Winning your customers’ minds and hearts: Disentangling the effects of lock-in and affective customer experience on retention

Experiential learning theory (Kolb, 1984) proposes that individuals learn through the experiences that they obtain from all parties, ranging from various product categories to multiple firms, including the competing alternatives. Such experiences can serve as a basis for reflection that allows the individual to obtain a wide range of information about various product categories provided by different firms (the reflection process). This information is later assimilated and distilled into abstract concepts, including the general perception of the experience with the focal firm (the conceptualization process), which can serve as a guide for carrying out actions, including the decision to stay with the focal firm or to switch to a competitor (the experimentation process). This theory suggests that, for a specific product or service, the customer’s decision to remain in a current relationship or switch to a competitor will be affected by the experience of that particular product or service. We refer to this as the main effect of the affective customer experience on retention (Keiningham et al., 2020; Lemon & Verhoef, 2016). In addition, the theory acknowledges that customer retention in a product category can also be affected by experiences in other (related) categories with the focal firm, which we refer to as affective customer experience spillover effects (Balachander & Ghose, 2003; Danaher et al., 2020; Dong & Chintagunta, 2016).

Social exchange theory regards exchange relationships as ranging across a continuum from purely transactional relations (at one extreme) to reciprocal relationships (at the other extreme) (Day, 2000). Depending on the exchange relationships established, there are differences in how customers encode, reflect, and conceptualize the perceived experiences (Puccinelli et al., 2009; Witell et al., 2020). In transactional relationships, exchanges are based on formal binding agreements in which both customers and firms agree on the terms of the discrete exchange event that gives both partners the benefit of equal value (Molm et al., 2003). Lock-in thus occurs when customers remain in the exchange relationship because of monetary factors or economic incentives (e.g., bundling offers or binding contracts). Lock-in mainly promotes transactional relationships in which customers rely on calculative-based reasoning, which elicits an analytic, detail-oriented processing strategy through which they carefully weigh the balance between the profit obtained from continuing the relationship and the loss caused by leaving (Aggarwal & Law, 2005). Conversely, reciprocal relationships are derived from relationship depth, which is “the deepening of the customer’s relationship with the firm through increased usage” (Bolton et al., 2004, p. 274) and reflects “the frequency of service usage over time” (p. 273). A deep relationship usually results from an ongoing process of exchanges and multiple interaction events between a customer and a firm (Witell et al., 2020), whereby the customer enjoys the psychological comfort of maintaining the deeply established exchange relationship (Bolton et al., 2004; Witell et al., 2020). Customers engaged in reciprocal relationships usually think in a broader, more abstract fashion, focusing on experiential benefits (Puccinelli et al., 2009). A figure (see Web Appendix A) has been elaborated to demonstrate the linkage between experiential learning theory and social exchange theory.

Having identified the two central customer retention strategies (lock-in and affective customer experience with its main and spillover effect), this study is concerned with understanding their joint effects on retention. In pursuit of this objective, first, and most importantly, we disentangle the effects of lock-in and affective customer experience and assess whether they strengthen or weaken each other. Second, we capture how the separate and joint effects of these strategies are determined by the depth of the established relationship, thereby addressing the question of when these effects take place. Figure 1 sets out the conceptual framework and the aforementioned effects.

As suggested by Bolton et al. (2004), economic reward programs with their monetary benefits propositions may sound attractive to customers. Lock-in characterized by the monetary factors or economic incentives (e.g., bundling offers or binding contracts) are very likely to retain customers due to the significant anticipated loss of benefits or switching barriers (Calvo-Porral et al., 2017; Malhotra & Malhotra, 2013). As there is accumulated evidence in the literature about the positive effect of lock-in on retaining customers (Blut et al., 2015; Johnson et al., 2003; Nitzan & Ein-Gar, 2019), we do not put forward specific hypothesis but instead empirically test this relationship. We will check the main effect (i.e., if it is indeed positive) for face validity.

