From a Homogeneous Service Environment to Consumer Learning

The focus of this study is based on the Stimulus-Organism-Response (S-O-R) model. Mehrabian and Russell (1974) asserted that individuals’ physical environments result in approach and avoidance behaviors. This depends on the individual’s feelings and thoughts. In analyzing service consumption, Bitner (1992) introduced the term “service environment” to represent the controlled, built physical environment (the stimulus) that an organization uses to influence the customer’s response. Furthermore, Han and Ryu (2009) described the service environment in restaurants as comprising three dimensions: décor and artifacts, spatial layout, and environmental conditions. Within this model, the physical environment acts as the stimulus; price perception and satisfaction are the organism’s internal states in response to the stimulus; and loyalty is the resulting behavioral and attitudinal response (Ahmed et al., 2023; Oliver, 1999).

The E-O-R model typically assumes that the stimulus-organism-response system operates at the same intensity levels for all customers. Consumer learning theory challenges this assumption. Hoch and Deighton (1989) suggested that the formation of experiential judgments is a four-stage process (hypothesis, exposure, encoding, and integration) with three moderators, one of which is familiarity with the domain, and is central to this theory. The more familiar an individual is with a given domain, the more they encode new information into sophisticated structures and the less they rely on the most salient and accessible pieces of information in the environment. Park et al. (2019) applied the theory in the context of a service environment and found that brand familiarity affects the influence of the physical environment on positive feelings and behavioral intentions, especially among customers who are less familiar with the brand. Therefore, by combining both theories, it is expected that the influence of the service environment will vary depending on the frequency of diners’ visits. The following sections define the relationships in the combined model (direct effects, simple mediating effects, and chains of mediating effects) and the expected variations between infrequent and frequent customers.

The Servicescape as a Predecessor to Price Perception

From the cognitive perspective of the S-O-R model, the components of the physical environment function as cues that diners interpret to infer whether the price charged is commensurate with the establishment’s category (Bitner, 1992; Han & Ryu, 2009). The décor and artifacts (r color schemes, furniture, table linens, visual art) constitute the most immediate component of this interpretation, as they communicate quality and identity from the very first contact and anchor the customer’s spending expectations (Jamaludin & Hashim, 2024; Liu & Jang, 2009). When these cues convey care and consistency, diners tend to view a higher price as reasonable. Based on this, we propose:

H1. Decor and furnishings positively influence price perception.

The arrangement of furniture and the design of traffic flow demonstrate how a business anticipates customer spending. A business that shows consideration in table spacing, furniture layout, and traffic flow exhibits superior service and business operations. These factors influence consumers’ perception that higher purchase costs are more reasonable (Huerta-Tantalean et al., 2024; Line & Hanks, 2019; Lu & Wang, 2017; Othman et al., 2025). Consequently, we hypothesize:

H2. Spatial layout positively influences price perception.

While operational conditions such as light, sound, and temperature—as well as smell, taste, and the degree of sensory stimulation—are absorbed, the location’s position in relation to these elements is equally communicative. Evidence shows that environmental stimuli appropriately calibrated for the designed experience enhance the experience and reinforce the perception of value received in relation to the environment provided (Apaza-Panca et al., 2023; Asghar Ali et al., 2021; Guillen et al., 2025; Ramos Farroñán et al., 2024). Therefore, we propose:

H3. Ambient conditions positively influence price perception.

Background on Customer Satisfaction

Perceived price is a consistent determinant of satisfaction in full-service restaurants, where the average check is high and price sensitivity is heightened. Ing et al. (2019) reported that, among the transactional characteristics evaluated by diners, perceived price was the strongest predictor of satisfaction, and Ahmed et al. (2023) confirmed that the perceived reasonableness of the price underpins the overall evaluation of the experience—a pattern replicated in various contexts (Avanica et al., 2025; Li et al., 2025; Nejati & Parakhodi Moghaddam, 2013). Hence:

H4. Perceived price positively influences customer satisfaction.

In addition to influencing price, décor can affect satisfaction through the hedonic gratification factored into diners’ judgments. According to Jamaludin and Hashim (2024), decorative elements designed to tell a story increase the perception of comfort and stimulation. Furthermore, the quality of the environment and the satisfaction of Peruvian diners are positively correlated. (Apaza-Panca et al., 2023; Chinelato et al., 2023). On this basis, the following hypothesis is proposed:

H5. Decor and artifacts positively influence customer satisfaction.

