Exploring Nonlinear Relationships Between Individual-Level Bank Customer Satisfaction and Revenue

Abstract

This study examines the nonlinear relationship between customer satisfaction (CS) and both the levels and growth of customer revenue (CR) at the individual level in the banking sector. Utilizing a unique data on 19,054 Swedish bank customers (2013–2017), the analysis combines subjective satisfaction measures with objective financial and demographic register data. Regression models test for diminishing returns at high satisfaction levels while assessing the persistence of these effects over a four-year period. The findings indicate that while CS is positively associated with both revenue level and revenue growth, the relationship with revenue level is nonlinear. Specifically, customers scoring 80–89 generate higher revenues than those scoring 90–100, providing weak evidence of a ceiling effect (at the 10% significance level) that is notably absent for revenue growth. Furthermore, CS explains less than 1% of revenue variation, highlighting the inherent limits of satisfaction-based revenue models. These ceiling effects are more pronounced among older, lower-income women without debt, whereas wealth has no observable impact. Finally, the nonlinear effects fade after one year, though gender remains a consistent moderator. These tentative findings suggest limited financial returns from maximizing satisfaction, thereby supporting the implementation of more differentiated customer segmentation strategies.

1. Introduction

Determining whether the observed positive relationship between customer satisfaction (CS) and financial performance is linear or nonlinear has significant implications for both academic research and managerial practice. Previous studies have identified a nonlinear relationship between CS and customer revenue (CR) at both the firm level and individual levels within the banking sector (Ittner & Larcker, 1998; Yu, 2007), as well as other industries, including telecommunications (Ittner & Larcker, 1998), transportation (Mittal & Kamakura, 2001; Strydom et al., 2020; Gao et al., 2021), and tourism (How & Lee, 2021). Notably, there is evidence of diminishing returns at the upper levels of CS. Whether such nonlinearities also characterize the banking sector at the individual customer level using larger samples, longer time periods, and comprehensive control variables is a question this study aims to address.

While Segerlind et al. (2026) confirmed that high CS sustains CR without diminishing returns, they noted a need for further research. This paper builds on their work by exploring nonlinearity more granularly across CS groups. Specifically, we analyze both the level and growth of individual revenue using combined register and survey data from the banking sector. Furthermore, we examine whether there are nonlinear effects over a longer time horizon and identify the underlying drivers of these patterns. By controlling for demographic and financial factors, this study offers a nuanced perspective on the relationship between CS and financial outcomes.

CS is assessed in accordance with Anderson et al. (1994), who conceptualize it as a construct comprising three elements: satisfaction, expectation, and comparison with an ideal standard. The resulting satisfaction index ranges from 0 to 100, with 100 indicating the highest level of satisfaction. CR includes income from net interest on various loans and deposits, funds transfer pricing, and other fixed and variable income streams. We select CR as our preferred measure of financial performance over alternatives such as customer profitability, as the latter is complicated by the distribution of overhead costs across channels and central IT systems. This distribution makes it difficult to attribute costs accurately to individual customers (Ittner & Larcker, 1998; Yu, 2007; Aurier & N’Goala, 2010; Terpstra et al., 2012; Eriksson et al., 2020). Focusing on CR allows us to better understand the financial implications of individual CS.

The empirical analysis draws on unique data from 19,054 Swedish bank customers, with CS matched to individual revenue figures. The data span the period from 2013 to 2017. Although online banking usage has increased since then, relational attributes, such as satisfaction, remain observable within objectively defined transactional customer groups (Eriksson & Hermansson, 2017). This validates the relevance of satisfaction measures regardless of whether customers interact in person or online. The sizeable dataset enables the partitioning of the sample into various satisfaction levels, revenue levels (or volumes), and growth rates, thereby facilitating a detailed examination of the relationship’s linearity. Recognizing that prior research has yielded different results depending on whether the focus is on revenue level or growth (Ittner & Larcker, 1998), our analysis considers both levels of CR in SEK and their corresponding percentage changes. Additionally, key control variables, such as age, gender, monthly income, debt, and wealth, are incorporated. Furthermore, the initial level of CR is controlled for in the growth analysis, given that growth rates vary significantly with a customer’s initial level of engagement with the bank.

Our findings demonstrate a statistically significant positive relationship between CS and both the level and growth of CR, although the proportion of explained variance is exceptionally low (below 1%). The relationship between satisfaction and revenue level is not strictly linear; specifically, customers with satisfaction scores between 80 and 89 generate higher revenue one year after the satisfaction measurement than those with scores between 90 and 100. Conversely, no significant nonlinearity is observed in the relationship between satisfaction and revenue growth. Moreover, variables such as gender, age, income, and debt appear to explain the observed nonlinearity in the satisfaction–revenue level relationship, whereas wealth and the initial revenue level do not. The nonlinear pattern, initially identified in 2014, is not sustained in subsequent years (2015–2017). These findings suggest that while CS positively influences financial outcomes, its explanatory power is limited and its effects vary across customer segments. Consequently, there is only weak evidence of nonlinearity in the satisfaction-revenue level relationship—a finding that warrants further investigation and carries distinct implications for theory, practice, and policy. These initial insights suggest that managerial implications might involve tailoring satisfaction-driven marketing and relationship management strategies within retail banking, though such initiatives must be carefully designed and targeted to maximize their long-term impact.

The remainder of this article is structured as follows: it begins with a literature review and the development of hypotheses, followed by a presentation of methodology and data. The empirical results are then discussed, and the article concludes with a summary of the core findings and their broader implications.

2. Literature Review and Hypotheses Development

Studies examining the relationship between CS and financial performance are typically grounded in relationship marketing theory (Berry, 1983) and, more specifically, in service marketing theory (Parasuraman et al., 1994). Within this framework, attributes such as CS, loyalty, trust, contextual factors, and relationship duration are viewed as key drivers of customer sales, revenue, and profitability (Eriksson & Hermansson, 2014). Recent research also suggests a link between digital banking satisfaction—derived from ease of use, efficiency, and security—and profitability indicators, including retention (Egala et al., 2021).

Several seminal contributions have shaped the conceptualization and measurement of CS. Fornell (1992) and Anderson et al. (1994) developed widely adopted models focusing on evaluations of satisfaction, expectations, and comparison with ideal standards. Oliver (2010, p. 8) defines satisfaction as “a judgement that a product or service feature, or the product or service itself, provided a pleasurable level of consumption-related fulfillment, including levels of under- or over-fulfillment.” Subsequent studies (Terpstra et al., 2012; Terpstra et al., 2014; Eklof et al., 2017) confirm that the CS construct developed by Fornell (1992) is valid and reliable within the retail banking context.

The potential nonlinearity in the satisfaction–revenue relationship can be interpreted through several theoretical lenses. We integrate these perspectives to guide the interpretation of our empirical data and facilitate pattern identification. Anderson and Mittal (2000) note that the satisfaction–revenue relationship may exhibit increasing or decreasing returns, eventually reaching a saturation zone where further investment yields flat or diminishing marginal gains, a phenomenon closely aligned with prospect theory (Tversky & Kahneman, 1979). Similarly, Oliver’s (2010) disconfirmation theory posits that satisfaction is determined by the discrepancy between expectations and perceived performance, producing a function with a shape comparable to the value function in prospect theory. Empirical studies across banking and other sectors support the existence of nonlinear relationships between CS and financial performance, particularly regarding diminishing returns at high satisfaction levels (Ittner & Larcker, 1998; Mittal & Kamakura, 2001; Helgesen, 2006; Yu, 2007; Terpstra & Verbeeten, 2014; Strydom et al., 2020; Gao et al., 2021; How & Lee, 2021). For instance, Ittner and Larcker (1998) demonstrated in the telecommunications industry that revenue levels increased with higher satisfaction up to a score of 100, but with diminishing marginal gains, dropping from $74.80 for scores between 88 and 92 to $25.81 for scores between 96 and100. In contrast, revenue growth plateaued around a satisfaction score of 80. Revenue level was utilized to examine whether highly satisfied customers purchased more services than less satisfied customers, whereas revenue growth was used to examine whether customers at higher satisfaction levels increased purchases more than those at lower levels. Based on this reasoning, both the level and growth of CR merit analysis. This leads to the following hypotheses, which focus specifically on the highest level of the satisfaction tier (CS scores ranging from 90 to 100):

Hypothesis 1 (H1a).

The relationship between customer satisfaction and the level of customer revenue in banking exhibits diminishing returns at the highest satisfaction levels.

Hypothesis 1 (H1a).

The relationship between customer satisfaction and the growth in customer revenue in banking exhibits diminishing returns at the highest satisfaction levels.

