with changes in customer behavior [20]. Many different models have been developed to predict churn risk; however, these models do not always align with the goals of MNO managers to improve profit rather than simply decrease churn, considering that not all customers are equally valuable to the company [2]. Clearly, big data analytics is a strategic capability that can be used by MNOs to gain deep insights into complex business issues prior to making formal decisions or commitments, helping to differentiate their own services from those of competitors while reducing business risk and improving financial outcomes [29].

Problem Statement, Hypothesis Statements, and Research Questions

Problem Statement

The problem to be addressed by this study is that the causes of customer churn in the wireless telecommunications industry are still not fully understood due to the lack of comprehensive studies on the relationship between three wireless network quality constructs and the rate of customer churn.

Hypothesis Statement

• H₀: The three network quality constructs cannot predict the rate of customer churn.

• H₁: The three network quality constructs can predict the rate of customer churn.

Research Question

Can the three network quality constructs predict the rate of customer churn?

Methodology

Method

The research method selected for this study is a quantitative non-experimental correlational analysis leveraging partial least squares structural equation modeling (PLS-SEM). The research method selected for this study is based upon the quantitative paradigm which utilizes observations in the form of empirical values in concert with statistical analysis to describe the phenomena being examined [30]. The quantitative paradigm is part of the postpositivist philosophy which suggests that researchers cannot be certain (or positive) about their claims, but rather that “causes probably determine effects or outcomes” [31]. This tradition demands the reduction of complex hypotheses into discrete questions which can be empirically measured and tested. This research study will utilize a non-experimental correlational approach, in which the use of secondary data results in the inability to manipulate the network experience parameters, leading to the identification of possible relationships between variables [32].

A conceptual framework for a quantitative research study combines the planned research parameters with a theoretical model predicting the expected nature of their relationships [31]. This approach ensures the research study is grounded in existing literature and reflective of previously developed frameworks and theories which are suitable for the planned research [33]. Numerous measures of service functionality exist to quantify each aspect of how a customer accesses and uses the mobile network, known as wireless network parameters [7], [23]. Factors such as how customers perceive their network experience, as compared to the factual data gathered which measures the actual experience, and customer satisfaction, which is informed by factors beyond the network experience such as word-of-mouth, the value of the service, and brand qualities, are identified as likely mediating factors but excluded from this study [34]. Customer experience of the wireless network will be categorized into three experience constructs, formed by the nine network parameters or measurement variables.

The conceptual framework for this study, shown in Fig. 2, adapts the SERVQUAL framework to the partial least squares structural equation modeling methodology, which in this framework is composed of three network quality constructs which will be examined in this research study. This framework enables a deeper understanding of how differing customer experiences of the network, reflected in the network parameters across different network quality constructs, or latent variables, correlate to customer loyalty measures, which directly impact MNO economics and subsequent network technology investment prioritization.

Population and Sample

The research study will be based on secondary data analysis, which reduces or eliminates the time and cost of primary data gathering but limits the researcher to existing questions, populations, and time periods based on the available data [35]. The PLS-SEM study design supports the analysis of large quantities of multivariate data, and the use of large data populations when available will increase the reliability of the prediction [36]. This study will rely upon secondary data gathered by an MNO during normal operations for the primary purpose of network and customer management. The study population will be comprised of all postpaid mobility customers of this MNO, which necessarily excludes customers of services such as fixed wireless broadband as well as connected devices or Internet of Things devices, for the selected study period. Throughout the selected study period the MNO reported having approximately four million mobility customers. This data will be extracted in aggregated form by county to ensure no personally identifiable information (PII) is reproduced or exported into the study dataset. The dataset is comprised of one month of data, from October 2023, for each of the nine selected network parameters, or measurement variables, grouped by approximately 664 counties in which the MNO actively offers services to customers.

Data Analysis

Data for the study was collected through common industry-standard systems which generate network performance parameters and metrics as an intrinsic function of the operational support system (OSS) required to operate and manage a wireless network. Customer churn data was collected through the MNO business support system (BSS) which is used to facilitate customer account activation, provisioning, billing, and record-keeping. The network metrics and customer churn data were extracted from the OSS and BSS, respectively, into comma delimited files which were subsequently cleaned and transformed manually using Microsoft Excel for use in the analysis, and then further analyzed using SmartPLS 4.0 as shown in Fig. 3.

Fig. 3. Research analysis steps.

The monthly network data was combined with data in UScellular’s customer relationship management (CRM) system, which uses a vendor-proprietary data table framework containing customer transactions, including termination of their service contract. The CRM data table export was aggregated to reflect customer churn statistics, calculated prior to data export as the number of customers terminating their service with the company in each month and in each county divided by the average number of customers obtaining or retaining service from the company in each month and in each county. Exported data shows customer churn figures by county to avoid the exposure of customer PII. The network and CRM datasets can each conceivably include millions of records representing millions of customers and months of data collection and were thus limited to a single month for analysis, October 2023.

Big data statistical analysis methodologies, including PLS-SEM, yield significance levels for model conclusions based in part on the amount of data used in the analysis [36]. This study will leverage the full dataset for all wireless network parameters, including those indicating technology investment density and customer churn, to maximize the precision and consistency of the prediction model. Incorporating data from all the counties in the full dataset further enables multiple analyses to evaluate whether the density of network technology investment alters the causal-predictive relationship between the network parameters and customer churn. A data sampling approach evaluating a subset of counties was considered as a mechanism to reduce the analysis time while maintaining the target significance level, but this approach was ultimately not selected to achieve maximum fidelity of the resulting model. Similarly, an approach incorporating multiple months of data in a longitudinal PLS-SEM model was considered, however higher order models, or hierarchical component models, were deemed to be beyond the scope of this study.