Ambient conditions play a similar role in the affective pathway. Warm lighting, moderate music volume, a comfortable temperature, and a pleasant aroma increase the length of stay and the overall evaluation of the visit. (Asghar Ali et al., 2021; Bujisic et al., 2025). Consequently, we hypothesize:

H6. Ambient conditions positively influence customer satisfaction.

The current model refers to spatial layout as a precursor to price perception rather than to satisfaction. The spatial layout of furniture is treated as a hygiene factor. The absence of furniture causes discomfort, while the presence of furniture—assuming that decorative and environmental factors are controlled—does not lead to positive satisfaction. (Asghar Ali et al., 2021; Han & Ryu, 2009; Rai & Anirvinna, 2019). This theoretical decision, which streamlines the original framework, aims for a more parsimonious model specification.

Satisfaction and Loyalty

Satisfaction and loyalty are perhaps the most well-established link in consumer behavior. Expectation confirmation theory suggests that an overall evaluation leads to a psychological bond with the service provider and, subsequently, to a commitment to return, recommend the service, and exceed expectations in terms of spending (Oliver, 1980, 1999). In the restaurant industry, this bond has been repeatedly demonstrated and is the immediate driver of attitudinal loyalty (Coelho & Henseler, 2012; Flores Curico et al., 2023; Mattison Thompson et al., 2014; Rivera Paredes et al., 2025). Therefore, the following hypothesis is proposed:

H7. Customer satisfaction positively influences customer loyalty.

The impact of price on loyalty is modeled both directly and indirectly. More recent studies suggest that when satisfaction is taken into account, the direct effect of price on loyalty decreases, but the indirect effect of price through satisfaction remains valid (Ahmed et al., 2023; Jin et al., 2012; Souki et al., 2023). This modeling approach is expressed in the following mediation hypotheses.

Simple Mediations

The cognitive logic of the S-O-R model implies that the physical environment does not always have a direct impact on satisfaction: part of its effect is channeled through the price judgment that the customer forms based on those cues (Bitner, 1992; Han & Ryu, 2009). When the environment is well-designed, the diner perceives the cost as appropriate, and this favorable price judgment increases their satisfaction. Since decor and environmental conditions have direct effects on price (H1, H3), it is plausible that price acts as a bridge to satisfaction; spatial layout is incorporated due to the symmetry of the model. On this basis, the following hypotheses are proposed:

H8. The effect of decor and artifacts on customer satisfaction is mediated by price perception.

H9. The effect of spatial layout on customer satisfaction is mediated by price perception.

H10. The effect of environmental conditions on customer satisfaction is mediated by price perception.

At the same time, satisfaction acts as the evaluative filter that either strengthens or inhibits loyalty. A pleasant environment or a price perceived as reasonable does not automatically translate into commitment: they must first crystallize into a positive overall evaluation (Oliver, 1999; Souki et al., 2023; Vera-Falcón et al., 2025). Since decor, environmental conditions, and price precede satisfaction (H4–H6), it is hypothesized that satisfaction mediates their effects on loyalty:

H11. The effect of decor and artifacts on customer loyalty is mediated by satisfaction.

H12. The effect of environmental conditions on customer loyalty is mediated by satisfaction.

H13. The effect of perceived price on customer loyalty is mediated by satisfaction.

Serial Mediations

The complete articulation of the S-O-R suggests a staggered causal chain in which two mediators act in sequence. The diner first processes the tangible cues from the environment, translates them into a perception of monetary value, integrates that judgment into an overall evaluation, and, finally, that satisfaction crystallizes into attitudinal commitment (Han & Ryu, 2009; Oliver, 1980). Previous literature has rarely formally tested complete serial chains in the restaurant industry, even though partial least squares modeling allows for rigorous estimation using the procedure by Preacher and Hayes (2008) and the typology by Zhao et al. (2010). On this basis, the study proposes the following three serial hypotheses:

H14. The effect of decor and artifacts on customer loyalty is transmitted sequentially through price perception and satisfaction.

H15. The effect of spatial layout on customer loyalty is transmitted sequentially through price perception and satisfaction.