Several studies have explored the temporal dynamics of the relationship between CS and financial performance. Yu (2007) found a positive effect of CS on CR in banking with a nine-month lag. Terpstra et al. (2012) identified positive effects on revenue growth after a one- to two-year period, while Terpstra and Verbeeten (2014) observed sustained nonlinear effects on customer value (defined as revenue minus costs) over a three-year horizon. In their study, the most profitable customers gained the most from higher satisfaction, illustrating enduring nonlinear benefits. Segerlind et al. (2026) demonstrated that higher CS is associated with a larger increase in CR than lower CS, with these effects persisting up to four years. Path dependence theory (see, e.g., North, 1994) provides a conceptual explanation for this temporal persistence; specifically, cognitive and financial lock-in mechanisms among satisfied customers prevent immediate switching in the absence of disruptive events, such as financial crises, that could prompt behavioral reframing. Consequently, we hypothesize that the nonlinear effects observed one year after the initial satisfaction measurement are sustained over time, extending up to four years following the baseline measurement.

Hypothesis 2 (H2).

The nonlinear effects observed one year after the satisfaction measurement persist for two, three, and four years following the baseline period.

Mental accounting and budget constraint theories suggest a ceiling effect, whereby diminishing returns to satisfaction emerge as customers hit budgetary or cognitive limits (Thaler & Shefrin, 1981; Dong et al., 2011). Accordingly, we explore how demographic and socioeconomic factors—specifically wealth, debt, income, age, and gender—influence this nonlinearity. While Eisenbeiss et al. (2014) suggest such effects are less pronounced for highly involved customers, such as those with substantial debt or wealth, evidence regarding age and gender remains mixed. For instance, although older customers may exhibit higher loyalty in the banking sector (Baumann et al., 2005), findings on age-related satisfaction are inconsistent (Mittal & Kamakura, 2001). Additionally, women often report lower involvement in banking and fintech services (Demirgüç-Kunt & Singer, 2017; Chen et al., 2023). We incorporate these variables, along with the baseline level of revenue, to examine their impact on the nonlinear effects of satisfaction.

3. Research Design

3.1. Data

Data were collected from one of Sweden’s largest retail banks (hereafter, “the Bank”), which maintains a market share (approximately 20%) and customer profiles comparable to those of the other three major banks in the country. Two types of data were utilized in this study: (1) customer telephone survey data (“survey data”), collected in October 2013, and (2) data from the Bank’s register of individual retail customers (“register data”) who participated in the survey, spanning the period from October 2013 to October 2017.

To be included in the final sample, customers had to meet specific inclusion criteria. First, they had to be between 18 and 78 years old at the time of the survey in the spring of 2013. Second, they had to identify the Bank as their primary financial institution. Additionally, eligible customers were required to have either conducted at least three transactions within the preceding seven months or maintained a minimum of 10,000 Swedish kronor (SEK) in total savings with the Bank. Respondents who did not regard the focal bank as their primary institution were excluded from the sample (n = 3020), given that secondary banking relationships exhibit fundamentally different engagement dynamics.

The telephone survey was conducted using stratified random sampling across 124 business units, with 200 interviews conducted within each unit. After removing outliers and ensuring complete data across all variables and all years, the final balanced sample consists of 19,054 respondents. Customers with negative revenue (n = 89) in 2013 were excluded from the study, along with customers (n = 10) exhibiting the most extreme positive CR values in the same year. Under normal operational circumstances, there are concerns that they may reflect data quality issues, such as recording errors or misclassifications, rather than genuine economic outcomes.

Regarding representativeness, the sample aligns closely with the broader Swedish population when comparing the survey data, the register data, and national demographic statistics. The mean age in the final sample is 50.6 years (with a standard deviation of 17.6 years), which is comparable to the average age of 49.7 years in the Bank’s overall register data and 41.2 years in the Swedish population in 2013. Notably, the national average age includes individuals under the age of 18, whereas this sample is restricted to adults. Because women are taxed individually in Sweden, it is standard practice within the Swedish context to analyze financial outcomes at the individual level rather than the household level. In our analysis, gender is treated as a binary indicator variable, with men coded as 1 and women as 0. The reported values for investment volume (wealth), loan volume (debt), and monthly net income align with those of the average Swedish population at the time, accounting for the sample’s age distribution. Consequently, the empirical models control for age, gender, wealth, debt, and income. Descriptive statistics for all included variables are provided in

Table 1. Descriptive summary statistics of the sample population (2014).

Variables	Mean	SD	Min	Max
%∆CR2014	11.49	110.03	−1366.67	6150
CR 2014	5804.32	785.48	0	150,000
CSI	66.16	20.09	0	100
CSI2	4780.79	2466.16	0	10,000
CSI_groups
0–49	0.19	0.40	0	1
50–59	0.15	0.36	0	1
60–69	0.16	0.36	0	1
70–79	0.26	0.44	0	1
80–89	0.13	0.34	0	1
90–100	0.10	0.30	0	1
Gender	0.50	0.50	0	1
Age	50.60	17.57	19.00	79.00
Log_Debt	4.75	6.15	0	16.12
Log_Wealth	11.31	2.43	0	18.32
Log_Monthly Net Income (MNI)	6.07	4.55	0	12.64

Note: N = 19,054, CR is customer revenue, CSI is the customer satisfaction index, and %∆ is the percentage change in CR from the previous period. All coefficients rounded to two decimal places.

.

3.2. Measurement of Customer Revenue and Customer Satisfaction

We analyze customer-level performance based on the Bank’s revenue. Total revenue includes income from net interest on various loans and deposits, funds transfer pricing, and other fixed and variable income streams. While our dataset encompasses both revenue and cost data at the individual customer level, individual profitability is challenging to assess accurately. This difficulty arises because the Bank allocates both variable costs (e.g., transaction costs, credit risk costs for own funds, and operational risk) and fixed costs (e.g., financial, physical, and marketing expenses) evenly across all retail customers. This structural issue aligns with the data limitations documented by Campbell and Frei (2010) and Eriksson et al. (2020). Consequently, consistent with Eriksson et al. (2020), we argue that evaluating customer-level performance is more reliable when focusing on revenue, as it directly reflects actual customer behavior.

Our primary focus is the level and growth of CR, drawing on Terpstra et al. (2012), who argue that future revenue growth is a suitable performance metric in retail banking. This choice is also informed by Ittner and Larcker (1998), who observed that empirical results differ significantly depending on whether the level or the growth in revenue is measured. For the empirical analysis, CR data from 2013, 2014, 2015, 2016, and 2017 are utilized.

The Bank regularly measures its customers’ CS by conducting structured surveys. CS is assessed according to the definition by Anderson et al. (1994), who conceptualize it as a construct comprising three core elements: satisfaction, expectation, and comparison with an ideal standard (see American Society for Quality Control, 1998). The customer satisfaction index (CSI) follows the foundational model developed by Fornell (1992), wherein these three latent variables are measured via a survey using a 10-point Likert-type scale. The survey responses are equally weighted, share the same 1-to-10 range, and are ultimately rescaled to a 0-to-100 index using the following formula:

$$SC score=\left(mean of the three CS questions-1\right) \times {\displaystyle \frac{100}{9}}$$

The three questions were phrased as follows:

• Considering all your experiences with the Bank, how would you rate it on a scale of 1 to 10, where 1 = very dissatisfied and 10 = very satisfied?
• To what extent does the Bank meet your expectations? Use a scale from 1 to 10, where 1 = does not match at all and 10 = fully meets expectations.
• Imagine a bank that is perfect in all respects. How close or far from that ideal is the Bank? Use a scale from 1 = very far to 10 = as close as possible.

For this study, we examine both the continuous and the squared CSI values, while also categorizing the index into six distinct satisfaction groups. The lowest group (CSI 0–49) comprises customers whose average response fell below 5 on the original 1–10 scale. Customers within the subsequent tiers, CSI-levels 50–59, 60–69, 70–79, 80–89, and 90–100, are evaluated relative to this baseline reference group (0–49). The categorization scheme, featuring a single reference group of less satisfied customers alongside five increasingly satisfied groups in increments of 10, was designed to facilitate a granular analysis of potential threshold effects.

As presented in

Table 2. Mean characteristics of the sample population across CS levels in 2014.

	CSI-Levels
Variables	0–49	50–59	60–69	70–79	80–89	90–100
%∆CR2014	3.38	9.95	11.05	12.97	20.39	14.54
CR2014	5640	6367	6443	6019	5434	4219
Age	55.65	51.65	49.73	48.90	47.92	48.36
Gender	0.52	0.54	0.52	0.52	0.46	0.41
Wealth	406,512	435,074	408,979	381,882	350,604	237,808
Debt	220,013	280,875	303,951	273,089	220,712	167,790
Monthly Net Income	11,131	12,643	12,996	11,812	10,304	8761

Note: N = 19,054. All values represent the mean within each CSI category. %∆ indicates percentage change. Raw financial figures (wealth, debt, income) are reported in currency units, unlike the log transformations in Table 1.