The importance of affective customer experience in retaining customers has long been recognized in the marketing literature (De Haan et al., 2015; De Keyser et al., 2020; Lemon & Verhoef, 2016). More specifically, this importance is reflected not only in the direct effect of delivering a good affective experience in one category offered by the focal firm, but also in delivering a good effect in a related category (Balachander & Ghose, 2003; De Keyser et al., 2020), that is, the spillover effect (Erdem & Sun, 2002; Janakiraman et al., 2009; Keller et al., 2020). Indeed, as noted previously, experiential learning theory clearly indicates that customer actions (e.g., switching decisions) are driven by grasping and transforming the concrete experiences acquired from all parties, including experiences from another product category. We therefore propose our first hypothesis:

Consistent with Bolton et al. (2004), relationship depth is one of the key indicators of relationship development. As the critical role of relationship depth in retaining customers, augmenting customer equity, and improving financial performance has been widely demonstrated by prior empirical research (e.g., Aurier & N’Goala, 2010; Burnham et al., 2003; Polo & Sesé, 2009), many firms have dedicated efforts in deepening current customer relationships. While the positive linkage between relationship depth and customer retention is straightforward, there is a lack of knowledge on how firm actions (i.e., lock-in) and the perceived affective customer experience may be assessed by customers under different relationship depth (Tarasi et al., 2013). As theoretically supported by social exchange theory, a deep relationship usually results from an ongoing process of exchanges and multiple interaction events between a customer and a firm (Witell et al., 2020). Depending on the degree of relationship depth, we expect that the way customers value economic-focused offerings and experiential benefits (Puccinelli et al., 2009) may also differ. Thus, it is essential to consider the moderating role of relationship depth under different situations (i.e., lock-in and affective customer experience).

To test the proposed conceptual framework and the associated hypotheses, we developed a set of multinomial logit models formulated using random utility theory (McFadden, 1973), with one model for each service category. This methodology allowed us to identify key determinants that affect customer retention probabilities across multiple firms (Elshiewy et al., 2017).

Following McFadden (1973), the model was derived as follows. Consider a set of customers I = {i|i = 1, 2, …, I} that faces a choice set of available alternatives, which can be denoted as M = {m|m = 1, 2, …, M}, from each of the two service categories S = {s|s = j,k}, where j refers to the mobile service category and k represents the broadband service category. The customers’ choices are observed over the period T = {t|t = 1, 2, …, T}, where T represents the observation window. From each of the alternatives, the customer would obtain a level of utility; let U_imt = {U_imjt, U_imkt} denote the overall utility in the mobile service category j and the broadband service category k that customer i would perceive from firm alternative m at time t. Researchers typically only observe actual customer choices and a set of attributes of the M alternatives (Elshiewy et al., 2017). Therefore, the utility of customer i for alternative m in each of the two service categories at time t is decomposed into the observable (deterministic) component and the unobservable component. The former is the true utility level perceived by customer i in the mobile and broadband service categories from the corresponding service provider m at time t − 1, which can be formulated as V_imt−1 = {V_imjt−1, V_imkt−1}. The latter is the error term associated with customer latent utility perceptions in the mobile service and broadband service categories, given as ε_imt = {ε_imjt, ε_imkt}, respectively; they follow an identical and independent (iid) Gumbel distribution. Most importantly, as noted previously, experiential learning theory suggests that customers tend to update their knowledge scheme through prior concrete experience within different product or service categories gained from the focal firm as well as from competitors. We thus assume that customers update the current overall utility level at time t on the basis of the previous affective customer experience in the mobile and broadband service categories received from the focal firm and observed from its competitors at the previous time period t − 1.

We specify the utility that customer i derives from firm alternative m in the mobile service j and broadband service categories k at time t in Eqs. (1) and (2):

Equation 1 is further specified in Eq. (2) via attributes to express the influence of lock-in, affective customer experience, and relationship depth in customer latent utility perceptions in the mobile service and broadband service categories.

Two dummy variables were created to indicate whether customers had responded to the survey question about their affective customer experience with firms in the mobile and broadband categories in December of each year. Controlmis_imt are the variables that control missing data relating to affective customer experience and binding contract. Firm_mt, Context_mt, and Demographic_it represent a vector of control variables including firm-related characteristics (market share, advertising expenditure, and social media mentions), context-related characteristics (acquisitions, new entrants, and iPhone release dates), and customer demographic information (gender, age, working status, and social class). Finally, as noted above, ε_imt is the error term.