H16. The effect of environmental conditions on customer loyalty is transmitted sequentially through price perception and satisfaction.

To formally test the assumed structural homogeneity of the original model and the replicas that have had the greatest influence (Han and Ryu, 2009; Rai et al., 2021; Ryu et al., 2012), a comparison of the a priori established subpopulations is sufficient. According to consumer learning theory, the frequency of model application is determined by the frequency of visits to the model (Burgos Cabanillas et al., 2026; Hoch and Deighton, 1989; Park et al., 2019). Customers with a limited history of local consumption (occasional customers) will judge the consumption experience based on accessible and cognitively economical cues (e.g., price); therefore, the price → satisfaction relationship will be more valid for this segment. In contrast, frequent customers, with richer cognitive frameworks, will place greater emphasis on aesthetic and decorative stimuli; thus, the relationship between decor and satisfaction will be more valid for this segment. From this, the first two hypotheses of difference emerge:

H17. The effect of perceived price on customer satisfaction is significantly greater among occasional customers than among frequent customers.

H18. The effect of decor and artifacts on customer satisfaction is significantly greater among frequent customers than among occasional customers.

This same logic should extend to loyalty through the channel of satisfaction. If decor carries more weight in the satisfaction of repeat customers, its indirect effect on loyalty should also be greater in that segment; and if price carries more weight in the satisfaction of occasional customers, its indirect effect on loyalty should be concentrated in that segment (Hoch & Deighton, 1989; Othman et al., 2025; Park et al., 2019). Based on this, we propose:

H19. The indirect effect of decor and artifacts on customer loyalty, via satisfaction, is significantly greater for repeat customers than for occasional customers.

H20. The indirect effect of perceived price on customer loyalty, via satisfaction, is significantly greater for occasional customers than for repeat customers.

These four differences are tested through a multigroup structural equation model estimated after establishing measurement invariance: each contrast is expressed as an explicit parameter difference with a bias-corrected bootstrap confidence interval, supplemented by an importance-performance analysis that translates the total effects into management priorities for each segment.

The conceptual model integrates the direct effects, the simple and serial mediations, and the hypothesized differences between segments developed above.

This study was conducted in accordance with the principles of the Declaration of Helsinki. It received ethical approval from the Institutional Ethics Committee 2026-IIICyT-ITCA of the Institute for Research, Innovation, Science, and Technology (IIICyT; registration number PE 11358714, Peru), under approval code 0081–2026-GM-IIICyT, granted on 13 January 2026. Written informed consent to participate was obtained from every participant before data collection. Because the questionnaire was administered online, this consent was recorded electronically: on the first screen, respondents read an information sheet describing the purpose, procedures, and voluntary and anonymous nature of the study, and then documented their agreement by selecting a mandatory acceptance checkbox, which functioned as a written record of consent and without which the survey could not be started. Electronic written consent was chosen instead of a handwritten signature because it suited the self-administered online format, left an auditable record, and avoided collecting any handwritten identifier that could compromise the anonymity of respondents. All participants were adults (18 years of age or older). Participation was voluntary and anonymous and could be withdrawn at any time without penalty, and no directly identifying data were retained.

Six constructs were measured using proxy indicators. These items were adapted from the scale validated by Han and Ryu (2009). Twenty-five items from the original scale were translated into Spanish and contextualized for the cuisine of northern Peru. The theoretical meaning of each item was preserved. This adaptation was based on Apaza-Panca et al. (2023) and Chinelato et al. (2023) and followed the guidelines for scale development (Churchill, 1979). This was done specifically for the context of Peruvian restaurant patrons. A seven-point Likert scale (Likert, 1932) was used for all items (1 = strongly disagree, 7 = strongly agree). Decor and furnishings were assessed using eight items; three items assessed spatial layout; environmental conditions were assessed using six items; and price perception was assessed using two items. Customer satisfaction and loyalty were each assessed using three items. To ensure linguistic equivalence, two bilingual researchers translated the survey items. A back-translation was performed to ensure faithful meaning, and a panel of three people (two academics specializing in consumer behavior and one manager) ensured that the items were clear, culturally relevant, and valid. This process was based on the criteria established by Aiken (1985) and Lawshe (1975). A pilot study was conducted with 30 restaurant patrons to ensure that the items were understandable.