, the change in CR increases progressively across the customer satisfaction groups up to the 80–89 CS bracket, after which it decreases for the highest group (90–100). Conversely, the absolute CR level peaks much earlier, specifically within the 60–69 CS range. A comparison of demographic attributes reveals that customers in the 0–49 baseline reference group are generally older than those in the highest satisfaction groups. Furthermore, female customers are more heavily represented in the two uppermost CS groups, particularly within the 90–100 range. Socioeconomic metrics display a downward trend at higher satisfaction levels: average wealth peaks in the 50–59 CS tier and steadily diminishes thereafter, reaching its absolute minimum in the 90–100 group. Similarly, both outstanding debt and monthly net income reach their maximums within the 60–69 CS range before tapering off to their lowest observed values in the 90–100 tier. In summary, the 90–100 CS group is uniquely characterized by the lowest relative levels of wealth, debt, and income alongside a higher concentration of female and younger customers relative to the least satisfied groups. This descriptive overview highlights that while the most satisfied customers exhibit the second-highest growth in CR, the peak growth in CR is localized within the 80–89 CS segment. Finally, a correlation matrix evaluating these variables is detailed in

Table 3. Correlation matrix of key variables in 2014.

Variables	CR2014	CR∆2014	CSI_Groups	CSI	CSI2	Gender	Age	Debt	Wealth	Income
CR2014	1.000
%∆CR2014	0.218 *	1.000
CSI_groups	−0.044 *	0.040 *	1.000
CSI	−0.029 *	0.041 *	0.940 *	1.000
CSI2	−0.045 *	0.039 *	0.972 *	0.972 *	1.000
Gender	−0.160 *	−0.008	0.059 *	0.053 *	0.063 *	1.000
Age	0.155 *	−0.104 *	−0.143 *	−0.150 *	−0.137 *	0.066 *	1.000
Log_debt	0.443 *	0.075 *	−0.039 *	−0.028 *	−0.040 *	−0.237 *	0.092 *	1.000
Log_wealth	0.378 *	0.008	−0.100 *	−0.090 *	−0.103 *	−0.004	0.296 *	0.115 *	1.000
Log_MNI	0.094 *	0.036 *	−0.008	0.005	−0.010	−0.038 *	−0.300 *	0.211 *	0.130 *	1.000

Note: N = 19,054. All coefficients rounded to three decimal places. MNI is monthly net income. * p < 0.01.

.

4. Empirical Analysis

4.1. Empirical Model

First, we test for nonlinearity by employing an ordinary least squares (OLS) regression to explain individual-level customer revenue based on CS and the square of CS. The following equation is estimated:

$$C{R}_i={\beta }_o+{\beta }_{1}C{S}_i+{\beta }_{2}C{S}_i^2+{\beta }_3 X_i+{\epsilon }_i$$

(1)

where CR_i represents the level (or volume in SEK) of CR in year i, CS is a continuous variable measuring CS (in line with CSI as explained above), and X is a set of control factors associated with the level of customer revenue for individual i.

The same equation is also used to test the growth in CR from CS.

$$\%\Delta C{R}_i={\beta }_o+{\beta }_{1}C{S}_i+{\beta }_{2}C{S}_i^2+{\beta }_3 X_i+{\epsilon }_i$$

(2)

where $\%\Delta C{R}_i$ is the percentage change in CR in year i compared to the previous year. Here is an example of the percentage change in 2014:

$$\%∆C{R}_i={\displaystyle \frac{C{R}_{i,2014} -C{R}_{i,2013}}{{CR}_{i,2013}}}\times 100$$

(3)

Second, we use an OLS approach again to explain individual-level customer revenue growth from CS, divided into six groups and examining a period of up to four years. The estimated equation is:

$$C{R}_{\left\{i, t \to t+n\right\}}= {\beta }_0+ {\beta }_{1}C{R}_{\left\{it\right\}}+ {\sum }_{k=2}^6{\beta }_k{D}_{\left\{ik\right\}}+ {\sum }_{j=1}^m{\gamma }_j{Z}_{\left\{ji\right\}}+ {\epsilon }_i$$

(4)

$$\%\Delta C{R}_{\left\{i, t \to t+n\right\}}= {\beta }_0+ {\beta }_{1}C{R}_{\left\{it\right\}}+ {\sum }_{k=2}^6{\beta }_k{D}_{\left\{ik\right\}}+ {\sum }_{j=1}^m{\gamma }_j{Z}_{\left\{ji\right\}}+ {\epsilon }_i$$

(5)

where $\%\Delta C{R}_{\left\{i, t \to t+n\right\}}$ denotes the percentage growth in CR for each individual i with t = 2013, and n = 1…4 represents the time periods between 2013 and 2017. $C{R}_{\left\{it\right\}}$ denotes the base level of CR in 2013, included to control for the initial CR level. D represents dummy variables for CS groups, as CS is divided into k groups (k = 1…6). $Z_{\left\{ji\right\}}$ is a set of control variables for individual i, and Ɛ_i is an error term. The same equation is also used to test the level (or volume in SEK) of CR from CS (Equation (4)).

We do not assert causality; instead, we focus on analyzing the statistical associations between CS and both the level and growth of CR. The relationship is conceptualized as an iterative process, whereby higher satisfaction may lead to increased revenue, which in turn could further enhance satisfaction, and so forth. Establishing strict causality, however, would require a longitudinal research design capable of fully capturing contemporaneous change in both CS and CR over time. Given the characteristics of our dataset, we aim instead to deepen our understanding by analyzing how revenue levels and growth over time are associated with distinct customer satisfaction tiers. Finally, because CR growth encompasses both positive and negative values, revenue is not modeled in logarithmic form.

4.2. The Effect of Customer Satisfaction on Customer Revenue

All regression models reported below are statistically significant at the 0.1% level (p < 0.001). The corresponding F-statistics and degrees of freedom for each model are available upon request. Although diagnostic tests indicate skewness in the residuals, we rely on ordinary least squares (OLS) estimation for two main reasons. First, given our large sample size (n = 19,054), the Central Limit Theorem guarantees that the parameter estimators are asymptotically normally distributed, thereby rendering standard hypothesis testing valid. Second, to address potential heteroscedasticity and arbitrary residual distributions, we utilize Huber–White robust standard errors via the vce (robust) option in Stata 17. This procedure ensures that our standard errors, (t)-statistics, and (p)-values are robust to violations of both homoscedasticity and normality.

The empirical results from Equation (1) and its variations are presented in

Table 4. OLS regression analysis: The association between CS, as well as CS², and the level and growth of customer revenue in 2014.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Variables	CR2014	CR2014	CR2014	CR2014	CR2014	%∆CR2014	%∆CR2014	%∆CR2014	%∆CR2014	%∆CR2014
CS	99.4626 ***	7.0742 ***	46.6794 ***	5.5446 ***	4.2274	0.2983 **	0.1467 ***	−0.0161	0.1499 ***	0.0721
	(10.0810)	(2.2626)	(8.3589)	(1.2947)	(4.7771)	(0.1324)	(0.0339)	(0.1369)	(0.0337)	(0.1375)
CSI2	−0.9298 ***		−0.3322 ***		0.0110	−0.0006		0.0014		0.0007
	(0.0838)		(0.0697)		(0.0403)	(0.0012)		(0.0012)		(0.0012)
Gender		−1084.8417 ***	−1071.8293 ***	47.3277	46.9425		4.4022 **	4.3487**	2.0480	2.0253
		(101.1642)	(100.9122)	(53.7579)	(53.6720)		(1.8435)	(1.8541)	(1.7693)	(1.7789)
Age		8.4126 ***	9.0054 ***	−11.5272 ***	−11.5477 ***		−0.8159 ***	−0.8183 ***	−0.7745 ***	−0.7757 ***
		(2.5442)	(2.5499)	(1.4659)	(1.4728)		(0.0564)	(0.0570)	(0.0557)	(0.0563)
log_debt		502.5887 ***	501.3462 ***	53.7577 ***	53.7802 ***		1.6694 ***	1.6745 ***	2.6027 ***	2.6040 ***
		(10.1727)	(10.1960)	(10.4055)	(10.3979)		(0.1900)	(0.1891)	(0.2464)	(0.2458)
log_wealth		1075.4669 ***	1069.6633 ***	185.9361 ***	186.0917 ***		1.8742 ***	1.8981 ***	3.7238 ***	3.7330 ***
		(33.9196)	(33.8830)	(17.5306)	(17.5411)		(0.4482)	(0.4525)	(0.4673)	(0.4695)
log_net monthly income		−50.3992 ***	−51.5488 ***	−16.6090 **	−16.5694 **		−0.6547 ***	−0.6500 ***	−0.7249 ***	−0.7226 ***
		(12.2143)	(12.2190)	(6.9565)	(6.9427)		(0.2048)	(0.2055)	(0.2060)	(0.2068)
CR2013_base				1.0007 ***	1.0008 ***				−0.0021 ***	−0.0021 ***
				(0.0195)	(0.0195)				(0.0002)	(0.0002)
Constant	3669.1813 ***	−7719.8004 ***	−8722.8935 ***	−1899.1661 ***	−1865.5635 ***	−5.2910	11.3211 *	15.4439 **	−0.7819	1.2030
	(292.7101)	(405.4110)	(435.9461)	(206.0548)	(229.7518)	(3.4930)	(6.0402)	(6.8700)	(6.1518)	(7.0980)
Observations	19,054	19,054	19,054	19,054	19,054	19,054	19,054	19,054	19,054	19,054
R-squared	0.0056	0.3112	0.3118	0.8021	0.8021	0.0017	0.0208	0.0209	0.0316	0.0317

