1. Introduction
In the last 20 years, in the era of new technologies and modernization, Internet use has increased significantly, especially for the purposes of communication, marketing, and electronic commerce (ecommerce) [
1,
2]. Business of marketing has always been fluid, continuously adapting to everevolving consumer preferences. The migration from traditional to electronic commerce was going on long before the events caused by the COVID19 pandemic, which has only hastened this transition. Indeed, the need for commerce digitization rapidly graduated for many businesses into means for survival once the pandemic hit. Within a matter of weeks to months, brands without online options were hastily implementing new ecommerce platforms, and those already using digital infrastructures were bracing their servers’ capacities for the impact of increased online traffic. Ecommerce is defined as a type of Internet use mainly to carry out business transactions in which parties communicate electronically instead of in person. These transactions significantly reduce costs, save time, increase profits, and simplify business activities, involving manufacturers, consumers, and service providers that use the Internet [
3,
4]. There is a clear expectation from consumers that companies should do their part to help them in their daily lives and to keep them informed. Brands must be able to meet consumers where they are and offer personalized services for their specific needs. According to Shaw [
5], considering the nature of transactions, there are five major categories of ecommerce: Business to Business (B2B), where ecommerce is done exclusively between companies; Business to Customer (B2C), in which company offers services to consumers; Business to Government (B2G), where companies offer government agencies products and services through online marketing and bidding for projects; Consumer to Business (C2B), where companies tender for projects posted by consumers; and Consumer to Consumer (C2C), where consumers sell their products to consumers online.
Knowing that B2C websites, where consumers directly buy products, present the lifeblood of B2C ecommerce, companies strive to design a successful B2C website and to ultimately make business considerably practical and effective. Amazon and Alibaba, followed by eBay, Walmart, Priceline, and Rakuten, are the most dominant and significant B2C ecommerce companies [
6]. The popularity and expeditious advancement of B2C ecommerce make these sorts of transactions the leading retailing channel for ordinary customers [
7], and therefore Internetbased commerce in general raises the question of awareness and vulnerability of consumers’ privacy and security on B2C platforms [
8,
9,
10]. Security and privacy of information provided by customers are very important, especially in risky and unpredictable ambiance [
11]. Factors such as transaction confidentiality, integrity, and authentication imply trust at the technology level. For the continual performance of B2C online commerce, customer relationship management plays an important role [
12] and trust becomes an inevitable factor [
13,
14].
In recent years, multicriteria decisionmaking (MCDM) has been applied in various fields of scientific research in cases where it is desirable to restructure a multicriteria problem. At the end of this process, the most optimal choice, or an alternative one, is selected. A formal framework for modeling multidimensional decisionmaking problems is therefore provided by applying MCDM, especially for problems that require systems analysis, the analysis of decision complexity, the relevance of consequences, and the need for the accountability of decisions made [
15]. Utilizing Fuzzy MCDM (FMCDM), an efficient approach for evaluating multiple criteria, can be achieved to support managers, experts, and other decision makers with the goal of balancing and measuring different factors, simplifying and clarifying decisions [
16].
In this paper, we study the factors for successful ecommerce platform design using Fuzzy Analytical Hierarchy Process (FAHP) based on triangular fuzzy numbers [
17]. In many realworld situations, when applying decisionmaking approaches, human judgment alone is often insufficient and not reliable. Therefore, the use of triangular fuzzy numbers presents a viable alternative for expert judgment regarding the qualitative factors and their importance. Similarly, trapezoidal, Pythagorean, znumbers, and the recently introduced Spherical fuzzy numbers [
18] may also be considered when applying the FAHP method. Although our research is based solely on triangular fuzzy numbers, we present an extension to the current model of optimism indexes. Firstly, we conduct a literature overview, and afterwards we select a number of factors and subfactors for prioritization, taking into account factors during the COVID19 pandemic. As a starting point, we include FAHP with three points of view and further extend to a novel, fivepointsofview ranking of subcriteria.