Fieldwork was conducted from February to April 2026 in five cities that constitute the gastronomic heart of northern Peru’s macroregion: Chiclayo, Trujillo, Piura, Cajamarca, and Tumbes. These cities have proven significant in the study of gastronomic consumers (Bautista et al., 2023; Esparza-Huamanchumo et al., 2025). Including Cajamarca and Tumbes alongside the three centers examined in earlier research covers the entire macroregion and adds socioeconomic diversity to the study framework, given the relatively low income levels of both areas. A total of 90 formal full-service restaurants covering different price categories and styles were visited across the five cities; establishments were selected by trained assistants within the formal segment of the region’s restaurant industry.

A non-probabilistic, purposive convenience sampling method combined with chain sampling was implemented. Participants were asked to take part at the end of their meal to capture the full dining experience; all were over 18 years of age and had made a purchase at the establishment under study. A total of 705 valid questionnaires were retained after eliminating incomplete responses. To minimize social desirability bias and presentation-order effects, the items were randomized into four unique questionnaires, which participants completed unsupervised. A screening for careless responding identified 46 cases (6.5%) with zero within-person variance across the 25 items; these were retained in the main analysis, and a sensitivity re-estimation excluding them (n = 659) reproduced every substantive conclusion.

For the multigroup analysis, participants were classified as customers who visit full-service restaurants occasionally (n = 304) versus frequently (n = 401). Both groups engaged with the restaurant’s service; the distinction captures low versus high behavioral engagement rather than first-time trial. Sample size comfortably exceeds conventional requirements for covariance-based estimation: with 25 indicators and 6 latent constructs, the N:q ratio surpasses 5:1 in each segment, and under the inverse square root method (Kock and Hadaya, 2018) the minimum sample to detect the weakest expected path (β ≈ .15) at 80% power and α = .05 is approximately 274, which each group exceeds on its own.

Data analysis combined two complementary environments, chosen to maximize the rigor and reproducibility of the evidence. Model estimation was performed in Mplus 8.3 (Muthén and Muthén, 2017) using covariance-based structural equation modeling (CB-SEM): the measurement model and the invariance tests employed the robust maximum likelihood estimator (MLR), warranted by the multivariate non-normality of the items (Mardia’s normalized multivariate kurtosis of 933.7 against an expectation of 675 under normality), while the structural and mediation models used maximum likelihood with 10,000 bootstrap draws and bias-corrected confidence intervals, the appropriate treatment for products of coefficients whose sampling distribution is asymmetric. All statistical graphics—the estimated path diagram, the importance-performance maps, and the between-segment difference plot—were produced in RStudio (R 4.4; R Core Team, 2024), where the full model was additionally re-estimated with the lavaan package (Rosseel, 2012) as an independent cross-validation of every Mplus estimate. Common-method variance was mitigated procedurally through anonymity and randomized item order, and examined empirically with a confirmatory single-factor (Harman) model, which fitted the data far worse than the six-factor solution (CFI = .863 vs. .954; RMSEA = .107 vs. .064; ΔAIC = 2,184), indicating that method variance alone cannot account for the observed covariance structure.

In accordance with the two-step approach (Hair et al., 2019, 2022)—first the measurement model, then the structural model—the following analyses were conducted. Standardized loadings above 0.708 were required for indicator reliability; internal consistency was assessed with Cronbach’s alpha (Cronbach, 1951) and composite reliability, and convergent validity with an average variance extracted (AVE) above 0.50 (Bagozzi and Yi, 1988; Fornell and Larcker, 1981). Discriminant validity combined the Fornell-Larcker criterion and the HTMT ratio (Henseler et al., 2015) with a stronger, covariance-based test: the hypothesized six-factor model was compared against rival specifications that collapsed adjacent constructs, using CFI, RMSEA, AIC, and BIC. Global fit was judged with CFI/TLI ≥ .90 (ideally ≥ .95), RMSEA ≤ .08, and SRMR ≤ .08. Because comparing structural coefficients across groups presupposes equivalent measurement, configural, metric, and scalar invariance were tested sequentially with ΔCFI ≤ .010 and ΔRMSEA ≤ .015 as decision criteria (Chen, 2007). The structural model reports standardized coefficients, bias-corrected 95% bootstrap intervals, and R² for each endogenous construct; simple and serial mediations were evaluated through specific indirect effects (Preacher and Hayes, 2008), with mediation type classified following Zhao et al. (2010).

