2.1. Conceptualization of analytics, customer analytics and CAC

The concept referred to as CAC still needs to be addressed in organizational management. To clarify such a concept, it is necessary to resort to foundational concepts, such as customer analytics, and frame them within the organization’s capabilities. The taxonomy presented in Figure 1 illustrates the relationship between different denominations that can be found in the literature to refer to customer analytics, as well as the differentiation between customer analytics and CAC.

Source: Authors’ own elaboration.

Figure 1 Representation of the conceptual difference between customer analytics and CAC 

Figure 1 shows that the study of customer analytics lies at the intersection between analytics and the customer. Analytics can be understood as the application of processes and techniques that transform raw data into relevant information for decision-making (^{Wilder and Ozgur, 2015}). Analytics is a primary concern for decision-makers in the current business environment. Due to organizational complexity, decision-makers cannot rely on their limited knowledge and are turning to analytics to improve the results of their decisions (^{Rezaei et al., 2022}). Analytics has been highlighted by several authors as a capability that helps organizations have better competitive performance. Thus, ^{Gunasekaran et al. (2017)} find that analytics positively affects organizational performance, ^{O’Neill and Brabazon (2019)} found a significant relationship between higher levels of analytics skill and the ability to generate organizational value and competitive advantage, and ^{Dubey et al. (2018)} also found that analytics is effective in gaining competitive advantage.

Customer analytics is one of the subdimensions of marketing analytics, which is, at the same time, a subdimension of analytics (^{Hossain et al., 2020b}) and has recently been gaining importance due to the rise of innovative technologies and because of the pressure that exists today, more than ever, to satisfy customers by leveraging structured and unstructured data (^{Sun et al., 2014}).

As seen in Figure 1, although some authors refer to customer analytics with the literal term, it is possible to find in the literature that the terms “analytics” and “customer” have different labels or names referring to the same essence: analytics in customer contexts. Thus, for example, it is possible to find references to “customer intelligence” (^{Dinh et al., 2022}), “client analytics” (^{Ediger et al., 2014}), “consumer analytics” (^{Erevelles et al., 2016}), among others, referring to analytics within the specific domain of the customer.

Given the lack of conceptual maturity that still exists regarding customer analytics, different conceptions underlying the attempts to define analytics in customer contexts can be identified. Thus, there is a conception that refers to it as an action or a method. For example, ^{Nethravathi et al. (2020)} define it as the discipline that deals with recognizing customer behavior and predicting their purchasing patterns to improve business and environmental sustainability. In this sense, ^{Fernández (2019)} suggests that organizations can thus develop personalized products and services and even anticipate their needs, which enhances the user experience. Another conception refers to analytics in customer contexts as a process. For example, ^{Hallikainen et al. (2020)} define it as the acquisition, storage, processing, and analysis of an immense volume, variety, and velocity of customer-related data, with the aim of creating meaningful information for decision-making in the company and discovering insights in a timely manner. A third conception of analytics in the customer context refers to it as a strategic resource. ^{Louro et al. (2019)} point out that customer analytics is the final insertion of the concept of analytics into the disciplines of marketing and strategy.

Under these considerations, CAC can be understood as a set of organizational routines that create value for the organization, within the framework of processes and methods for exploring and exploiting customer data and their interactions with the organization.

2.2. CAC as a source of competitive advantage using the dynamic capabilities approach

^{Teece et al. (1997)} define dynamic capabilities as the ability to integrate, build, and reconfigure internal and external competencies to rapidly cope with environmental changes. ^{Pavlou and Sawy (2011)} propose that they are an essential driver of an organization’s competitive advantage, through the renewal of operational capabilities. Consistent with that conceptualization, ^{Louro et al. (2019)} define CACs as a dynamic capability present in organizations that continuously respond to trends in their markets and remain attentive to potential customer opportunities.