The advances in this paper are summarized in the following:
New subcriteria influencing ecommerce websites are introduced.
The FAHP method is extended by introducing two new points of view for the decisionmaker, namely, semipessimistic and semioptimistic views, with corresponding optimism indexes $\lambda =0.25$ and $\lambda =0.75$, respectively.
The estimation and analysis of ranking similarities in the extended model is conducted and discussed.
As of writing this paper, the authors have not found any article or study regarding ecommerce platform design using FAHP in the region of the Western Balkans. Therefore, our main goal is to provide insights to the decisionmaking process and further extend one of the wellknown MCDM methods. According to this goal, we formulated four research questions (RQs).
RQ1: Can the results presented in our paper help ecommerce companies of the Western Balkans region?
RQ2: Does a highly influencing subfactor during the COVID19 pandemic exist?
RQ3: Are there significant changes in the subfactors ranking when the three values of an optimism index in the FAHP method are expanded to the finite or countable set of values?
RQ4: Do we have complete insight into the interrelations of subcriteria using Extended FAHP?
The rest of the paper is organized as follows.
Section 2 presents the criteria for evaluation of B2C websites, divided into factors and subfactors.
Section 3 deals with methodology used, namely, Fuzzy AHP. Finally,
Section 4 gives the results, while the concluding remarks are given in
Section 5.
3. Methodology
The fuzzy set theory has been known for over half a century, ever since its proposal by Zadeh in 1965. Even then, it has been employed as guidance for fuzzy decisionmaking problems. Their original inception was intended for linguistics, and it has enabled uncertainty and imprecision to be represented, and, more importantly, constructed in a deterministic manner [
59,
60]. Sets defined in such a manner could therefore be identified as a generalization of the wellknown set theory, enabling a decisionmaker to include incomplete or partially unknown information in the decision model [
61].
Whereas in the classic set theory, an element can either belong to a set or not belong at all, in fuzzy sets, the membership of an element can be described by a number from the interval $[0,1]$. Each element of this set can hence be mapped on this interval with a membership function (MF), denoted by $\mu $. In addition, a fuzzy set can have an infinite number of different MFs.
Let all fuzzy sets defined on the set of real numbers
$\mathbb{R}$ be represented as
$F\left(\mathbb{R}\right)$. The number
$A\in F\left(\mathbb{R}\right)$ is a fuzzy number if there exists
${x}_{0}\in \mathbb{R}$ so condition
${\mu}_{A}\left({x}_{0}\right)=1$ holds, and
${A}_{\lambda}=\left[x,{\mu}_{{A}_{\lambda}}\left(x\right)\ge \lambda \right]$ is a closed interval for every
$\lambda \in [0,1]$(see [
17,
62]). The membership function, a component of a triangular fuzzy number (TFN)
A, is a function
${\mu}_{A}:\mathbb{R}\to [0,1]$, defined as
where inequality
$l\le m\le u$ holds. Variables
l,
m, and
u are the lower, middle, and upper value, respectively, and when
$l=m=u$, TFN becomes a crisp number. In the sequel, the triangular fuzzy number will be denoted by
$\tilde{A}=(l,m,u)$.
Assume two TFNs,
${\tilde{A}}_{1}=({l}_{1},{m}_{1},{u}_{1})$,
${\tilde{A}}_{2}=({l}_{2},{m}_{2},{u}_{2})$, and scalar
$k>0$,
$k\in \mathbb{R}$. The arithmetic operation properties are defined as [
63,
64,
65]:
Left and right side of the membership function of triangular number
$\tilde{A}=(l,m,u)$, as shown in
Figure 2, are denoted by
${\mu}_{\tilde{A}}^{l}={\displaystyle \frac{xl}{ml}}$ and
${\mu}_{\tilde{A}}^{r}={\displaystyle \frac{ux}{um}}$, and their matching inverse functions are
Left and right integral values of the triangular fuzzy number
$\tilde{A}$, according to [
66], are defined as
and
and the total integral value, according to [
66] as a combination of left and right integral values, is
where
$\lambda $ represents an optimism index. The pessimistic, semipessimistic, balanced, semioptimistic, and optimistic points of view of the decisionmaker are, respectively, expressed by the values 0,
$0.25$,
$0.5$,
$0.75$, and 1.
