AI-Assisted Sustainability Intelligence and Decision Support in Manufacturing Organizations Using Expert-Validated Indicators Under the Triple Bottom Line Framework

Abstract

Industrial sustainability assessment needs a transparent and operational framework that can manage multidimensional indicators, expert uncertainty, weighting complexity, and managerial interpretation. This research presents the Sustainable Industrial Measurement (SIM) Model, an AI-assisted sustainability intelligence architecture for manufacturing organizations. The model comprises literature-based indicator synthesis, Fuzzy Delphi Technique (FDT), Pareto 80/20 screening, Group Analytic Hierarchy Process (Group AHP), Utility Value Analysis and the web-based AI-enabled decision support system (AI-DSS) under the Triple Bottom Line (TBL) framework. In FDT validation by consensus, threshold and fuzzy score criteria, 33 experts accepted 64 indicators. The Pareto screening reduced the set to 50 high-impact indicators, consisting of 10 economic, 22 social and 18 environmental indicators. The priority weights were derived from the group AHP weighting by 21 experts and checked for consistency. The environmental and economic indicators represent the dominant sustainability priorities. The weighted structure was embedded in the web-based AI-DSS to enable automated scoring, visualization, gap diagnosis and AI-based managerial recommendations. Thirty industrial practitioners reported excellent perceived usability of the SIM Model, with a System Usability Scale score of 86.0. However, the evaluation assessed usability only, not the accuracy, effectiveness, or organizational impact of AI-assisted recommendations for manufacturing sustainability decisions and future implementation.

1. Introduction

Manufacturing is a key engine of economic growth, jobs, technology development and value creation [1,2]. However, industrial production also consumes large amounts of energy, water, raw materials, and labor and generates emissions, waste, occupational hazards and broader social impacts [2,3]. There is an increasing pressure on manufacturing organizations to move beyond the traditional measures of productivity, cost, and profitability to assess their performance. Sustainable manufacturing addresses this requirement by integrating economic viability, environmental responsibility, and social well-being within the Triple Bottom Line (TBL) framework [1,2,3]. The conceptual development of sustainable manufacturing is important but translating the principles into measurable and decision-relevant information is still challenging [3,4]. The literature contains a multitude of indicators regarding financial performance, production efficiency, resource consumption, occupational health and safety, labor practices, stakeholder relations, emissions, water management, waste, and environmental governance [3,4,5]. This diversity reflects the multidimensionality of sustainability but also presents challenges for the development of useful assessment systems [4,5]. Indicators may overlap in concept, have different measurement units, represent positive and negative performance directions, require data that are not always available, or have limited relevance to particular industrial contexts [4,5]. Empirical evidence also indicates that manufacturing organizations are still challenged in terms of consistently selecting, measuring and applying appropriate sustainability indicators [4].

Previous studies have proposed different approaches to structure and aggregate sustainability indicators. Helleno et al. [2] integrated TBL indicators with Lean Manufacturing and Value Stream Mapping, and Saad et al. [6] proposed a framework that includes indicator identification, normalization, weighting, aggregation and interpretation. Neri et al. [7] proposed a balanced set of TBL indicators for industrial supply chains. These studies provided the critical groundwork for multidimensional sustainability assessment. Nonetheless, the relevance of indicators is determined by the industrial sector, the size of the organization, regulatory conditions, stakeholder priorities, assessment level, and data availability [4,7,8,9]. Therefore, indicators from the literature need to be validated prior to being integrated into an operational framework. Expert validation can enhance the contextual appropriateness of candidate indicators. However, expert judgment is subjective by nature and often expressed in linguistically uncertain terms. To tackle this problem, the Fuzzy Delphi Technique (FDT) is used, which transforms the linguistic assessments into fuzzy values and applies explicit criteria for consensus and indicator acceptance [10]. Lin et al. [10] showed how it can be used for validation of sustainability indicators for employee activities related to production. FDT can decide whether an indicator is sufficiently supported by experts, but the accepted set may still be too large for efficient data collection, interpretation and subsequent weighting [10,11].

A large set of indicators increases data needs, respondent burden and analytical complexity, especially if pairwise comparisons are required [8,9,11]. The weighting is therefore based on screening of indicators. For example, Gani et al. [11] showed that Pareto analysis can reduce complexity by keeping only the indicators that contribute to the largest part of assessed importance. However, due to the study being mainly based on environmental indicators, there is a lack of studies applying Pareto screening to the whole TBL framework, especially as a link between fuzzy validation and hierarchical weighting [8,10,11]. Moreover, the retained indicators should have different weights, as they do not equally contribute to the overall sustainability performance [8,9,12]. Although methods such as Best–Worst Scaling have been used [8], the Analytic Hierarchy Process (AHP) offers a very structured way to derive hierarchical priorities and to evaluate the consistency of expert judgments [12]. But weights alone do not represent actual performance. Since sustainability indicators differ in terms of units, scales and desired directions (e.g., more financial return is positive, while more emissions or accident rates are negative), values must be normalized and combined with weights to generate comparable sustainability scores [6,7,12]. The practical value of these scores, however, lies not only in their mathematical aggregation but also in their communication. Most approaches end with scoring calculations and do not incorporate the results into an operational platform [9,12,13]. Furthermore, while Industry 4.0 technologies can improve the stages from data acquisition to decision-making [13], standard dashboards may not be able to provide actionable managerial guidance as they may not be able to clarify the root causes of weak performance, distinguish priority gaps, or identify complex relationships among TBL dimensions [9,13].