Two additional analyses address the study’s central question. The first is the multigroup structural model: with loadings constrained to the metric-invariant solution, the seven structural paths were freed in both segments, and the four heterogeneity hypotheses (H17–H20) were operationalized as explicit parameter differences estimated with 10,000 bootstrap draws and bias-corrected 95% confidence intervals in Mplus, supplemented by a Wald omnibus test of the joint equality of the two focal paths. A hypothesis is supported only when the interval for its difference excludes zero in the predicted direction, a more transparent criterion than the convergence of distribution-based tests used in earlier PLS applications. The second is an importance-performance analysis computed in RStudio as the covariance-based analogue of the IPMA (Ringle and Sarstedt, 2016): importance is the standardized total effect of each construct on loyalty, performance is the construct mean rescaled to a 0–100 metric, and the resulting maps are drawn for the full sample and for each segment. Three robustness checks complete the analytical strategy: a re-estimation of the structural model with a second-order servicescape factor, a specification test constraining all seven structural paths to equality across segments (Satorra-Bentler scaled chi-square difference), and a re-estimation adding city indicators as covariates of the endogenous constructs.

The sample consists of 705 adult diners from the northern macroregion of Peru, distributed by city of survey administration as follows: Chiclayo (30.1%), Piura (22.1%), Trujillo (21.6%), Cajamarca (15.6%), and Tumbes (10.6%). The profile is balanced by gender (50.5% male, 49.4% female) and predominantly young (50.1% between 18 and 24 years old and 26.7% between 25 and 39), with 54.9% reporting university or technical education. Income concentrates in the lower strata: 38.3% report up to S/ 1,130 per month and a further 29.9% between S/1,131 and S/2,260, consistent with the inclusion of Cajamarca and Tumbes, two of the areas with the lowest relative income in the macroregion. Based on visit frequency, 401 participants (56.9%) form the frequent group and 304 (43.1%) the occasional group, a division that comfortably supports the multigroup analysis. Table 1 details the sociodemographic profile.

Table 2 offers a first descriptive look at the two segments: across the six constructs, frequent customers show a somewhat higher proportion of high scores than occasional customers, with the widest gaps in satisfaction (49.6% vs. 43.1%) and loyalty (47.4% vs. 43.1%). These are differences of level, not of structure; whether the relationships among constructs differ between segments is examined formally in Section 3.4.

The measurement model meets reflective criteria. Standardized loadings range from.856 to.937, all above the.708 threshold, and Cronbach’s alpha (.927–.967), composite reliability (.929–.968), and AVE (.788–.877) confirm reliability and convergent validity ( Table 3). Global fit of the six-factor model is adequate: χ²(260) = 1,011.86 (MLR), CFI = .954, TLI = .947, RMSEA = .064 (90% CI.060, .068), SRMR = .019. Discriminant validity deserves close attention: four pairs of constructs show HTMT ≥ .90, and the latent correlations among the servicescape dimensions reach.94 ( Table 4). Model comparison nonetheless supports the six-factor specification: collapsing the three servicescape dimensions into a single factor worsened fit markedly (CFI = .924; RMSEA = .081; ΔAIC = 709; ΔBIC = 668), and merging satisfaction with loyalty was also inferior (CFI = .948; ΔAIC = 152; ΔBIC = 129). A second-order servicescape specification fitted virtually identically to the first-order model (ΔAIC = 7); the first-order solution was retained for hypothesis granularity, and the overlap is acknowledged as a limitation. Finally, measurement invariance across segments was fully supported: fit did not deteriorate from the configural to the metric model (Δχ²(19) = 19.47, p = .427) nor from the metric to the scalar model (Δχ²(19) = 17.70, p = .543), with ΔCFI ≤ .001 at each step, licensing the structural comparisons of Section 3.4. As a further safeguard, the full structural model was re-estimated with a second-order servicescape factor: the evaluation chain held intact (servicescape → price perception: β = .825; servicescape → satisfaction: β = .658; price perception → satisfaction: β = .320; satisfaction → loyalty: β = .941; all p < .001, with the serial indirect effect remaining significant), confirming that the overlap among servicescape dimensions does not alter the substantive conclusions.

