What Is ACSI? The Customer Satisfaction Measurement Model
What Is the ACSI Model
ACSI stands for the American Customer Satisfaction Index, a model for measuring customer satisfaction. The model was developed to measure satisfaction at the national and organizational levels, and it later became a foundation for research on services, products, and customer behavior.
When you use ACSI in your thesis, you usually use it to explain why customers are satisfied or dissatisfied, then examine how satisfaction relates to complaints and loyalty. The main variable groups usually include customer expectations, perceived quality, perceived value, satisfaction, complaints, and loyalty.
ACSI fits studies in which customer satisfaction is the dependent variable, such as satisfaction with a bank, e-commerce platform, telecommunications provider, hospital, or educational service. You need to adapt the wording of the items to your research context instead of copying questions from another study word for word.
If your topic focuses on service quality through five specific dimensions, you can also compare the SERVQUAL model, the SERVQUAL scale, or the SERVPERF model. ACSI and SERVQUAL can be used in the same study, but they answer different research questions.
Components of the Model
In the ACSI model, the constructs are arranged as a chain of causes and outcomes. Expectations, perceived quality, and perceived value affect satisfaction. Satisfaction may be related to complaints and loyalty. You need to define the specific relationships again based on your industry and research model.
| Component | Meaning in the model | Common role in a thesis |
|---|---|---|
| Customer expectations | The level of performance customers predict or expect before using a product or service | Independent variable or antecedent variable |
| Perceived quality | Customers' evaluation of actual quality after the experience | Independent variable |
| Perceived value | Customers' evaluation of the balance between benefits received and costs paid | Independent variable |
| Customer satisfaction | Customers' overall evaluation after use | Mediator or dependent variable |
| Customer complaints | Customers' response when the experience does not meet expectations | Outcome variable |
| Loyalty | The intention to continue using, return to, or recommend the product or service to others | Dependent variable |
You should distinguish perceived quality from satisfaction. A customer may rate delivery quality highly but still feel dissatisfied because the price is high. Conversely, a customer may be satisfied with the overall experience even though one aspect of quality is not particularly strong.
Perceived value is also more than the selling price. It reflects how customers weigh benefits, quality, convenience, and cost. If you only ask whether the price is cheap, your scale may be measuring price perception rather than the full concept of perceived value.
When you build the model, treat each construct as a latent variable measured by multiple items. For example, satisfaction can be measured through overall evaluation, the extent to which expectations are met, and the extent to which the decision to use the service was appropriate. These items need to be written as a clear Likert questionnaire that respondents can answer easily and that fits your sample.
The Original Model and Common Extensions
The original ACSI model focuses on the chain from expectations, perceived quality, and perceived value to satisfaction. Satisfaction is then linked to complaints and loyalty. When you apply the model to a specific industry, you may reduce or extend it to fit your research question.
One common extension is to add factors that explain perceived quality. For example, in banking services, perceived quality may be formed by reliability, responsiveness, assurance, empathy, and tangibles. This connects with the five-gap service quality model, but you need to state clearly which model is the theoretical base and which variable groups have been added.
Another option is to keep satisfaction as a mediator between perceived quality and loyalty. In that case, your research question is not only whether quality affects loyalty, but also whether that effect is transmitted through satisfaction. If you test mediation, define the direct and indirect paths in advance.
You can also extend the model by adding variables that fit the context, such as trust, perceived risk, brand image, or switching costs. Each added variable needs a theoretical and practical reason. The more variables your model contains, the longer the questionnaire becomes, and the greater the sample-size and interpretation requirements are.
ACSI does not have to be copied in its original form for every topic. In service-quality research, you can compare ACSI with SERVQUAL or SERVPERF to choose a suitable structure. The SERVPERF model usually focuses on perceived service performance, while ACSI emphasizes the relationships among expectations, quality, value, satisfaction, and behavioral outcomes.
A common problem is mixing every variable from several familiar models into one diagram. Each arrow needs to answer three questions: what argument supports this relationship, which scale measures the variables, and can your data test the relationship?
Scales Commonly Used for Each Construct
ACSI is a model framework, not one fixed questionnaire that can be used identically in every industry. You need to find scales that fit your context, translate and adapt the wording, and then test them again on your own data. Scales for service quality can draw on SERVQUAL by Parasuraman et al., while ACSI provides the logic connecting perceived quality, value, and satisfaction.
| Construct | What should be measured | Source or reference direction |
|---|---|---|
| Customer expectations | Expectations before use and what customers predict about quality and benefits | ACSI structure, adapted to the research context |
| Perceived quality | Overall evaluation of quality after the experience | The service-quality dimensions in (Parasuraman et al., 1988) can be consulted |
| Perceived value | The balance among benefits, quality, and cost | ACSI structure, worded for the specific product or service |
| Satisfaction | Overall evaluation, the extent to which expectations are met, and whether the choice was appropriate | ACSI structure, with reliability tested on the actual sample |
| Complaints | The likelihood of making a complaint and the response when the experience falls short | ACSI structure, used only when the context provides sufficient variation |
| Loyalty | The intention to continue using, repurchase, and recommend | ACSI structure, potentially combined with arguments about customer behavior |
When you write items, avoid two common errors. The first is putting two ideas into one item, such as “the service is fast and the employees are friendly.” A respondent may evaluate these two parts differently. The second is using an overly broad phrase such as “the service is very good” without saying which aspect is good.