Fuzzy AHP
Since its creation [
67], the AHP had a respectable application in MCDM, enabling the decision makers to solve complex problems by decomposing them into a hierarchical structure, creating the comparison matrix and determining the importance of one indicator above others. The specified level of uncertainty of a team of experts (or even one expert) [
68] due to the inability to express the significance of some criteria has led to the introduction of FAHP [
69,
70] enabling conversion of linguistic statements into mathematical expressions.
The summarized steps in FAHP are as follows [
17,
71]:
Step 1. Establishing the main goal and hierarchical appearance of criteria. In general, the hierarchical structure has been organized vertically: the main goal is, as the most important component, at the top; the criteria that contribute to the goal are at the intermediate levels; and the subcriteria are at the lowest level.
Step 2. Determining the pairwise comparison matrix
$\tilde{D}$ in terms of TFNs. In this step, a positive fuzzy reciprocal comparison matrix
$\tilde{D}={\left({\tilde{d}}_{ij}\right)}_{n\times n}$ with a total of
$\left(\genfrac{}{}{0pt}{}{n}{2}\right)$ comparisons of elements from a higher level with elements from a lower level is developed. The fuzzy value
${\tilde{d}}_{ij}$ represents the degree of relative importance between criteria;
$i=j$,
${\tilde{d}}_{ij}=(1,1,1)$, and
${\tilde{d}}_{ij}=1/{\tilde{d}}_{ji}$, otherwise.
Table 1 shows the fuzzy scale for constructing pairwise comparisons.
As it was recommended in [
72], a fuzzy distance of 2 and odd values as boundaries for all nonintermediate values are applied in order to achieve better consistency. There are also different scales of triangular fuzzy numbers applicable in the previous case [
73,
74,
75].
The graphic representation of the used FAHP scale with all three values (lower, median, and upper) is presented in
Figure 3.
Step 3. Matrix consistency review.
For a matrix
$D={\left({d}_{ij}\right)}_{n\times n}$, the consistency index
$CI$ and consistency ratio
$CR$ are calculated using eqs. from [
76]:
where
${\lambda}_{max}$ corresponds to a maximal eigenvalue of matrices
D and
$RI$ is a random index, as shown in
Table 2.
The value
$CR<0.1$ confirms the comparison matrix consistency, while otherwise the reason for inconsistency should be found and calculations repeated [
77].
Step 4. The fuzzification process.
Using the triangular fuzzy numbers from the comparison matrix
$\tilde{D}={\left({\tilde{d}}_{ij}\right)}_{n\times n}$, applying
and
the value of the fuzzy synthetic extent is obtained as follows [
64]:
Step 5. The defuzzification process.
Next, in this step, using
the total integral value for the TFNs
${\tilde{S}}_{i}$ is calculated [
78].
Step 6. Normalization of weight vector w and obtaining the vector for each criterion.
Using
the weights for all criteria are obtained.
Step 7. Ranking the weights for all subcriteria.
The weights for each subcriterion are obtained by multiplying the weights of the criteria and subcriteria. Then, arranging the obtained weights, the subcriteria ranking is received.
These steps can also be presented in algorithm form [
79], shown below in Algorithm 1.
Algorithm 1 Steps in the FAHP process. 