A market-oriented organization places importance on understanding both the expressed and unexpressed needs of its customers, as well as on building unique capabilities to satisfy their desires (^{Slater and Narver, 1999}). CAC, by acting on the operational routines of marketing and related areas (^{Hossain et al., 2020a}) to reconfigure them, contributes to achieving competitive advantages to the extent that it enables the organization to address all customer- management fronts in a rare, valuable, imperfectly imitable, and difficult-to-substitute manner, in accordance with the theory of resources and capabilities.

2.3. Psychometric properties for the development of measurement scales

According to the ^{Real Academia Española (s.f.)}, a construct is a theoretical framework used to comprehend a particular issue. CAC is conceptualized as a construct, serving as an underlying factor in a series of observed manifestations.

The traditional paradigm for the development of construct measurement scales includes: generation of items; definition of the evaluation scale; evaluation of content validity; selection of a sample; execution of exploratory factor analysis; evaluation of internal consistency; execution of confirmatory factor analysis; and the evaluation of convergent and discriminant validity (^{Swanson and Holton, 2005}). Following these steps leads to the development of measurement scales that meet the following psychometric properties:

Content validity: the degree to which a measure covers the content domain of what it is trying to measure (^{Yaghmale, 2003}; ^{Pérez-Rave, 2021a})

Construct validity: the degree to which the measurement of a given construct evaluates what it is supposed to measure. Thus, it is expected that the measures among themselves reflect convergence towards the underlying construct, but also that they help to discriminate with respect to the measures of other constructs. It includes convergent and discriminant validity (^{Pérez-Rave, 2021a}; ^{Ployhart and Schneider, 2012}; ^{Sackett et al., 2012}).

Reliability: has to do with the measurement error in the measurement process. In general, reliability is the tendency towards consistency of scores, which gives an idea of the precision of the instrument and its ability to generate stable scores (^{Martínez, Hernández and Hernández, 2006}; ^{Pérez-Rave, 2021a}a.

Equity: this is the ability of the instrument to evaluate individual differences impartially, independent of personal characteristics such as gender, age, position held, etc. (^{AERA et al., 1999}; ^{Pérez-Rave, 2021a}).

3.1. Formulation of the items

The generation of the items was based on the identification of the manifestations or uses of analytics in customer contexts, as found in the 42 studies subjected to the previously conducted systematic literature review (^{Maya-Restrepo, 2023}). From these manifestations, an adjustment process was conducted considering the inclusion, where appropriate, of different phases of the analytical process in which such manifestation could occur (e.g., collecting, analyzing, evaluating). That is, for example, if the use of analytics in a customer context was found as “identifying the best customers for the company,” it was transformed into “analyzing transactional data to identify the best customers for the company.”

Twenty-eight items were developed to represent CAC. An example of these is “How often does the organization where you work analyze data on the performance of competitors on customer-related issues?” The response format was a 5-point Likert scale: 1= Not carried out or carried out more frequently than once a year; 2 = At least once a year; 3 = At least once every six months; 4 = At least once every three months; 5 = At least once a month. In Table 1, the items associated with the reference studies from which they were extracted are observed. The items of the final scale after psychometric validation are presented in Appendices 1 and 2.

3.2. Content validity analysis

Following the above, the analysis of the content validity of the proposed items was conducted. The method used to carry it out was the expert panel, in which a group of experts in the area of interest (in this case, organizational administration) assessed the attributes of the items. ^{Egaña et al. (2014)} conducted a literature review of 40 studies to capture the methods used to carry out the analysis of content validity in the design of measurement scales. According to them, the expert panel is the most used method for evaluating this type of validity. The same authors cite references indicating that the appropriate number of experts can range from 7 to 30 and suggest that the most important aspect when using this method is to select experts with the appropriate profile. The present study involved 10 experts in organizational management, who evaluated the attribute of clarity of the items. Although the attribute of relevance is also often evaluated as part of content validity, it was not considered on this occasion, as the items were obtained from empirical evidence provided by the 42 studies previously reviewed.