In this study, we are especially interested in parameters β₃–β₈, which gauge the separate and joint effects of lock-in strategies and affective customer experience. We are also interested in parameters β₁₀–β₁₇, which measure the extent to which the impacts of lock-in and affective customer experience, and their joint effects might vary depending on the level of relationship depth. Of these, the parameters β₁₀–β₁₁ capture the moderating role of relationship depth in the linkage between lock-in and customer retention. In the same vein, β₁₂–β₁₃ correspond to its moderating impact on the main effects (i.e., direct and spillover effect) of affective customer experience on customer retention. Finally, parameters β₁₄–β₁₇ capture how the joint effects of lock-in and affective customer experience differ with the depth of the established relationship, that is, the three-way interactions across lock-in, affective customer experience, and relationship depth.

The multinomial logit model, as its function form reflects, captures the possibility of customer i choosing firm m instead of the alternatives. For the estimation of the logit parameters, the maximum likelihood estimation method was applied. In order to represent choice probabilities, Eq. (3) was elaborated as follows:

Let Y_imt = {f_it1, f_it2, …, f_itM} denote the index vector of the firm alternatives chosen by customer i for the mobile and broadband service categories j and k, respectively. Consequently, Pr (Y_imt) represents the possibility of observing the choice profile that customer i would choose firm alternative m across the M alternatives at time t in the corresponding service category. Following Elshiewy et al. (2017), this possibility is conditioned as follows:

To demonstrate the contribution of the variables to explaining the variance in customer retention, we applied a hierarchy approach, introducing different categories of variables set by set. In total, three models were estimated. Model 0 is the baseline model that examines the impact of the control variables. Model 1 adds the main effects of lock-in, affective customer experience, and relationship depth. Model 2 also takes into account the joint effects of lock-in and affective customer experience and the moderating role of relationship depth. The same set of models was estimated for the broadband service category, yielding six models in total.

We provide some model-free evidence of our findings in Web Appendix  B. To do this, we took the mobile service category as the reference category and used the unadjusted sample of 656,208 observations. We calculated the average switching rates, given the different situations in terms of (1) the main effects of affective customer experience, (2) the joint effects between lock-in and affective customer experience, and (3) the effects under the moderating role of relationship depth. Drawing on prior research (i.e., Fader & Hardie, 2010; Nitzan & Ein-Gar, 2019), the average churn rate was obtained by dividing the number of customers who left the providers by the number of active customers. In general, the patterns illustrated in Web Appendix B confirm the hypothesized relationships.

The results of the regression models (Models 0–2) are presented in Table 3 as a series of nested models. The fit statistics indicate that the addition of each set of variables improves the model fit significantly, confirming the incremental power of lock-in, affective customer experience, the joint effects between them, and their relative importance under the moderating role of relationship depth. Log-likelihood value and Akaike’s information criterion (AIC; Akaike, 1998) were used to assess the adequacy of the three models. With log-likelihood, the higher the value, the better the fit of the model to the data; AIC indicates that the model with the lowest AIC is the optimal option. To assess whether our estimates were affected by multicollinearity, we followed standard practice by computing variance inflation factor (VIF) scores for each regression. As shown in Web Appendix  C, each of the VIFs was below the recommended cutoff of 10 (the maximum score is 5.70), which suggests that multicollinearity did not severely affect the regression results (Hair et al., 1998). Additionally Web Appendix, D shows the correlations between the key variables, which do not indicate multicollinearity.

Although we did not put forward any hypotheses about the main effects of lock-in on customer retention, we empirically tested this relationship. Consistent with prior research (Blut et al., 2015; Johnson et al., 2003; Nitzan & Ein-Gar, 2019), the results demonstrate the positive impacts of lock-in (β^M₁ = 1.517, P < 0.01; β^B₁ = 2.624, P < 0.01; β^M₂ = 0.637, P < 0.01; β^B₂ = 0.156, P < 0.01) on customer retention.