The model explains a substantial proportion of variance in the three endogenous constructs: R² = .671 for price perception, .874 for satisfaction, and.883 for loyalty. In the price-perception block, decor (β = .424; p = .001) and ambient conditions (β = .472; p < .001) are significant antecedents, whereas spatial layout is not (β = −.063; p = .671). The three hypothesized paths toward satisfaction were confirmed: price perception (β = .348), decor (β = .230), and ambient conditions (β = .410), all with p ≤ .001. The strongest estimate in the model is the satisfaction → loyalty path (β = .940; 95% CI [.915, .960]), which positions satisfaction as the near-exclusive antecedent of declarative loyalty. Simple and serial mediations confirm that the effects of the physical environment and price on loyalty travel through satisfaction, with significant serial chains for decor (β = .146) and ambient conditions (β = .153) but not for spatial layout (β = −.020). Table 5 summarizes hypotheses H1 through H16. The estimates are also robust to sample composition: re-estimating the model with city indicators as covariates of the three endogenous constructs left every structural path virtually unchanged (maximum shift in β ≤ .01) and produced no significant city effect (all p ≥ .102).

The multigroup analysis addresses the core question of the study: whether the structure of the model changes with visit frequency. With metric invariance imposed on the loadings, the seven structural paths were freed in both segments and each hypothesized difference was estimated with a bias-corrected bootstrap interval. None of the four contrasts reached significance. The price → satisfaction path is descriptively stronger among occasional customers (β = .399 vs. .304), but the difference does not exclude zero (Δ = .101; 95% CI [−.056, .262]), so H17 is not supported. The decor → satisfaction path shows the reverse pattern—significant only among frequent customers (β = .331, p < .001, vs. β = .097, p = .389)—yet the between-group difference again includes zero (Δ = .232; 95% CI [−.065, .524]), so H18 is not supported either ( Table 6).

The same picture extends to loyalty through satisfaction. The indirect effect of decor on loyalty is significant only among frequent customers (unstandardized estimate.327; 95% CI [.151, .503], vs. .096 [−.128, .316] among occasional customers), while the indirect effect of price is significant in both groups. In neither case, however, does the between-segment difference exclude zero (H19: Δ = .231 [−.052, .510]; H20: Δ = .083 [−.064, .236]), and the Wald omnibus test of joint equality of the two focal paths is likewise non-significant, χ²(2) = 3.56, p = .169. The four multigroup hypotheses are therefore not supported: with a sample more than twice the size of previous regional studies, the evaluation chain proves structurally invariant across segments, and the segment-specific activation of the decor channel remains a suggestive but statistically unconfirmed pattern. A complementary specification test reinforces this reading: constraining the seven structural paths to equality across segments did not deteriorate model fit (Satorra-Bentler scaled Δχ²(7) = 10.83, p = .146), turning the absence of significant differences into positive evidence of structural invariance.

The importance-performance analysis translates the total effects into managerial coordinates. In the full sample, satisfaction combines the highest importance (.940) with the highest performance (65.3), acting as the universal anchor of loyalty. Among frequent customers, decor (importance = .432) and ambient conditions (.442) sit close together as secondary levers while price perception recedes (.285); among occasional customers, ambient conditions dominate the actionable space (.678), followed by price perception (.375), with decor further back (.254). Spatial layout carries no practical importance in either segment. Consistent with the invariance results, the two maps differ in emphasis rather than in kind: the same constructs order loyalty in both segments, and the contrast is a matter of degree concentrated in the decor and price channels ( Table 7).

Figures 1 and 2 show the importance-performance maps for frequent and occasional customers, respectively, and illustrate the shift in emphasis between the two segments.

Table 8 summarizes the twenty hypothesis tests: thirteen of the sixteen direct and mediation hypotheses are supported (the three paths involving spatial layout did not reach significance), and none of the four multigroup hypotheses is confirmed. Figure 3 depicts the estimated structural model with all standardized path coefficients.