For the evaluation of the attribute of clarity, a matrix-type format was developed, which included the formulated items (rows) and five possible ratings for the clarity of the items: very clear (5); clear (4); moderately clear (3); unclear (2); and very unclear (1). As mentioned, it was administered to 10 experts in the field of organizational management. Table 2, shows the descriptive statistics of the scores. Items rated 4 or 5 were immediately accepted and those rated 3 or less were subject to modifications to improve clarity.

According to Table 2, 11 out of the 28 items underwent some form of modification after evaluation by the experts, as they received a rating equal to or less than 3.

Table 3 shows the basic information of the experts who served as validators of the proposed instrument: gender; activity with which their current professional work is most related; highest academic degree; training undergraduate degree; and years of work experience. Table 3 corroborates the suitability of the experts to carry out the evaluation, due to their academic degree, field of training related to organizational administration, and years of professional experience.

Table 4 shows a sample of the items before and after content validation with the experts. The changes made corresponded to the suggestions and evaluation of the experts to improve the clarity of the item. As seen in Table 4, clarity adjustments primarily involved substituting those words that held some degree of ambiguity and précising the action that was intended to be talked about (e.g., determine, monitor, etc.).

3.3. Sample description

Between September and October 2022, the questionnaire was administered to professionals working in areas related to customer data analysis (e.g., marketing, customer-centric data analytics, consumer insight, market surveillance). Most of the professionals were contacted through LinkedIn, the largest professional social network in the world with more than 200 million members (^{Sumbaly et al., 2013}), and others through professional networks in Colombia. In total, approximately 520 requests were sent out, of which 101 completed questionnaires were obtained (response rate: 19%, boosted by the promise of receiving an executive report upon completion of the study). The respondents belong to 101 different companies and include individuals working in analytics (26.7%; analysts: 7.9%; supervisors or coordinators: 11.8%; director or manager: 6.9%), in marketing or related functions (72.2%; analysts: 10.8%; supervisors or coordinators: 27.7%; director or manager: 33.6%) and in general management (1%). Eighty-two percent of the organizations to which the respondents belong are engaged in commerce or service provision, while the remaining 18% are involved in industrial activities.

3.4. Analytical framework

The study is cross-sectional, employing a design for the development and validation of scales with psychometric rigor. The methodology used to refine and validate the scale follows the MinerConstructo framework (^{Pérez Rave, 2021b}), which contains seven stages, four of which are useful in the present study: observe ( a descriptive diagnosis of the variables is carried out using measures of central tendency, dispersion, and correlation); explore (the necessary conditions for conducting an exploratory factor analysis are assessed, potential underlying factors in the data are identified, an exploratory factor structure is provided, and the Cronbach’s Alpha reliability measure is calculated); confirm (a confirmatory factor analysis is performed in which the functional relationships between constructs and their factor loadings are identified, as well as the analysis of convergent and discriminant validity between factors); and communicate (the results are expressed through scientific article format). The other stages of the framework are not applied in this work (apply, explain, and predict), as covering structural relationship analysis and practical application (e.g., diagnoses) is beyond the scope of the study. This framework combines analytical capabilities (e.g., machine learning methods, automated tasks, among others) and psychometric analysis of latent variables, integrating resources from various R packages, and has been applied in other constructs, such as: dynamic capabilities of continuous improvement, (^{Pérez-Rave, Guerrero et al., 2022}); quality of health services (^{Pérez-Rave, Figueroa et al., 2022}) and dignified treatment for health workers (^{Pérez-Rave, González-Echavarría et al., 2022}).

4.1. Observe

Table 5 displays the descriptive statistics of the responses provided by the survey participants to the 28 items of the CAC scale.

As can be seen in Table 5, nearly all item scores range from 1 to 5, except for cac2, with mean values ranging between 2.901 (cac13) and 4.782 (cac1), and standard deviations ranging from 0.642 (cac1) to 1.577. (cac12). Ten of the 28 items had an average rating higher than 4, 17 of them between 3 and 4, and only one had an average rating lower than 3. The above shows that, although the survey managed to capture different positions among the respondents, there is a tendency to positively value customer analytics practices in organizations. Furthermore, all items have negative asymmetry values, which also indicates the tendency to adopt favorable positions regarding organizational practices related to customer analytics. In most cases (except for cac1 and cac5), kurtosis is between -2 and 2, suggesting that there are no extreme deviations from normality.