Results, Interpretation and Applications

Results

For the research question and hypothesis, the analysis suggests that the indicator variables have a statistically significant (p < 0.001) but very small explanatory value (R² = 0.076) for changes in customer churn, and little predictive value for this relationship (Q² = 0.076). The findings test Hypothesis 1 and indicate that it should be rejected in favor of the null hypothesis, that the nine network indicators and three network quality constructs cannot predict the rate of customer churn. Full bootstrapping analysis results are shown in Fig. 4.

Fig. 4. Results of bootstrapping analysis.

Interpretation and Applications

This study of the causes of customer churn in the wireless telecommunications industry, specific to three network quality constructs, adds substantial new knowledge to the published body of knowledge associated with wireless subscriber churn. Specifically, the research study question and hypothesis evaluated the predictive value on customer churn of three network quality constructs, formed from nine commonly measured network factors [23]. Based on the conceptual model and the analytical methodology variances in the three network quality constructs explains only 7.6% of the variance in customer churn decisions leading to a determination that these constructs cannot meaningfully predict the rate of customer churn. This surprising result suggests that the selected network quality factors have little impact on customer loyalty decisions, and that other factors which were not studied play a larger role.

This study presents evidence that improvements to network quality factors are not broadly predictive of improvements to customer churn. Much of the existing body of knowledge associated with wireless customer retention focuses on factors such as company brand, competitive offers, service pricing, and promotional incentives [7], [8], [17]. This study suggests that these non-network quality factors may have a stronger predictive value on customer churn than the studied network quality factors. This may be a result of wireless network maturity as the industry reaches a point where each network in a geographical area is nearly indistinguishable from others from network coverage, network strength, and network experience perspective. A competitive environment in which all MNOs offering service in an area are effectively at network quality parity might explain the apparent indifference of customers to network quality factors when they are making churn decisions.

Limitations

Every study is bound by constraints, necessarily including some specific considerations and variables while excluding others. In a similar fashion every study faces challenges in the design, method, or execution which may limit the accuracy or extensibility of its conclusions. These factors are considered the study limitations, reflecting the potential weaknesses of the study resulting from its design and the external conditions which may degrade the value of its findings [37].

The key limitation of the study is the data set used for the analysis. Secondary data gathered by the researcher’s employer was used for the analysis which prevents expansion of the study to additional factors or extension of the study to alternate time periods of data. Selection of the data available from a single MNO risk limiting the extensibility of study conclusions to the degree the relationship between the measured network parameters and measurements of customer churn differs greatly at other MNOs. Neither historical data nor longitudinal analysis were used for the study, which also necessarily used a dataset from a year prior to the completion of the study. This limitation was mitigated by ensuring the selected dataset was comprehensive and fine-grained with respect to the selected factors, including data by parameter and county. The time-bound nature of this study, required to fit within a doctoral study program, is a further limitation, and was mitigated through a research design which was structured to complete data analysis within the boundaries of the program.

Recommendations

The surprising result found in evaluating the research question and hypothesis yields multiple avenues for productive future exploration. First, a longitudinal model, incorporating multiple months or years of data to counteract the potential effects of seasonality, would be valuable means of further confirming this studies results. The assessment could be completed using a higher-order model or hierarchical component model for PLS-SEM analysis and would address this study’s limitation of using only a single month of data. This type of follow up study would benefit from using either the same month across multiple years, or multiple consecutive months extending for up to a year, to demonstrate results that reflect the longer-term relationship between network quality factors and customer churn.

This study utilized secondary data from a single MNO. Future studies might evaluate the extent to which the predictive value of network quality is associated with the degree of MNO competition in a county. This relationship could be further evaluated by enriching the dataset with network quality factors and customer churn data from multiple operators, though this is likely to be challenging considering the competitive nature of this data.

Finally, this study did not consider the different relative economic value of customers in counties with different population densities. MNO investment should not be based solely on customer churn but should also be informed by the lifetime value of customers, insofar as an investment which improves the retention of lower economic value customers is less beneficial to the MNO than an investment which improves the retention of higher economic value customers. Incorporation of average revenue per user or average revenue per account measurements into the dataset might support an analysis more focused on the economic implications of network quality investments in different counties [2].

Conclusion

The purpose of this quantitative non-experimental study was to determine the relationship between three specific wireless network quality constructs, formed by a total of nine network parameters, and the rate of customer churn. The problem addressed by this study was that the causes of customer churn in the wireless telecommunications industry are still not fully understood due to the lack of comprehensive studies on the relationship between three wireless network quality constructs and the rate of customer churn. This problem was addressed through a partial least squares structural equation modeling analytical methodology using one month of secondary data from a single MNO.

The study provides evidence that, overall, wireless network quality factors do not have substantial predictive value on customer churn (7.6%, p < 0.001). The implication that network quality cannot be used to broadly predict customer churn is a surprising outcome suggesting that at this scale networks are not the primary driver of customer churn decisions.

Future research should assess these results further, extending the study to include all types of devices and customers (e.g., prepaid, Internet-of-Things, enterprise accounts) over a longer time to validate this study’s findings more broadly. In addition, future studies should include data from multiple MNOs for the same time and geographical area. This would enable an assessment of whether the findings apply to multiple MNOs and would also support the evaluation of predictive value based on the level of network quality competitiveness in each county. Finally, the inclusion of customer economic value data, such as average revenue or lifetime margin per customer, would enable future studies to assess the relative value of improvements to customer retention supporting further targeting of investments to maximize economic returns.