You should use the same scale for most items, such as a five-point or seven-point Likert scale, and state clearly what the low and high points mean. The Likert scale originated from a technique for measuring attitudes described in (Likert, 1932). After collecting data, you can run Cronbach's Alpha, EFA, or a PLS-SEM measurement model depending on your research design.
If you use SPSS, the process usually begins with Cronbach's Alpha and Corrected Item-Total Correlation, followed by EFA if your model and method require it. Run Analyze > Scale > Reliability Analysis to assess reliability, then use Analyze > Dimension Reduction > Factor if EFA is needed. Cronbach's Alpha from 0.7 upward is generally considered acceptable (Nunnally, 1978), while Corrected Item-Total Correlation from 0.3 upward is used as a reference point (Nunnally and Bernstein, 1994). For a new or exploratory scale, Alpha from 0.6 may be considered according to (Hair et al., 2010).
Applying the Model to Your Topic
Suppose you are studying students' satisfaction with an online learning platform. You could set perceived quality, perceived value, and expectations as independent variables, satisfaction as a mediator, and loyalty as the dependent variable. This is an illustrative model, not the result of an actual study.
The conceptual model can be stated as follows: customer expectations, perceived quality, and perceived value affect satisfaction, and satisfaction affects loyalty. If there is a suitable basis, you can also test the direct effects of perceived quality and perceived value on loyalty.
Four illustrative hypotheses are:
- H1: Customer expectations have a positive effect on students' satisfaction with the online learning platform.
- H2: Perceived quality has a positive effect on students' satisfaction with the online learning platform.
- H3: Perceived value has a positive effect on students' satisfaction with the online learning platform.
- H4: Satisfaction has a positive effect on students' loyalty to the online learning platform.
Before distributing the questionnaire, replace “online learning platform” with the specific object of study and make sure respondents have enough experience to evaluate it. For example, someone who only registered an account but has not attended a class will have difficulty evaluating perceived quality. The screening question should appear before the main scale section.
Sample size also needs to be considered based on the number of items and the analysis method. The rule of 5 to 10 observations per item is stated in (Hair et al., 2010). If you use regression with m independent variables, the reference formula of n at or above 50 + 8m is presented in (Tabachnick and Fidell, 2013). This supports the initial design and does not replace checking data quality after collection.
How to write this in your thesis
You can adapt the following paragraph for the theoretical foundation chapter: “Based on the ACSI model, this study proposes that [customer expectations], [perceived quality], and [perceived value] affect [satisfaction]. At the same time, [satisfaction] is assumed to affect [loyalty] among [survey respondents] toward [the product or service].” Then present the operational definitions, scale sources, and reason for adapting each construct.
Testing the Model with SPSS or SmartPLS
SPSS is suitable when your model uses composite scales, linear regression, mediation testing with PROCESS, or EFA. You can run Analyze > Scale > Reliability Analysis to assess reliability, then use Analyze > Dimension Reduction > Factor if EFA is needed. For regression models, the usual path is Analyze > Regression > Linear.
When reading the Coefficients table, focus on Standardized Coefficients Beta, t, and Sig.. The Beta coefficient shows the direction and strength of the relationship in the regression model, while Sig. is commonly used to assess statistical evidence for each path at the significance level you selected. Report the coefficient together with the standard error or confidence interval when the procedure provides it, rather than writing only that a hypothesis was accepted.
When the model contains latent variables and multiple items, SmartPLS is more suitable if you have clearly specified the measurement model and structural model. In SmartPLS 4, create a project, import the .csv file, assign items to constructs, connect the constructs according to the hypotheses, and run Calculate > PLS-SEM Algorithm. To test the significance of the paths, run Calculate > Bootstrapping and select an appropriate number of subsamples. The current PLS-SEM procedure commonly uses 5,000 bootstrap subsamples according to (Hair et al., 2022).
Read Outer Loadings, Reliability and Validity, Discriminant Validity, Path Coefficients, R Square, f Square, and Collinearity Statistics. An outer loading from 0.7 upward is a commonly used benchmark in PLS-SEM according to (Chin, 1998) or (Hair et al., 2022). CR from 0.7 and AVE from 0.5 upward are referenced in (Fornell and Larcker, 1981). HTMT below 0.85, or below 0.90 for closely related constructs, is stated in (Henseler et al., 2015).