As the theoretical direction of the influence of relationship depth on customer retention is not unambiguous, no hypothesis has been established for this linkage. Still, we empirically test the relationship between relationship depth and customer retention. The results confirm the findings of prior research, indicating that relationship depth positively and significantly affects customer retention (β^M₉ = 0.449, P < 0.01; β^B₉ = 0.170, P < 0.01).

We ran different models, including (1) several multinomial logit models with alternative specifications in key variables (i.e., affective customer experience, relationship depth, and customer switching) to check the robustness of the results; (2) one Tobit model to assess the ceiling effect of affective customer experience; and (3) a system of equations model via three-stage least squares (3SLS) to assess other intertwined effects among key variables. The set of multinomial logit models encompasses one estimation without control variables and three benchmark models whereby the measurement of relationship depth is replaced with alternative options (i.e., the customer’s bill as an alternative measure for relationship depth, customer tenure as relationship length, and relationship breadth reflected in “add-on” buying) (Bolton et al., 2004), one with an accumulative measure of relationship depth, and one with a balanced sample between churners and non-churners.

The Tobit model, as highlighted by prior research (e.g., McBee, 2010; Wang et al., 2008), is considered a potential way to deal with the ceiling effect. Here, the potential ceiling affective customer experience data were deleted for the estimation to assess the ceiling effect of affective customer experience. Finally, we developed a system of equations model using three stage least squares (3SLS), which accounts for correlations between equations (Amemiya, 1985). This allowed us to determine other intertwined effects among key variables (i.e., lock-in, affective customer experience, and relationship depth).

A problem with studying lock-in empirically is that it is often not exogenous. Firms target certain customers with lock-in offerings as they believe that these customers will accept the offers. Customers accept the offers because they see benefits in them (e.g., a lower price), but they might already have the intention to stay longer with the firm, which reduces the negative aspects of the lock-in. This endogeneity issue, as shown in Table 1, has rarely been addressed in previous studies of lock-in and affective customer experience. The few papers that have conducted an endogeneity assessment have simply developed a probit model or a structural model (e.g., De Haan et al., 2015; Dong & Chintagunta, 2016; Jones et al., 2007). In the present study, to control for the endogeneity bias across lock-in, affective customer experience, and customer retention, we adopted propensity score matching (PSM) via a greedy matching algorithm. This is a technique that has been widely applied in the literature to address endogeneity issues (Rosenbaum & Rubin, 1985) and has proved advantageous in many fields, including economics, medical studies, and marketing (Garnefeld et al., 2013; Rutz & Watson, 2019). The method has been used in prior research (Titus, 2007) specifically to address the problem of the limited distributional assumption of the errors inherent in the endogenous switching and independent variables estimation, which is similar to the situation under study here.

From the estimation results for the firm alternative specific intercept β₀ (Table 3), we found considerable heterogeneity in the intrinsic propensity to maintain the established exchange relationship with different firms. Unmeasured customer-specific factors may influence customer retention decisions. To account for customer heterogeneity, following the study of Gönül and Srinivasan (1993), we estimated two mixed multinomial logit models. In mixed logit models, customer heterogeneity is recovered by assuming that coefficients in the utility function are randomly distributed.

We adopted different approaches to address the issue of missing data. First, we used the multiple overimputation (MO) approach from Venkatesan et al. (2019) to replace missing values for affective customer experience. This method was applied by Venkatesan et al. (2019) in a similar situation, namely when missing customer mindset metrics (e.g., affective customer experience) were imputed to solve the missing data and measurement problem. Second, for customers who do not participate in the survey for a specific year, affective experience was imputed using the individual customer’s average score rather than the average values across customers when customers do not participate in the survey for a specific year. Third, to further tackle the issue of missing data, we developed a set of very conservative robustness checks. Among them, we estimated the model without observations in cases of missing values in affective customer experience. More conservatively, we estimated the multinomial logit models by keeping only customers who filled out the survey during the four-year covered study window. The results of the different approaches remain consistent with the patterns captured in the original multinomial logit models.