Robust standard errors in parentheses. All coefficients rounded to four decimal places. CR2013_base is the customer revenue measured in 2013 as a baseline reference. *** p < 0.01, ** p < 0.05, * p < 0.1.

, with Columns 1–5 showing regressions for the absolute level of CR (CR2014) and columns 6–10 reporting the growth of CR (%ΔCR2014).

In Column 1 of Table 4, a linear model without control variables shows that a one-unit increase in CS (measured on a 0–100 scale) is associated with a 99.5 SEK higher level of CR, which is statistically significant at the 1% level. Although not explicitly tabulated in a purely linear model, the introduction of the squared term of CS yields a negative and significant coefficient, indicating a concave nonlinear relationship. In Column 2, demographic and financial control variables are included, revealing that CR is negatively associated with gender (male) and income, whereas age, debt, and wealth are positively associated with CR. In Column 3, the coefficients for both CS and CS-squared remain statistically significant at the 1% level; however, the linear coefficient of CS decreases to 46.7 SEK while still confirming a nonlinear pattern. When the baseline level of CR2013 is included in Column 4, CS remains significantly and positively related to CR2014. Conversely, in Column 5, when both CS and CS-squared are simultaneously included alongside the baseline revenue control, neither coefficient remains statistically significant. Overall, the R² increases to 80.2% when all control variables are incorporated, but less than 0.6% of the variance in CR is explained by CS alone.

In Column 6 of Table 4, the relationship between CS and the CR growth is examined. The linear model without control variables indicates that a one-unit increase in CS raises the CR growth by 0.2983 percentage points, which is statistically significant at the 5% level. The quadratic term of CS is not statistically significant, suggesting no evidence of a nonlinear relationship. When control variables are included in Column 7, CR growth is found to be positively associated with gender (male), debt, and wealth and negatively associated with age and income. In Column 8, when both CS and CS-squared are included alongside these controls, neither coefficient achieves statistical significance. Controlling for the baseline level of CR2013 in Column 9 yields a significant linear effect of CS; however, in Column 10, where all variables are simultaneously included, neither CS nor CS-squared remains significant. Across all growth specifications, the explanatory power remains remarkably low: the R² values hover around 3%, with less than 0.2% of the variance in CR growth explained by CS alone. Consequently, there is no evidence of nonlinearity in the relationship between CS and CR growth, regardless of the model specification.

To further investigate the relationship between CS and both the level and growth of CR across different satisfaction levels, CS is divided into six distinct groups. Each of the five groups scoring 50 or above is compared with the baseline reference group, which comprises scores between 0 and 49. The results from Equation 2 and its variations for the level and growth in CR in 2014 are presented in

Table 5. OLS regression analysis: The association between CS groups and the level and growth of customer revenue in 2014.

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	CR2014	CR2014	CR2014	%∆CR2014	%∆CR2014	%∆CR2014
Relative 0–49
CSI 50–59	727.1632 ***	588.6532 ***	136.2975 *	6.5695 ***	3.0035	3.9437 *
	(178.6490)	(148.9627)	(79.0410)	(2.2438)	(2.2040)	(2.1954)
CSI 60–69	802.8919 ***	767.8337 ***	208.1307 **	7.6719 ***	2.5871	3.7504
	(193.3888)	(165.5162)	(82.8785)	(2.8395)	(2.9368)	(2.9622)
CSI 70–79	378.8050**	561.7800 ***	267.5055 ***	9.5901 ***	4.2529 **	4.8646 ***
	(160.7929)	(138.3398)	(71.2676)	(1.8435)	(1.8340)	(1.8230)
CSI 80–89	−206.1640	678.1137 ***	433.4842 ***	17.0127 ***	12.0446 ***	12.5531 ***
	(209.8215)	(183.9639)	(113.3081)	(3.1981)	(3.1559)	(3.1483)
CSI 90–100	−1421.0847 ***	124.2092	225.0214 ***	11.1632 ***	7.6581 ***	7.4486 ***
	(185.9440)	(161.1441)	(85.2674)	(2.8273)	(2.7647)	(2.7479)
CR2013_base			1.0006 ***			−0.0021 ***
			(0.0195)			(0.0002)
Controls	No	Yes	Yes	No	Yes	Yes
Constant	5640.0378 ***	−7693.9201 ***	−1739.3429 ***	3.3797 ***	16.5020 ***	4.1255
	(112.5309)	(362.0978)	(184.1449)	(1.2959)	(5.6770)	(5.8184)
Observations	19,054	19,054	19,054	19,054	19,054	19,054
R-squared	0.0065	0.3121	0.8021	0.0021	0.0212	0.0320

Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. Columns 1–3 report regressions where the dependent variable is the level of CR, whereas Columns 4–6 report regressions where the dependent variable is the growth of CR.

In Column 1, the relationship between discrete CS tiers and the level of CR (estimated without control variables) is examined. The results show that higher CS levels are associated with increased CR up to the 60–69 range relative to the baseline reference group (scores below 50). Beyond this threshold, the association weakens; coefficients for the 70–79 tier remain significant but decline in magnitude, the 80–89 range is no longer statistically significant, and the highest tier (90–100) is negatively and significantly associated with CR. When control variables (excluding the baseline level of CR) are incorporated in Column 2, a similar pattern emerges, although the effect sizes are smaller; notably, scores in the 80–89 bracket are significantly higher than those of the baseline group, whereas the coefficients for the 90–100 tier become statistically non-significant. Finally, in Column 3, after incorporating the baseline level of CR2013, the results demonstrate significantly higher revenue for all satisfaction groups up to the 80–89 tier, followed by a significant decline for the 90–100 group.

In Column 4, the relationship between the discrete CS tiers and CR growth is analyzed without control variables. The resulting pattern mirrors that observed in Column 3: higher CS levels are associated with elevated CR growth up to the 80–89 group, after which the relationship weakens. The strongest association is localized within the 80–89 CS bracket, whereas the 90–100 group exhibits a smaller increase, though it remains larger than that of the 70–79 group. Column 5 reveals the same pattern when control variables are added (excluding the baseline level of CR), and this trend remains consistent in Column 6 when all controls, including the baseline level of CR2013, are incorporated.

Examining the statistical associations in Table 5, customer revenue increases less in the highest satisfaction tier (90–100) compared to the baseline reference group than it does in the second-highest tier (80–89). Specifically, customers with CS scores between 80 and 89 exhibit a CR increase of 12.55 times (bootstrap standard error = 3.15) compared to the baseline group (0–49), while customers in the 90–100 range exhibit an increase of 7.45 times (bootstrap standard error = 2.75).

As shown in

Table 6. Contrasts compared between CS groups and the level and growth of customer revenue in 2014.