| Results table | Column to read | Illustrative output | Interpretation |
|---|---|---|---|
| Path Coefficients | Original Sample, T Statistics, P Values | H2: β = 0.42, t = 5.18, p = 0.000 | Perceived quality has a positive relationship with satisfaction in the example |
| R Square | R Square | Satisfaction: 0.48 | The explanatory variables in the illustrative model explain 48% of the variance in satisfaction |
| Outer Loadings | Original Sample | CL1: 0.78 | The illustrative item has a loading above 0.7 |
| Reliability and Validity | Composite Reliability, AVE | CR = 0.86, AVE = 0.61 | The illustrative scale exceeds the corresponding reference benchmarks |
| Discriminant Validity | HTMT | CL and HL: 0.79 | The illustrative construct pair is below the 0.85 benchmark |
The numbers in the table above are illustrative output, not the results of any study. For R², 0.75, 0.50, and 0.25 are interpreted as substantial, moderate, and weak, respectively, in the PLS-SEM literature of (Hair et al., 2011). Do not use the CFI, TLI, or RMSEA benchmarks for CB-SEM to evaluate a SmartPLS model. Those indicators belong to the CB-SEM fit-index group, with benchmarks discussed in (Hu and Bentler, 1999).
If you test satisfaction as a mediator, report the indirect effect and the bootstrap confidence interval. An indirect effect is considered significant when the bootstrap confidence interval does not contain 0 according to (Preacher and Hayes, 2008). Do not conclude that mediation exists simply because both separate regressions have small p-values.
Common Errors When Using This Model
The first error is calling every customer satisfaction study an ACSI study. If your model only measures five service-quality dimensions and predicts satisfaction, describe the structure you are actually using. ACSI may be the theoretical foundation, but the specific variables may have been changed substantially.
The second error is taking a scale from another industry without explaining the adaptation. Items about digital banking, hospitals, and restaurants may all refer to quality, but the experience and interpretation of the questions are different. Record the original version, translation, adapted version, and reason for retaining or removing each item.
The third error is adding too many arrows to make the model look complete. Each hypothesis needs an argument, measurable variables, and a relationship that can be tested with data. If the questionnaire measures respondents only once at one point in time, use strong causal language carefully.
The fourth error is removing an item simply because one indicator is below the benchmark without examining its theoretical content. When Cronbach's Alpha increases after deleting an item, you still need to check whether the item is conceptually necessary. Re-running the analysis until the output looks attractive without recording each processing round will make the results difficult to defend when you are questioned.
The final error is mixing SPSS and SmartPLS without explaining the role of each software package. SPSS can be used for data cleaning, sample description, Cronbach's Alpha, and EFA. SmartPLS can be used for the measurement model and structural model. Present the analysis sequence consistently with the method you selected.
Frequently asked questions
What is ACSI, and is it a fixed scale?
ACSI is a customer satisfaction measurement model that commonly connects expectations, perceived quality, perceived value, satisfaction, complaints, and loyalty. You should not treat ACSI as one fixed questionnaire that can be copied unchanged for every industry.
Choose constructs that fit your research question, find the scale sources, adapt the context, and test the scales again on your sample.
Can the ACSI model be used for a service-quality topic?
Yes. You can use ACSI to explain how perceived quality relates to satisfaction and loyalty. If you need to measure service quality through specific dimensions, you can combine or compare it with SERVQUAL, SERVPERF, and scales suitable for the industry.
In the methods chapter, state clearly whether ACSI is the foundation model or the complete research structure. This prevents readers from assuming that every item in the questionnaire is an original ACSI component.
Should ACSI be run in SPSS or SmartPLS?
SPSS is suitable if you use regression, EFA, mediation testing, or analyses based on composite variables. SmartPLS is suitable when you want to assess the measurement model and structural model simultaneously with latent variables and multiple items.
The software does not choose the model for you. Choose based on the research question, scale type, sample size, analytical assumptions, and your supervisor's requirements.
Does the ACSI model require Cronbach's Alpha testing?
If each construct is measured by multiple items, you need to assess the reliability and validity of the scale. In an SPSS procedure, Cronbach's Alpha and Corrected Item-Total Correlation are usually reviewed before EFA or regression. In PLS-SEM, you also need to examine outer loading, CR, AVE, and discriminant validity.
Do not retain or remove an item based on one number alone. The item content, scale source, and fit with the context also need to be explained.
Can I remove complaints and still call it an ACSI model?
You can remove complaints if the research question or context does not generate enough data for this construct. However, describe the result as an adapted version of the ACSI model and explain why the construct was removed.
A reduced model with a clear rationale is usually easier to defend than a complete model with an overly long questionnaire or a complaints construct that does not fit the respondents.
Can ACSI be used for a study of loyalty?
Yes, if you argue that satisfaction, perceived quality, or perceived value relates to loyalty. Define loyalty with suitable items, such as the intention to continue using, repurchase, or recommend the product or service.
If you test satisfaction as a mediator, define the direct and indirect paths before running the model. A mediation conclusion should be based on the indirect effect and bootstrap confidence interval, not only on one individual p-value.
Open your data file, redraw the ACSI diagram for your specific research object, list the items, and mark the source of each scale before running the analysis. If you need to run the model on your own .sav or .csv file, M4 analysis by DoThesis carries out this step.