Table 6 contains the Pearson correlation coefficients for all pairs of items. The mean value of the correlation coefficients is 0.40, and the first and third quartiles are 0.34 and 0.504, respectively. The moderate correlation observed among most of the items suggests the presence of underlying patterns to be discovered in the following steps.

4.2. Explore

The result of the Bartlett test contrasts the hypothesis that the correlation matrix is the same (null hypothesis) or different (alternative hypothesis) from an identity matrix. The test yields a p value < 0.05 (0.000; chi.square = 1888.690, degrees of freedom = 378), so the null hypothesis is rejected. The Kaiser-Meyer-Olkin (KMO) test provides a measure of the proportion of global and individual variance that is possibly due to one or more latent factors. Values less than 0.5 are usually considered unacceptable, which is why cac1 is excluded. Table 7 details the KMO coefficients for each item.

As Table 7 shows, the overall suitability value is 0.872 with individual values ranging from 0.548 (cac2) to 0.943 (cac5). These results justify proceeding with exploratory factor analysis to uncover potential association patterns.

According to the Kaiser criterion (eigenvalues greater than 1) and Horn’s parallel analysis (corrected eigenvalues greater than 1), 6 and 1 factor(s) should be considered, respectively. Table 8 presents the criteria used to select the number of factors.

As Table 8 shows, based on Horn’s parallel analysis, a single-factor model is suggested (a single eigenvalue greater than 1). Likewise, under the Kaiser criterion it is suggested to have 6 factors (6 eigenvalues greater than 1).

The items from the single-factor model are presented in Table 9, in the supplemental material. The items and their respective factor loadings (correlation between the item scores and the respective factor) are presented there.

The one-factor model is composed of 26 items. Although the accumulated variance explained by this single factor is less than 50%, it will continue to be considered to verify the relevance of item refinement actions and compliance with the confirmatory criteria of psychometric quality.

The six-factor model contains two items with a factor loading of less than 0.45 (cac1 and cac2), so they are excluded. By excluding them, the model becomes one of five factors. The items and their respective factor loadings are presented in Table 10.

The five-factor model comprises 20 items in total and accounts for 63% of the total variance of the data.

4.3. Confirm

The results of the initial confirmatory factor analysis are presented in Table 11. Two structures were examined: one factor (M1) and five factors (M2).

With regard to Table 11, although neither of the two models meet the plausibility conditions in this initial test, a refinement is carried out considering items with confirmatory factor loadings below 0.7, as well as the analysis of modification indices. The plausibility conditions to be met are: RMSEA < 0.06, SRMR < 0.08, CFI and TLI > 0.95, Chisq/df < 2, AVE > 0.5, Fiab.Comp >= 0.7.

Following this refinement, Model M1 (one factor) remains with 14 items that satisfy the goodness-of-fit indicators, convergent validity, and composite reliability, while Model M2 (five factors) remains with 15 items but the criterion for discriminant validity between two of the factors is not met, despite satisfactory goodness-of-fit indices. Given this, starting from model M2, a third model (M2A) is generated, which includes the fusion of the two factors that did not pass the discriminant validity test. After a new refinement process like the one described, the model ends up with 10 items and achieves discriminant validity and fit.

Table 12 shows the comparative summary of the fit metrics, discriminant validity, and composite reliability of the three resulting models.

According to Table 12, models M1 and M2A satisfy the minimum criteria for model fit, validity, and reliability. Figures 2-3 show the two plausible models (M1 and M2A) to represent CAC. It should be noted that the factors labeled in M1 and M2A, despite being named in the same way (F1, F2, F3), represent different constructs after undergoing the aforementioned refinement process. The complete items of both scales (M1 and M2A) can be seen in Appendix 1 and 2.

Figure 2 displays model M1, unidi­mensionally, composed of 14 items. It is observed that all items exhibit high loadings close to or greater than 0.7. This CAC scale has an average variance extracted (AVE = 0.516) greater than 0.5 and a composite reliability greater than 0.7 (0.937).