	Levels				Growth
CSI Groups	Contrast	Std.Err	[95% Conf. Interval]		Contrast	Std.Err	[95% Conf. Interval]
50–59 vs. 0–49	136.2975	79.0410	−18.6298	291.2249	3.9437	2.1954	−0.3594	8.2467
60–69 vs. 0–49	208.1307	82.8785	45.6815	370.5798	3.7504	2.9622	−2.0558	9.5566
70–79 vs. 0–49	267.5055	71.2676	127.8146	407.1964	4.8645	1.8230	1.2914	8.4378
80–89 vs. 0–49	433.4842	113.3081	211.3902	655.5781	12.5531	3.1483	6.3821	18.7240
90–100 vs. 0–49	225.0214	85.2674	57.8898	392.1530	7.4486	2.7479	2.0625	12.8346
60–69 vs. 50–59	71.83315	87.9854	−100.6261	244.2924	−0.1933	3.1162	−6.3012	5.9147
70–79 vs. 50–59	131.208	77.1904	−20.0921	282.508	0.9209	2.2240	−3.4383	5.2801
80–89 vs. 50–59	297.1866	117.0604	67.7380	526.6353	8.6094	3.4061	1.9332	15.2856
90–100 vs. 50–59	88.72392	91.2689	−90.1712	267.6191	3.5049	3.0848	−2.5416	9.5514
70–79 vs. 60–69	59.37481	79.7867	−97.0142	215.7638	1.1142	2.8333	−4.4393	6.6677
80–89 vs. 60–69	225.3535	117.6595	−5.26950	455.9765	8.8027	3.8450	1.2660	16.3393
90–100 vs. 60–69	16.8908	91.4454	−162.3504	196.1319	3.6981	3.6303	−3.4175	10.8138
80–89 vs. 70–79	165.9787	108.9167	−47.5078	379.4652	7.6885	3.1595	1.4957	13.8813
90–100 vs. 70–79	−42.4840	80.4868	−200.2453	115.2772	2.5840	2.8032	−2.9104	8.0784
90–100 vs. 80–89	−208.4627	114.6531	−433.1930	16.2676	−5.1045	3.7974	−12.5478	2.3388

Note: Bold font implies significance at the 5% level; italics implies a 10% significance level.

, pairwise comparisons between CS groups are presented for the model specification that includes all control variables, including the baseline level of CR. Regarding the absolute level of CR, only the 80–89 group exhibits a significantly higher value (at the 5% significance level) than the 50–59 group. At the 10% significance level, the 90–100 group displays a significantly lower value than the 80–89 group. Additionally, the 70–79 group is significantly higher than the 50–59 group, and the 80–89 group is higher than the 60–69 group. For CR growth, the 80–89 group demonstrates significantly higher values compared to all tiers below 80 (i.e., 0–49, 50–59, 60–69, and 70–79). However, the 90–100 group is not significantly lower than the 80–89 group.

To summarize the findings presented in Table 4, Table 5 and Table 6, significant nonlinear effects are observed only for the level of CR and not for its growth. This nonlinearity is evident when analyzing CS and CS-squared in models both excluding and including control variables, except when the baseline level of CR2013 is incorporated in the quadratic model. Similarly, the analysis using six distinct satisfaction tiers and all control variables indicates a nonlinear pattern for the level of CR. However, when comparing these CS groups directly (Table 6), this nonlinearity is only statistically significant at the 10% level, suggesting that the highest CS group (90–100) exhibits a lower CR level than the second-highest group (80–89). Conversely, no significant nonlinearity is found for CR growth, as the highest CS group does not differ significantly from the second-highest group. Finally, the overall explanatory power of CS across all models remains low (R² < 1%). Taken together, these findings provide weak support for H1a but do not support H1b.

4.3. Sustained Effects up to Four Years

As shown in

Table 7. Summary of findings regarding the significance of nonlinearity and factors explaining nonlinearity, 2014–2017.

Type of CR	Type of CS	Controls/Factors	2014	2015	2016	2017
CR Level	CS and CSI2	No controls	1%	1%	1%	1%
		Controls (1)	1%	1%	1%	1%
		Controls (2)	No	No	No	No
CR Growth	CS and CSI2	No controls	No	No	No	No
		Controls (1)	No	No	No	No
		Controls (2)	No	No	No	No
CR Level	CS levels in	No controls	1%	1%	1%	1%
	groups	Controls (1)	No	No	No	No
		Controls (2)	1%	5%	No	No
CR Growth	CS levels in	No controls	1%	1%	1%	1%
	groups	Controls (1)	1%	No	5%	No
		Controls (2)	1%	No	No	No
CR Level	Contrast	Controls (2)	10%	No	No	No
	90–100/80–89
CR Growth	Contrast	Controls (2)	No	No	No	No
	90–100/80–89
CR Level	CS levels	Gender	5%	10%	5%	5%
	90–100/80–89	Age	10%	No	No	No
		Debt	5%	No	No	No
		Wealth	No	No	No	No
		Income	5%	No	No	5%
		CR Base level	No	No	No	10%
CR Growth	CS levels	Gender	10%	No	No	No
	90–100/80–89	Age	10%	No	No	No
		Debt	No	No	No	No
		Wealth	No	No	No	No
		Income	10%	No	No	No
		CR Base level	5%	No	5%	5%

Note: Controls (1) include gender, age, debt, wealth, and income. Controls (2) include gender, age, debt, wealth, income, and the baseline reference of CR.

regarding Hypothesis 2, the support for H1a, which establishes a nonlinear relationship between CS (measured in 2013) and the absolute level of CR (in 2014), remains consistent across subsequent years (2015–2017) in models both with and without control variables, except when the baseline CR level is included. When the baseline CR level (2013) is incorporated, the nonlinear relationship disappears, a result that holds uniformly from 2014 through 2017. Similarly, the lack of empirical support for H1b, proposing a nonlinear relationship between CS2013 and CR growth in 2014, is sustained when analyzing CR growth through 2015, 2016, and 2017.

When dividing CS into six groups, the only indication of nonlinearity in 2014 is that the CR level for the CS 90–100 CS tier is lower than that of the 80–89 tier, which is significant only at the 10% level. Consequently, these results provide no empirical support for H2, which posits that nonlinear effects are sustained over time.

Regarding the underlying drivers, gender emerges as the most consistent factor explaining variations among the highest satisfaction groups within the CS-CR level relationship, with its moderating effects persisting throughout the four-year observation window. In contrast, the effects of age and income, while significant in 2014, are not sustained beyond the first year. For CR growth, only the baseline CR level (2013) consistently explains differences at the highest satisfaction levels over time.

4.4. Factors Contributing to the Explanation of Nonlinearity

The descriptive summary across CS levels in Table 2 shows that customers in the highest CS groups are more likely to be women, younger, less wealthy, less indebted, and earn lower incomes than those in the baseline reference group. These demographic and socioeconomic differences persist relative to the second-highest CS group, with the sole exception of a minor increase in average age. To better understand the unique consumer characteristics associated with the nonlinear patterns observed in Section 4.2, these variables interact with the discrete CS levels by extending Equation (2).

Table 8. Interactions between CS groups and demographic and financial factors on the level of customer revenue (2014).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CR2014	CR2014	CR2014	CR2014	CR2014	CR2014	CR2014
	No Interactions	Gender	Age	Debt	Wealth	Income	CR 2013 Base
Relative 0–49
2.CSI 50–59	136.2975 *	168.8123	237.4976 *	35.7160	100.3050	116.2422	85.9819
	(79.0410)	(126.7589)	(129.2944)	(90.7277)	(84.4777)	(108.0833)	(73.6099)
3.CSI 60–69	208.1307 **	209.5135	253.8201 **	246.0990 **	114.7105	354.2869 ***	35.5876
	(82.8785)	(132.2920)	(122.9361)	(109.1848)	(73.4507)	(130.2234)	(67.3026)
4.CSI 70–79	267.5055 ***	301.8357 ***	320.7699 ***	242.3207 ***	235.4468 ***	440.2613 ***	17.3251
	(71.2676)	(113.7658)	(110.5187)	(85.4248)	(70.1392)	(105.6153)	(61.3947)
5.CSI 80–89	433.4842 ***	412.6723 **	401.0233 ***	490.7569 ***	276.8188 ***	576.3324 ***	122.7194
	(113.3081)	(173.8983)	(128.4579)	(144.7463)	(86.6710)	(181.9150)	(100.2391)
6.CSI 90–100	225.0214 ***	433.5829 ***	411.9379 ***	222.3412 **	204.8343 ***	117.4105	195.1176 ***
	(85.2674)	(152.5274)	(119.0122)	(97.7849)	(79.1275)	(102.9342)	(73.5647)
2.CSI 50–59#D		−67.7432	−163.5137	261.8144	70.9599	23.9088	800.6464 ***
		(151.9313)	(162.0938)	(167.9040)	(149.6194)	(155.5400)	(259.3542)
3.CSI 60–69#D		−2.0014	−33.1148	−71.0845	173.5236	−284.7108*	1287.6252 ***
		(160.3460)	(170.4004)	(171.7813)	(157.3597)	(167.4985)	(292.9282)
4.CSI 70–79#D		−71.2094	−38.4481	83.0769	63.8374	−343.3527**	1111.9804 ***
		(135.8057)	(144.2643)	(149.4584)	(134.7516)	(139.1016)	(245.9452)
5.CSI 80–89#D		31.7484	171.8509	−157.5962	339.1008	−302.6133	1457.6611 ***
		(218.7436)	(230.5262)	(232.9929)	(237.9143)	(214.4959)	(379.1242)
6.CSI 90–100#D		−363.0248**	−308.5906 *	11.0339	−60.5331	246.1560	−190.1848
		(172.6482)	(170.3617)	(188.9369)	(185.8622)	(167.1329)	(340.3902)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	−1739.3429 ***	−1720.7291 ***	−2099.1540 ***	−1670.8941 ***	−136.6060	−1844.0754 ***	−4248.7792 ***
	(184.1449)	(193.0425)	(190.2123)	(194.2473)	(124.4028)	(196.2849)	(299.7997)
Observations	19,054	19,054	19,054	19,054	19,054	19,054	19,054
R-squared	0.8021	0.8022	0.8020	0.8017	0.8005	0.8023	0.3732

Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

and

Table 9. Interactions between CS groups and demographic and financial factors on the growth of customer revenue in 2014.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CR%∆2014	CR%∆2014	CR%∆2014	CR%∆2014	CR%∆2014	CR%∆2014	CR%∆2014
	No Interactions	Gender	Age	Debt	Wealth	Income	CR 2013 Base
Relative 0–49
2.CSI 50–59	3.9437 *	1.8353	8.4794 **	0.2582	6.1109	2.2365	8.2515 *
	(2.1954)	(2.4816)	(3.7474)	(2.3804)	(3.8254)	(2.6619)	(4.5313)
3.CSI 60–69	3.7504	1.2139	7.7938 **	1.9764	5.8626	7.1336	7.9727
	(2.9622)	(2.7396)	(3.2981)	(2.5990)	(5.4694)	(5.5043)	(6.1457)
4.CSI 70–79	4.8646 ***	6.7920 ***	11.2839 ***	1.8792	6.7465 **	3.6130	8.3748 **
	(1.8230)	(2.4801)	(3.1022)	(2.1724)	(3.0070)	(2.2840)	(3.5411)
5.CSI 80–89	12.5531 ***	11.3305 ***	22.2821 ***	10.1883 ***	12.9262 ***	10.8315 **	16.8693 ***
	(3.1483)	(4.3090)	(4.7090)	(3.8802)	(4.3199)	(4.5248)	(5.2832)
6.CSI 90–100	7.4486 ***	10.9871 ***	20.3611 ***	6.1582 **	9.3197 **	2.1993	13.1767 ***
	(2.7479)	(4.2371)	(4.8891)	(3.0593)	(4.0033)	(2.7055)	(4.6127)
2.CSI 50–59#D		4.5849	−5.2671	9.8075 **	−3.6875	3.4581	−8.6508 *
		(4.5026)	(4.5244)	(4.8288)	(4.5712)	(4.3951)	(4.7652)
3.CSI 60–69#D		5.3132	−2.5823	5.2864	−3.8105	−5.8683	−9.3474
		(5.7865)	(5.8809)	(6.2890)	(5.9306)	(6.0910)	(6.3768)
4.CSI 70–79#D		−4.0636	−6.8353 *	8.2422 **	−3.1831	2.6150	−8.2094 **
		(3.6235)	(3.6926)	(3.9037)	(3.6855)	(3.6674)	(3.8129)
5.CSI 80–89#D		2.2590	−12.4398 *	6.4824	−0.0621	3.6544	−11.9983 **
		(6.2808)	(6.4771)	(7.1502)	(6.3706)	(6.3665)	(6.0594)
6.CSI 90–100#D		−5.9457	−19.7500 ***	3.0457	−5.2016	11.6638 **	−13.6363 ***
		(5.6901)	(5.5235)	(6.6995)	(5.1843)	(5.8919)	(4.9788)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	4.1255	6.2722	−23.0620 ***	8.3437	33.4892 ***	3.4345	−11.6586 *
	(5.8184)	(5.4049)	(5.4266)	(5.8749)	(5.2509)	(6.2864)	(6.6629)
Observations	19,054	19,054	19,054	19,054	19,054	19,054	19,054
R-squared	0.0320	0.0323	0.0265	0.0287	0.0283	0.0321	0.0401

Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

present the interaction effects for the level and growth of CR, respectively, addressing the exploratory research questions discussed in the literature review. In Table 8, when gender is interacted with CS, the absolute CR level decreases for the highest CS scores (90–100). This suggests that women with the highest CS scores generate lower CR relative to men or those in the baseline reference group. A similar pattern emerges for age, where older customers at the highest CS exhibit lower CR values than their younger counterparts or those with lower CS scores. No significant interaction effects are observed for debt or wealth, whereas higher income reduces revenue across the 60–69 and 70–79 CS ranges relative to the baseline group. Finally, the baseline CR level (2013) exerts a strong positive influence across the 50–89 CS groups, peaking in the 80–89 tier, but the effect becomes statistically insignificant for the 90–100 group.

In Table 9, the interaction effects for CR growth are analyzed, despite the absence of significant nonlinearities in the direct group contrasts in Table 6. The interactions involving gender and wealth do not achieve statistical significance. However, higher debt volumes increase CR growth for 50–59 and 70–79 CS tiers relative to the baseline reference group, whereas higher income levels increase CR growth specifically for the highest CS group (90–100). As anticipated, the most substantial effect stems from the interaction with the baseline CR level (2013), which significantly reduces CR growth for nearly all CS groups compared to the baseline group, with the sole exception of the 60–69 bracket, and exhibits the most pronounced reduction within the 90–100 group.

To further assess the contrasts between CS levels, each control variable is partitioned into two distinct categories: women versus men, customers with versus without debt, and individuals below versus at or above the sample median for age, wealth, income, and baseline level of CR2013.

Table 10. Pairwise contrasts of customer revenue levels across customer satisfaction groups: Accounting for demographic and financial covariates in 2014.

CS-Groups		Full	Gender		Age		Debt		Wealth		Monthly Net Income		Customer Revenue Base
		Sample	Men	Women	Low	High	No Debt	Debt	Low	High	Low	High	Low	High
50–59	0–49	136.2975	151.9650	117.7155	217.8561	43.0994	70.1447	217.2963	107.9130	114.5102	116.1564	143.3334	92.5482	150.2922
60–69	0–49	208.1307	191.1971	230.0826	214.5943	193.2455	303.7099	90.1229	85.0823	257.1601	357.5581	72.1113	30.8902	311.6008
70–79	0–49	267.5055	284.8142	251.5217	216.2894	260.1996	317.0783	254.0676	175.8459	277.6495	450.8198	92.2129	68.2931	372.8883
80–89	0–49	433.4842	406.2808	463.321	247.8776	549.8603	578.8082	276.2401	186.3466	647.9745	587.1650	272.9701	132.9920	621.8230
90–100	0–49	225.0214	445.6002	72.8636	207.7256	132.5764	304.9582	164.1937	104.1843	291.8382	134.6781	363.9011	25.8475	296.0761
60–69	50–69	71.8332	39.2321	112.3670	−3.2618	150.1462	233.5652	−127.1735	−22.8307	142.6499	241.4016	−71.2221	−61.6580	161.3086
70–79	50–69	131.2080	132.8492	133.8062	−1.5667	217.1002	246.9336	36.7713	67.9329	163.1383	334.6634	−51.1205	−24.2550	222.5961
80–89	50–69	297.1866	254.3158	345.6055	30.0215	506.7609	508.6635	58.9438	78.4336	533.4643	471.0085	129.6367	40.4439	471.5308
90–100	50–69	88.7239	293.6352	−44.8519	−10.1305	89.4770	234.8135	−53.1026	−3.7287	177.3280	18.5216	220.5677	−66.7007	145.5308
70–79	60–69	59.3748	93.6171	21.4392	1.6951	66.9540	13.3684	163.9448	90.7636	20.4884	93.2618	20.1016	37.4030	61.2875
80–89	60–69	225.3535	215.0837	233.2384	33.2833	356.6147	275.0983	186.1172	101.2643	390.8145	229.6069	200.8588	102.1019	310.2222
90–100	60–69	16.8908	254.4031	−157.2189	−6.8687	−60.6691	1.2484	74.0708	19.1020	34.6781	−222.8800	291.7898	−5.0427	−15.5246
80–89	70–79	165.9787	121.4666	211.7993	31.5882	289.6607	261.7299	22.1724	10.5007	370.3260	136.3451	180.7573	64.6989	248.9346
90–100	70–79	−42.4840	160.7859	−178.6581	−8.5637	−127.6231	−12.1200	−89.8739	−71,6616	14.1897	−316.1418	271.6882	−42.4456	−76.8122
90–100	80–89	−208.4627	39.3194	−390.4574	−40.1519	−417.2839	−273.8499	−112.0465	−82.1623	−356.1363	−452.4869	90.9310	−107.1445	−325.7468
No. of observations		19,054	9611	9443	9619	9435	11,756	7298	9576	9478	9431	9623	9595	9459

N.B. Bold font implies significance at the 5% level; italics implies a 10% significance level.

presents these subsample results for the level of CR, focusing explicitly on the comparisons between the highest CS group (90–100) and the adjacent tiers (80–89 and 70–79). Female customers exhibit lower CR in the CS 90–100 bracket compared with the 70–89 bracket (p < 0.05). Similarly, older customers display lower CR in the 90–100 range compared to the 80–89 range (p < 0.10). Customers without debt and those with lower income levels also generate lower absolute CR in 2014 within the 90–100 range compared to the 80–89 range 89 (p < 0.05); conversely, high-income customers exhibit the opposite pattern, generating significantly higher CR in the 90–100 group (p < 0.05). Finally, the baseline CR level (2013) reveals no statistically significant differences across these satisfaction groups.