The main finding of this research is not the heterogeneity it set out to document but the robustness it found instead. With measurement invariance secured and a sample of 705 diners, none of the four hypothesized between-segment differences excluded zero, and the Wald omnibus test corroborated the joint equality of the focal paths. The evaluation chain of the servicescape model (Han and Ryu, 2009) proves structurally stable across visit-frequency segments, which qualifies—for this market—the expectation derived from consumer learning theory (Hoch and Deighton, 1989) that experience reorders the weights diners assign to price and to the physical environment. One trace of that expectation persists: the decor → satisfaction path is significant only among frequent customers (β = .331 vs. .097), and the corresponding indirect effect on loyalty is confined to that segment. The prudent reading is that learning effects, if present, operate at the margin of a fundamentally common structure; whether they consolidate into significant differences may require stronger experience contrasts—first-time versus habitual patrons, for instance—than the frequency split used here.

A second reading concerns statistical discipline in segment research. The direction of every contrast matched the theoretical prediction—price weighed descriptively more among occasional customers and decor among frequent ones—yet the interval estimates kept all differences compatible with zero. Findings of this kind are vulnerable to overinterpretation when tested with less conservative procedures; the combination used here (invariance testing, explicit difference parameters with bootstrap intervals, and an omnibus Wald test) offers a template for separating suggestive patterns from confirmed heterogeneity. The importance-performance analysis, in turn, converts the invariant structure into management coordinates: satisfaction anchors loyalty in both segments, and the maps differ in emphasis rather than in kind.

For the entire sample, the model replicates and extends the logic of the servicescape framework. Decor and ambient conditions predict price perception and satisfaction, while spatial layout shows no significant effect on either outcome, consistent with its status as a hygiene factor (Asghar Ali et al., 2021). Satisfaction has near-total predictive power over loyalty (β = .940; R² = .883), exceeding the coefficients typically observed in international studies and consistent with recent meta-analytic evidence on the centrality of satisfaction (Chi and Phan, 2025). The serial chains servicescape → price → satisfaction → loyalty are significant for decor and ambient conditions, completing the evaluation chain and placing this study among the few that test it formally in the restaurant industry—and the first to do so in this region with a covariance-based estimator and full measurement-invariance safeguards.

This study makes three theoretical contributions. First, it converts the homogeneity assumption of the servicescape model from a premise into a hypothesis and, contrary to the drift of the segmentation literature, retains it: the structure survives a well-powered, invariance-protected multigroup test. Second, it refines the scope of consumer learning theory in this domain: experience does not reweight the whole evaluation chain but concentrates its observable influence in a single channel—decor—whose effect on satisfaction emerges only among frequent customers. Third, it extends the regional evidence base by covering the five gastronomic centers of Peru’s northern macroregion, including Cajamarca and Tumbes, with a sample and an estimation strategy that strengthen the external and statistical validity of the servicescape model in a developing market.

The managerial implication reverses the segmentation playbook: because the drivers of loyalty are structurally common, restaurants do not need parallel value propositions for occasional and frequent diners. Investment should follow the importance-performance ordering—first protect satisfaction, then ambient conditions (warm lighting, balanced aroma and noise, comfortable temperature) and a price perceived as fair, the levers that weigh most where new customers are concerned. Decor is the one calibrated exception: its satisfaction effect concentrates among frequent customers, so decor upgrades—ideally expressing the regional culinary identity—are best understood as retention spending. The exceptional satisfaction → loyalty coefficient also means that any operational failure eroding satisfaction translates almost one-to-one into lost loyalty, an argument for consistency over spectacle.

The results should be read with caution given the non-probabilistic convenience and chain sampling, which limits generalizability, and a sample skewed toward young adults. The cross-sectional design precludes claims of temporal causality. Discriminant validity is a substantive limitation: four pairs of constructs exceed the .90 HTMT benchmark and the servicescape dimensions correlate up to .94; although the six-factor model outperformed rival specifications, a second-order alternative fitted equivalently, and the structural conclusions were unchanged under that hierarchical specification, instrument refinement remains advisable in future work. The frequency split (up to once a month versus twice or more) may understate experience contrasts; panel designs following first-time customers, probability sampling, and models incorporating culinary quality, interpersonal service quality, and consumer emotions are natural extensions. Finally, 46 straight-lined questionnaires were retained in the main analysis; a sensitivity re-estimation without them left every conclusion unchanged.