 1:
Establish the main goal  2:
Identify $Xi,Xij$ ▹ Criteria and subcriteria  3:
Construct $\mathbf{D}$ ▹ Fuzzy correlation matrix  4:
Calculate $CR$  5:
if$CR\ge 0.1$then  6:
Adjust values  7:
go to 3  8:
else  9:
Fuzzification, calculate $\tilde{S}{}_{i}$  10:
Defuzzification, calculate ${w}_{i}$  11:
Calculate ${w}_{i}^{*}$ ▹ Normalization vector  12:
$Xij$ ranking  13:
end if

One of the main general drawbacks of the AHP methods (FAHP included) is the existence of incomparable criteria. This shortcoming may be overcome using the networklike presented ANP, where all the criteria, subcriteria, and alternatives are presented as nodes, grouped in clusters, enabling them to be compared to each other as long as an interrelation exists there. In this paper, we have chosen the FAHP method only, due to the fact that it enables the expert to decompose a complex problem into a few simplified steps. We have, however, extended the model to include five points of view instead of the usual three. The decision maker can hence easily express their opinion using descriptive grades, and these linguistic values can be further explained with a mathematical approach.
4. Results and Discussion
We firstly discuss the main criteria ranking, both for AHP and FAHP, with three points of view (pessimistic, balanced, and optimistic). Afterwards, we rank individual subcriteria. Finally, we conduct the ranking of all nineteen subcriteria using the extended FAHP, with semipessimistic and semioptimistic points of view, and test our ranking using the Spearman rank correlation coefficient [
80].
In the FAHP process, we have firstly calculated a fuzzy comparison matrix and weights for the main five criteria, as shown in
Table 3, and since
$CR=0.008117<0.1$, the matrix is consistent.
In addition, the ranking of main criteria for both AHP and pessimistic, balanced, and optimistic FAHP points of view (with corresponding
$\lambda =0$,
$\lambda =0.5$ and
$\lambda =1$, respectively) is presented in
Figure 4. In both AHP and all three FAHP cases, criteria
$X3$Service Quality ranked highest, while criteria
$X5$ (customer support) ranked lowest. This is somewhat expected, as Quality of Service, and its superset Quality of Experience, are increasing factors in the Internet presence of B2C websites. This corresponds to the finding of [
42] for the pandemic shopping trends. In AHP ranking, our results show that
$X5$ and
$X2$ have the same rank, while in all three cases of FAHP, no two criteria are ranked the same. Using FAHP, a decisionmaker can finetune their actions to increase an aspect of their B2C website.
Comparing points of view for main criteria, we can observe that a optimistic point of view (i.e., $\lambda =1$) does not always yield a higher rank when compared to the pessimistic view. For instance, for $X3$ and $X5$, the highest and lowest criteria, the pessimistic point of view ranked higher when compared to the corresponding optimistic view.
The ranking of subcriteria is firstly conducted in the same manner as the ranking of the main criteria, and fuzzy comparison matrices are given in
Table 4,
Table 5,
Table 6,
Table 7 and
Table 8. The results show that all matrices are consistent. Similar to the main criteria comparison, the AHP method yields equal ranking in some cases, namely, in
$X2$Information. Out of five subcriteria, we have two pairs of equal rankings, in the highest and second highest rank. This can also be observed in the triangular fuzzy numbers for
$X2$ in
Figure 5. Subcriteria
$X21$ and
$X22$, corresponding to Comparison and Search functions, respectively, have equal ranking. These two subcriteria both require customer input, and a B2C website with a better userfriendly design would most certainly aid in these functions. The second highest ranking pair,
$X23$ and
$X25$, referring to contact information and multilingual options, respectively, do not require user input like the previous case but rather serve to increase customers’ trust in the company. Both groups are very important in the postCOVID world as etrust, as stated in the introductory section, is an everincreasing factor for B2C websites.
Similar to the main criteria case, FAHP distinguishes between subcriteria; their rankings are unique, regardless of point of view.