Table 11. Pairwise contrasts of revenue growth across customer satisfaction groups: Accounting for demographic and financial covariates in 2014.

CS-Groups		Full	Gender		Age		Debt		Wealth		Monthly Net Income		Customer Revenue Base
		Sample	Men	Women	Low	High	No Debt	Debt	Low	High	Low	High	Low	High
50–59	0–49	3.9437	1.3220	7.2354	5.4139	2.4278	1.2789	8.3147	4.9361	2.7406	2.5315	5.2065	6.0136	2.1154
60–69	0–49	3.7504	0.3677	7.5034	3.0769	4.5598	4.0257	5.1461	3.5033	3.4888	7.9298	0.3921	4.0769	2.5932
70–79	0–49	4.8646	6.0543	3.9273	3.9887	4.0373	4.6608	8.1487	3.4082	4.9380	4.7551	4.6835	2.3960	3.5825
80–89	0–49	12.5531	10.3107	15.0741	12.0806	9.4726	13.8557	14.4008	9.2237	14.0441	12.3812	12.3337	8.5938	7.3175
90–100	0–49	7.4486	9.8176	6.2923	8.3839	1.8703	10.2967	4.9964	5.9862	6.4039	4.0678	10.5206	4.9392	0.9290
60–69	50–69	−0.1933	−0.9543	0.2680	−2.3370	2.132	2.7378	−3.1685	−1.4328	0.7481	5.3983	−4.8144	−1.9367	0.4778
70–79	50–69	0.9209	4.7324	−3.3081	−1.4252	1.6096	3.3729	−0.1660	−1.5279	2.1974	2.2236	−0.5230	−3.6176	1.4671
80–89	50–69	8.6094	8.9887	7.8387	6.6667	7.0448	12.5678	3.3183	4.2876	11.3035	9.8498	7.1273	2.5802	5.2021
90–100	50–69	3.5049	8.4956	−0.9431	2.9700	−0.5575	9.0088	3.0025	1.0502	3.6633	1.5363	5.3141	−1.0744	−1.1864
70–79	60–69	1.1142	5.6866	−3.5761	0.9118	−0.5224	0.6350	9.2547	−0.0951	1.4493	−3.1747	4.2914	−1.6809	0.9893
80–89	60–69	8.8027	9.9430	7.5761	9.0037	4.9128	9.8300	−0.1497	5.7204	10.5554	4.4514	11.9416	4.5169	4.7242
90–100	60–69	3.6981	9.4499	−1.2111	5.2070	−2.6895	6.2709	6.2522	2.4829	2.9152	−3.8620	10.1285	0.8623	−1.6642
80–89	70–79	7.6885	4.2564	11.1468	8.0919	5.4353	9.1949	6.2522	5.8155	9.1061	7.7626	7.6502	6.1978	3.7350
90–100	70–79	2.5840	3.7633	2.3650	4.3952	−2.1671	5.6359	3.1522	2.5781	1.4659	−0.6873	5.8371	2.5431	−2.6535
90–100	80–89	−5.1045	−0.4931	−8.7818	−3.6967	−7.6023	−3.559	9.4044	−3.2374	−7.6402	−8.3135	−1.8131	−3.6547	−6.3884
No. of observations		19,054	9611	9443	9619	9435	11,756	7298	9576	9478	9431	9623	9595	9459

N.B. Bold font implies significance at the 5% level; italics implies a 10% significance level.

presents the subsample results for CR growth. Female customers, older individuals, and low-income groups all exhibit significantly lower CR growth in the highest satisfaction tier (90–100) compared to the adjacent tier (80–89) (p < 0.10). Additionally, customers with an above-median baseline CR level in 2013 display significantly lower CR growth within the 90–100 CS bracket relative to the 80–89 bracket (p < 0.05).

In summary, the findings indicate that gender, age, debt, and income play a vital role in explaining the nonlinearity observed within the relationship between CS and the absolute level of CR. In contrast, wealth and the baseline CR level (2013) do not appear to exert a statistically significant influence. Although no nonlinear relationships are established for CR growth, these subsample results suggest that gender, age, income, and the baseline CR level merit further empirical investigation as potential explanatory factors when analyzing the variations between the highest and second-highest CS groups.

5. Discussion

Drawing on a sample of 19,054 Swedish bank customers, this study investigates the relationship between CS and both the level and growth of CR in the banking sector. Specifically, we hypothesize that this relationship exhibits diminishing returns at the highest level of CS. Following the observation by Ittner and Larcker (1998) that substantial differences may exist between the level and growth of CR, our analysis explicitly distinguishes between these two dimensions. Furthermore, we hypothesize that any observed nonlinearity persists over time, extending up to four years after the initial measurement of CS. Finally, to fulfill the study’s exploratory objective of understanding relationship nonlinearity, we interact CS with demographic and socioeconomic variables, as well as the baseline level of CR—the inclusion of the latter following the empirical recommendations of Terpstra et al. (2014).

A notable initial finding concerns the low explanatory power of CS. While higher satisfaction is associated with both elevated revenue levels and growth, the independent variables explain only a modest share of the variance. The full model examining the relationship between satisfaction and revenue level yields an R² of 0.3118, whereas the model without control variables explains only 0.56% of the variance (R² = 0.0056). For revenue growth, the explanatory power is even lower (R² = 0.0209 for the full model and 0.0017 without controls). These results align with previous studies (Ittner and Larcker, 1998; Yu, 2007; Terpstra et al., 2014), underscoring that although CS is positively and significantly associated with CR, its effect size is small. Consequently, other observed and unobserved factors appear to play a more substantial role in determining revenue outcomes. Nonetheless, CS remains an actionable metric that banks can actively influence, and estimated coefficients demonstrate a higher financial return from increased satisfaction compared to the baseline reference group (with scores below 50).

Our first hypothesis posited that the relationship between CS and both revenue level (H1a) and revenue growth (H1b) displays diminishing returns at higher satisfaction levels. This hypothesis receives only weak support for revenue levels and no support for revenue growth. When nonlinearity is modeled quadratically using CS and CS², the relationship loses significance once the baseline level of revenue is included as a control alongside demographic and financial variables. However, when satisfaction is segmented into six distinct groups, and the two highest categories are contrasted, a diminishing returns pattern emerges at the 10% significance level for revenue levels. This provides weak support for H1a and none for H1b.

This ceiling effect, characterized by diminishing returns at the peak satisfaction levels, may be linked to mental accounting (Thaler & Shefrin, 1981) and budget constraints. Empirically, we find evidence that lower income levels contribute to reduced revenue within the highest satisfaction tier. Prior research provides a plausible interpretation of this pattern, demonstrating that lower customer involvement is associated with weaker outcomes at comparable satisfaction levels (Eisenbeiss et al., 2014). Specifically, among customers with no debt and lower income, revenue declines at the highest satisfaction level relative to the second-highest satisfaction group. Interestingly, neither low nor high wealth appears to influence these revenue discrepancies. Accounting for the baseline level of revenue in the regression models proves essential. Once this control is included, the quadratic nonlinearity in the relationship between revenue and CS and CS² disappears. Moreover, when interacting baseline revenue with the discrete CS groups to examine the slope of the relationship, no significant interaction is observed at the highest satisfaction level (scores 90–100).

Interpretations regarding the effects of gender and age can be informed by Eisenbeiss et al. (2014) and Baumann et al. (2005), respectively. In our dataset, the gender effect is statistically significant, implying a ceiling effect among women at the highest satisfaction levels. However, disentangling this effect from confounding customer characteristics remains challenging. Older women with low income, no debt, and low wealth likely exhibit lower financial involvement and a limited capacity to increase their spending baseline despite high satisfaction. We find that gender (p < 0.05) and age (p < 0.10) significantly contribute to the observed diminishing returns at high satisfaction levels when these control variables are interacted with discrete CS groups. Furthermore, when splitting the sample by gender, age, debt status, wealth, income, and baseline revenue, significant effects persist for age, debt, and income (p < 0.05), with age remaining marginally significant at the 10% level. Conversely, wealth and the baseline level of revenue do not contribute to the observed diminishing returns.