This study integrates experiential learning theory with social exchange theory to offer a comprehensive framework of the linkages across lock-in, affective customer experience, and relationship depth in customer retention. The results show that, given poor affective customer experience, lock-in helps firms to reduce customer churn. However, the impact of lock-in decreases alongside improvement in affective customer experience, and it becomes insignificant when affective customer experience reaches the threshold level. More importantly, lock-in, affective customer experience, and their relationship vary depending on the level of relationship depth, being stronger when there is low relationship depth and weaker when a deeper relationship has been established. The proposed framework and associated findings go beyond previous studies (e.g., Ataman et al., 2010; Kim & Kumar, 2018), allowing us to contribute to the literature in the following ways.

First, despite the merits of previous studies in advancing knowledge about customer retention, the two key marketing strategies for managing retention (i.e., lock-in and affective customer experience) have largely been studied separately (see Table 1), leading to a fragmented view of the role played by each of them. By taking into account the joint effects on customer retention of different types of lock-in situations and various affective customer experience effects (i.e., direct and spillover effects), we address this important research gap. As highlighted in prior research, a conceptual distinction between cognition-based and affect-based strategies is useful for theoretical investigations (Edwards, 1990; Panksepp, 2003) as a means of improving the fundamental understanding of the relationships of different strategies (lock-in and affective customer experience). Our study provides a nuanced understanding of the effects of lock-in and affective customer experience on retention. In this way, we respond to a highly relevant yet hitherto unanswered research question, namely whether lock-in (transactional) and affective customer experience (relational) strengthen or weaken each other.

The results of this study allow us to address two issues of managerial interest for marketing practitioners. The first issue concerns how lock-in and affective customer experience, as crucial strategies in pursuit of the same goal (i.e., customer retention), can be optimally deployed under different circumstances. The second concerns for whom firms should implement their actions. The proposed guidelines can be generalized and applied in a large variety of contexts, given the prevalence of these marketing strategies in diverse industries and different product categories (e.g., Ascarza & Hardie, 2013; Chen & Hitt, 2002; De Keyser et al., 2020; Malhotra & Malhotra, 2013; Nitzan & Libai, 2011).

We acknowledge several limitations of our study, and note that these indicate promising lines for future research. First, we measured affective customer experience using a single-item metric (the scale item of NPS in the mobile service category and a similar five-point Likert scale in the broadband category). Although simple measures are easily understood by marketing practitioners (Lemon & Verhoef, 2016), and the superior predictive power of NPS for customer retention in comparison to other perception metrics has been established (De Haan et al., 2015), we recommend that future research adopts other metrics. Second, we measured affective customer experience in annual terms in December of each year. In real business practice, as highlighted by Venkatesan et al. (2019), it is difficult and costly to collect affective customer experience information on a monthly basis, and yearly measurements are commonly used by companies. Nevertheless, future studies may seek to capture this information on a monthly basis, with a reduced sample size if necessary, in order to verify the proposed conceptual model. Third, we mainly focus on customers’ inner feelings derived from affective CX, as opposed to specific customer-initiated or firm-initiated factors regarding the affective-related information. Although, our database does not provide information about these factors. We do believe that determining the impacts of these factors (i.e., factors related to the affective-related information) would make for a very interesting complement to our study in the future. This information may enable companies to design activities to increase customer retention based on lock-in, affective customer experience, and relationship depth. We acknowledge the difficulty of tracking and implementing such a system. However, we believe that, given the availability of advanced information and communication technology, there is a huge opportunity for service providers to expand the array of detailed records of customer transactions and perceptions (e.g., affective customer experience) (Du et al., 2007; Venkatesan et al., 2019). Fourth, our results reveal affective customer experience measured by the scale item of NPS positively and significantly affects customer retention. However, we acknowledge while confirming such positive linkage, it is necessary to check for profitability, since in prior research, NPS has shown to be an ineffective measure in terms of predicting future profitability (Baehre et al., 2022). Fifth and lastly, future studies should collect information about customers’ perceptions of the focal firm’s competitors. Compared to a quantitative approach based on the difference between the experience gained from the focal firm and the average value for the rest of the firms, this offers a better reflection of the influence of competitors.