Finally, we extend the process of FAHP by adding the semipessimistic (
$\lambda =0.25$) and semioptimistic (
$\lambda =0.75$) points of view. The final ranking subcriteria is given in
Figure 6. We observe that while the subcriteria from
$X3$, namely
$X31$, still ranked highest; however, the remaining subcriteria from Service Quality,
$X32$,
$X34$, and
$X33$ ranked third, fifth, and eleventh, respectively. Furthermore, the highest ranked subcriteria from customer support, namely,
$X51$, corresponding to item tracking, ranked seventh overall, is
$2.6$ times higher than
$X52$ and
$4.5$ times higher for the AHP case and about
$2.301$ and
$4.395$ times higher in the FAHP case (averaged over all points of view). The high overall ranking of tracking and tracing reflects the shift in online transactions, which includes item delivery during the pandemic and, more importantly, in the postCOVID world.
We have applied different solving techniques in this paper, which can, in general, lead to inconsistencies or disagreement. For the purpose of estimation and analysis of ranking similarities applying the AHP and the Extended FAHP to all subcriteria influencing ecommerce platforms, we have conducted fifteen different rankings using the Spearman rank correlation coefficient: [
80]:
where
n is the number of elements in the ranking and
${R}_{{x}_{k}}$ and
${R}_{{y}_{k}}$ represent the ranks of the
kth element in the compared rankings. By applying Equation (
17), all compared results are presented in
Figure 7, and since
$min\left\{{r}_{s}\right\}=0.964912$, it can be concluded that all rankings have high similarity [
81].
5. Conclusions
Due to the ongoing pandemic, the previous two years have made a significant impact in all aspects of life—everyday activities, education, healthcare, security, economy, and trade, and have hence acclimated people to a new form of reality. Certain situations, such as lockdowns, paved the way for the evergrowing online presence. The struggle of small and medium enterprises to compete on the market therefore heavily relied on their shift towards online commerce. Reports that show multiple increments of online sales are an indicator that the trend of online commerce will continue to exist and grow in the postCOVID world.
The paper has investigated the problem of ecommerce management and the influence of various different factors. Indicators associated with digital platforms have been divided into five groups, involving security, information, quality, design, and support aspects. Using AHP and Extended FAHP, nineteen subcriteria were ranked to determine the preferred ones in the process of ecommerce evaluation. The obtained results indicated that the proposed methods are entirely capable of estimating the influence of factors and subfactors on online trade commerce. Considering the obtained weights for each subcriteria and all five values of $\lambda $ in the FAHP case, factors such as trust and loyalty, safe payment, exchange or return, and account security have the most significant effect on successful ecommerce platform design, while the FAQs, multilingual option, and contact information deemed the least significant. In the AHP case, the same subcriteria from the service quality and security, privacy, and authority are ranked highest, while those from the information sector are of least importance.
Although our proposed method gives insight into several advantages and potentials in the field of B2C ecommerce, there are limitations to this work. However, these limitations might lead to future possibilities and steps forwards in our future research. Our use of AHP, because of its topdown direction structure and comparisons of criteria from one level with all criteria from the upper level, can lead to incomparable subcriteria. This challenge can be overcome utilizing ANP, which allows clusters and elements, as well as interactions between the elements of hierarchy. This approach can further lead to more accurate results in the subcriteria comparisons.
Another limitation presents the impossibility to examine the market solely from the customers’ point of view, which was the main reason for conducting interviews with experts in management and sales from companies with a range of up to two million clients. Based on their experience in ecommerce companies, we have obtained the requirements and experts’ judgments related to ecommerce businesses similar to those given by their clients.
The findings in this paper present a starting point for our continual research in the ebusiness area. Depending on the type of ebusiness and/or etrade, we plan to add or remove certain factors or subfactors. Furthermore, an extension to this research could focus on the practical application for the ranking of the alternatives of given websites. Finally, various MCDM methods, such as AHP, TOPSIS, and VIKOR, could be applied with fuzzy logic, Pythagorean fuzzy numbers, Intuitionictic fuzzy numbers, zNumbers, and/or Spherical fuzzy numbers.