Furthermore, we find no support for H2, which posits that the nonlinear effects observed after one year persist over longer horizons, as suggested by path dependence theory (North, 1994). This finding directly contrasts with the earlier research by Yu (2007) and Terpstra et al. (2014). Specifically, the weak support for H1a (p < 0.10) identified in the first year (2014) vanishes entirely in subsequent periods (2015–2017). Among the control variables, only the gender effect remains stable and consistent over time.

In sum, our findings reveal that the nonlinearity is strictly confined to first-year revenue volumes, exhibiting a pattern consistent with diminishing marginal returns and potential over-saturation. For revenue levels, we find weak support for a decrease in revenue at the highest levels of customer satisfaction, paired with a complete, unhypothesized absence of any revenue increase. This pattern aligns with the concept of diminishing returns characterized by, for example, Anderson and Mittal (2000). It suggests that pursuing extreme satisfaction may hit a saturation zone; once customers are fully satiated, their purchasing volume stabilizes or slightly contracts as they reach their maximum wallet share capacity with the firm. Furthermore, achieving these peak satisfaction scores often requires resource-intensive promotional incentives that can transiently depress absolute revenue volume in the short term. For percentage revenue growth, the relationship displays a flat, symmetric neutrality, with no significant increases or decreases on either end of the satisfaction spectrum. Because the minor volume contraction at peak satisfaction is a transient, first-year phenomenon, it lacks the sustained momentum required to alter the compounding velocity of revenue. Once the baseline revenue is controlled for, percentage growth remains entirely insulated from these localized satisfaction dynamics. Ultimately, this isolation reveals that path dependence theory, contrary to conventional assumptions, lacks explanatory power for revenue growth, as short-term satisfaction fluctuations fail to anchor or dictate long-term financial velocity.

For bank leadership, a key implication of customer satisfaction’s low explanatory power is that it may constitute a necessary, but ultimately insufficient, condition for driving CR. Accordingly, management might consider a more cautious approach when interpreting, incentivizing, and operationalizing satisfaction metrics within performance systems. Given these tentative findings, bank leadership could explore integrating satisfaction scores with objective behavioral indicators, such as product usage, digital engagement, and transaction intensity, while carefully factoring financial capacity into predictive revenue models. Combined, these indicators might provide a more nuanced, real-time representation of actual customer engagement. Crucially, while our evidence regarding the upside of maximizing satisfaction remains weak, prior research (e.g., Segerlind et al., 2026) demonstrates that lower satisfaction levels can exert persistent negative effects lasting up to four years. Consequently, maintaining these attitudinal measures may still have strategic relevance.

To further refine managerial insight, bank leadership could explore measuring CS at a more granular level, potentially incorporating domain-specific dimensions such as product holdings, channel usage, and service interactions. Synthesizing these localized satisfaction metrics with observed behavioral data may facilitate a more balanced assessment of both the depth and intensity of the bank–customer relationship. Ultimately, this integrated diagnostic approach could offer a helpful framework for predicting customer retention and informing broader strategic decisions.

Preliminary results suggest that revenue peaks within the satisfaction interval of 80–89, with only weak evidence of diminishing returns beyond the 90 threshold. As an exploratory study, these findings provide initial insights into the satisfaction–revenue relationship that require further validation. Given the tentative nature of these findings, managers might consider investigating whether satisfaction is better conceptualized as a threshold variable rather than a strictly monotonic, “more-is-better” objective.

The analysis hints at the possibility that customers with moderate-to-high satisfaction levels could represent an important pocket of revenue potential. There is a tentative suggestion that improvements within this specific segment might generate more notable revenue increases than incremental gains among already highly satisfied customers. These preliminary findings may point to the value of exploring refined customer segmentation strategies that identify customers in the mid-satisfaction range. Future initiatives could explore targeting this group to deepen relationships, increase product adoption, and stimulate cross-selling before satisfaction deteriorates into lower tiers. Rather than focusing resources solely on “delighting” already satisfied customers, banks may explore whether higher returns on marketing investments exist in converting moderately satisfied customers into strongly satisfied ones. Because investments aimed at moving customers from high satisfaction levels (90–100) to maximum satisfaction levels may yield limited financial returns, resource allocation could potentially shift from maximizing top-box scores to ensuring a broader base of customers reaches a high, but not extreme, satisfaction range.

In addition, the findings indicate that demographic and financial characteristics, including income, debt status, age, and gender, are associated with revenue outcomes. Revenue capacity appears partly constrained by customers’ financial resources and life-cycle stage. For instance, low-income older women without outstanding debt exhibit ceiling effects at high satisfaction levels, suggesting that high satisfaction cannot fully compensate for limited financial capacity or low product involvement.

In summary, the relationship between CS and CR appears more complex than a simple linear association. Although higher satisfaction generally corresponds to higher revenue, this exploratory study provides tentative evidence that the effect may plateau within the 80–89 satisfaction range. Beyond this level, additional increases in satisfaction may not yield significant revenue gains, and diminishing returns emerge only at the 10% significance level when analyzing absolute revenue levels. The nonlinear pattern appears most pronounced among women, low-income customers, and individuals without debt; support for age-based variations is weak, and no evidence was found for wealth effects. Overall, CS shows limited explanatory power, suggesting that it may not serve as a universally robust driver of CR.

A key methodological limitation involves the potential for omitted variables bias, which could introduce endogeneity and affect the estimation results. Future research might, therefore, incorporate additional psychological, behavioral, and contextual constructs to better explain customer behavior and revenue variations at the upper bounds of the satisfaction spectrum. In particular, linking customer satisfaction and revenue to specific digital and face-to-face service channels represents a promising avenue for future investigation.

While these preliminary findings offer only tentative implications for strategic customer management, they suggest that for already highly satisfied customers, further improvements in satisfaction may not produce proportional revenue gains. Resource allocation could instead be more effectively explored within low- and medium-satisfaction customer segments to strengthen future CR. This highlights a potential opportunity to develop alternative marketing and segmentation strategies aimed at maintaining or improving engagement and loyalty among these specific groups. Ultimately, CS may be conceptualized not as a standalone metric but as a foundational driver of CR that is embedded within a broader, analytically integrated customer relationship strategy.

6. Conclusions

Customer satisfaction (CS) has long been recognized as a key performance indicator in marketing theory and practice. This study empirically investigates the nonlinear relationship between CS and both the level and growth of customer revenue (CR) at the individual customer level within the banking sector. Specifically, it tests the hypothesis of diminishing returns at the upper end of the satisfaction scale and examines how customer characteristics explain this nonlinearity. In addition, it explores the temporal dynamics by assessing whether these nonlinear effects persist for up to four years after the initial satisfaction measurement.

Drawing on a unique dataset of 19,054 Swedish bank customers, which integrates subjective survey data with objective financial records and demographic controls, the results reveal a significant positive association between CS and CR. Consistent with previous research, this relationship is nonlinear. The data suggests diminishing returns in revenue levels among the most satisfied customers: those with satisfaction scores between 80 and 89 exhibit higher revenue levels than those scoring between 90 and 100. However, this ceiling effect is weak (significant only at the 10% level), and no diminishing returns are observed for revenue growth.

Another finding highlights the relatively low explanatory power of CS. Although positively associated with revenue, its effect size is small (less than 1%), indicating that other observed and unobserved factors exert a stronger influence on revenue outcomes. Furthermore, older, lower-income women with no debt are more likely to exhibit ceiling effects at high satisfaction levels, whereas wealth does not influence these outcomes. Finally, the analysis finds no evidence that the first-year nonlinear effects persist over subsequent years, although gender emerges as the most consistently influential factor over time.

By utilizing an extensive dataset, this study contributes to both theoretical research and managerial discussion, enabling an exploratory analysis of nonlinearity while accounting for key control variables. Our results provide only weak, transient evidence of a ceiling effect in absolute revenue levels, with no such pattern emerging for percentage revenue growth. Furthermore, the overall explanatory power of customer satisfaction remains limited. These patterns offer cautious insights for strategic management, tentatively suggesting that expanding relationships with moderately to highly satisfied customers may yield more observable revenue variations than seeking incremental gains among top-tier satisfied segments. Consequently, future research could explore refined segmentation models tailored to mid-range satisfaction intervals. For exceptionally satisfied accounts, alternative marketing and segmentation strategies may be worth exploring to optimize revenue generation. Ultimately, these results underscore the value of evaluating consumer sentiment at the individual relationship level, motivating future research to replicate these patterns and investigate alternative mechanisms driving revenue generation.