Source: Generated by Minerconstructo.

Figure 2 Model M1 to represent CAC  

Figure 3 presents the M2A model, made up of three factors and 10 items. It is observed that all items exhibit high loadings (greater than 0.7). This CAC scale presents an average variance extracted (AVE = F1: 0.655; F2: 0.578; F3: 0.734) greater than 0.5 and a composite reliability (0.855) greater than 0.7. Furthermore, the overall Cronbach’s alpha for the final factors (F1, F2, and F3) was 0.866, 0.835, and 0.829, respectively.

Source: Generated by Minerconstructo.

Figure 3 M2A model to represent CAC 

The M1 model is useful for measuring CAC globally and parsimoniously, which is valuable when one wants to measure the most representative latent trait behind the existence of these capabilities in organizations, without delving into specific subtopics that compose them. The M2A Model is useful when the organization wants to explore in more depth the dimensions that constitute them and into the specific ways to materialize them. The interpretation of the tridimensional model is further elaborated below.

The M2A model suggests that CACs in organizations have three underlying dimensions (factors):

Factor 1 (F1): Customer maintenance analytics capability. This is manifested through the practices of evaluating the customer experience across all touchpoints through data analysis, assessing the performance of loyalty strategies derived from data analysis, and designing cross-selling strategies based on customer consumption patterns. These manifestations can be interpreted as a set of routines that occur after the sale, when the customer has already established a commercial relationship with the organization and it aims to evaluate their purchasing experience, retain them for future returns, and attempt to sell complementary products or services to those already acquired. This front aims to generate outcomes that the customer will eventually perceive directly.

Factor 2 (F2): Customer acquisition analytics capability. This is manifested through practices of evaluating different marketing mixes considering customer data, monitoring the impact of marketing campaigns based on data, evaluating the performance of new products or services in the market, reconfiguring the value offer based on customer data analysis results and defining commercial strategies derived from customer-related data analysis. These manifestations can be interpreted as a set of routines that the organization develops prior to the sale, when it is attempting to capture customers with new value offers, new commercial strategies, campaigns with appropriate mixes of pricing, promotions, placements, and products.

Factor 3 (F3): Customer economic evaluation analytics capability. This is manifested through the identification of the organization’s best customers based on commercial transactions and through the evaluation of customer retention levels through data analysis. These manifestations can be interpreted as the organizational routines conducted to measure the economic contribution of customers and their value to the organization. Unlike the first front, this one does not seek to generate results that can be perceived directly by customers, but rather by the organization, in order to adjust its strategy in pursuit of profitability.

Next, confirmatory factor analysis is conducted on the models, incorporating control variables to verify the equity of the scale. In this case, three control factors are considered: the position of the person answering the survey; the area to which the person answering the survey belongs (marketing or analytics); and infrastructure for analytics. The variable “position” has three response levels: marketing/analytics analyst; marketing/analytics head or coordinator; and marketing/analytics director or manager. The variable “area” has two levels of response: marketing; and analytics. The variable “infrastructure for analytics” is a binary variable that summarizes the rating given by the respondents on the level of maturity that the organization has for carrying out analytics activities at the level of human effort (manual, semi-automatic, and automatic), tools used (spreadsheet, licensed software and free software), processing capacity (individual computer, computer network and cloud), and analysis methods used (statistical description, association between pairs of variables, association between more than two variables or multivariate).

Tables 13 and 14 present the confidence intervals based on bootstrapping (resampling technique) with 2000 replications with the aforementioned control variables, to verify the equity of the scales.

The absence of significance among the control variables for any of the factors, in both scales, as depicted in Tables 13 and 14, implies that organizational factors are not prioritized by the scales. In other words, the proposed measurement models for CAC enable robust scores of CAC to be generated regardless of (a) factors such as the area or position of the individuals responding to the instrument, and (b) the maturity levels of the analytical infrastructure available to the organization of the respondent.