One possibility to bridge this interpretability divide is Generative artificial intelligence (GenAI). Gholami [14] used fuzzy logic, computational tools, and large language models in a manufacturing decision support application. Ghobakhloo et al. [15] identified potential contributions of GenAI to sustainable and human-centered manufacturing, such as analytical insight, knowledge accessibility, and operational decision support. Additionally, AI has been integrated with fuzzy multi-criteria techniques for evaluating ESG strategies in manufacturing [16]. These developments suggest that GenAI can add value beyond automated reporting. However, outputs produced directly from unstructured or unweighted data may be generic, inconsistent with organizational priorities or difficult to verify [14,15,16]. A more defensible role for AI is to interpret results that have been produced through an explicit process where the relevance, priority, scoring direction and performance level of indicators have already been set.

Despite recent progress in sustainability assessment and digital decision support systems, existing studies generally address indicator validation, indicator reduction, priority weighting, score calculation, system implementation, AI-assisted interpretation, and usability evaluation as separate or incomplete components. This creates a research gap in developing an integrated framework that can move from expert-validated TBL indicators to weighted scoring, operational system implementation, and AI-assisted managerial interpretation within a single workflow. The distinguishing feature of the proposed SIM Model is its sequential architecture, which combines FDT-based expert validation, Pareto-based indicator reduction, Group AHP weighting with consistency verification, Utility Value Analysis, web-based AI-DSS implementation, AI-assisted interpretation of structured sustainability scores, and formal usability evaluation using SUS. In this study, the SIM Model is developed as a sequential AI-assisted sustainability intelligence architecture for manufacturing organizations. The novelty of this work therefore lies not in any single method, but in the integration of these methodological and operational components into a transparent and usable sustainability assessment workflow.

2. Materials and Methods

The Sustainable Industrial Measurement (SIM) Model was developed through a sequential mixed-methods research design in this study. The methodology was designed to translate general sustainability notions into a validated, simplified, weighted, digitally implemented and AI-supported decision support architecture for manufacturing entities. The methodological workflow was composed of seven stages: (i) identification of candidate sustainability indicators based on the literature; (ii) expert-based validation using the Fuzzy Delphi Technique (FDT); (iii) reduction of indicators using Pareto 80/20 principle; (iv) hierarchical priority weighting using Group Analytic Hierarchy Process (Group AHP); (v) calculation of sustainability scores using Utility Value Analysis (UVA); (vi) implementation of the weighted assessment model within a web-based AI-enabled decision support system (AI-DSS); and (vii) assessment of usability and practical applicability using the System Usability Scale (SUS). The sequential design is based on the logic of systematic evidence synthesis, expert validation, multi-criteria prioritization, composite indicator construction and evaluation of the decision support system. The research workflow is shown in Figure 1.

2.1. Literature-Based Sustainable Industrial Indicator Identification

The first step was to generate a large set of candidate sustainability indicators for manufacturing organizations. A systematic literature review was performed to find the indicators used in sustainable manufacturing, industrial sustainability assessment, Triple Bottom Line evaluation, sustainability performance measurement, and multi-criteria decision-making frameworks in the past. The review process was guided by systematic review principles to improve transparency, traceability and reproducibility in indicator identification [17,18]. The search for peer-reviewed studies was conducted via Scopus, ScienceDirect and other academic databases. The search was limited to studies published between 2001 and 2024, clearly targeting the sustainability assessment in manufacturing or industrial contexts and operationalized indicators in economic, social and environmental dimensions. Studies were selected if they satisfied at least one of the following criteria: (i) proposed sustainability indicators for manufacturing processes or organizations; (ii) developed a sustainability assessment framework based on measurable indicators; (iii) applied multi-criteria decision-making methods for prioritizing sustainability indicators; or (iv) integrated TBL indicators into an industrial decision support framework. Studies were excluded when they were purely conceptual, focused on sustainability at the national level without organizational indicators, or did not provide measurable variables applicable to manufacturing organizations [17,18]. The extracted indicators were classified into economic, social and environmental dimensions, based on the TBL framework. Indicators that overlapped conceptually were merged based on their operational meaning rather than their wording. The merging process followed four criteria. First, indicators were merged when they referred to the same sustainability construct, such as resource efficiency, occupational safety, or environmental compliance. Second, indicators were merged when they used a similar measurement logic, for example, the same numerator–denominator relationship or the same type of performance ratio. Third, indicators were merged only when they had the same performance direction, meaning that higher or lower values indicated the same sustainability interpretation. Fourth, the broader and more operationally measurable indicator label was retained to support practical data collection. Indicators were not merged when they differed in measurement unit, organizational level, stakeholder meaning, or managerial implication. This process reduced redundancy while preserving indicators with distinct operational relevance for manufacturing sustainability assessment. For example, in order to reduce redundancy, similar indicators with different labels were merged into one representative indicator before validation by experts. This process resulted in the generation of 65 candidate sustainability indicators, including 13 economic, 28 social and 24 environmental indicators. These indicators were the initial input for the FDT validation stage.

2.2. Expert Recruitment and Data Collection

In this study, two panels of experts were used. The first panel was used to validate indicators based on the FDT, and the second panel was used to weight the hierarchies based on the AHP. Experts were recruited using purposive expert sampling because the study required specialized judgments regarding manufacturing sustainability rather than statistical representation of a general population. The inclusion criteria were as follows: (i) direct professional, managerial, technical, or academic experience in manufacturing, sustainability assessment, industrial management, environmental management, occupational health and safety, or multi-criteria decision-making; (ii) at least five years of relevant experience, where applicable; (iii) familiarity with industrial sustainability indicators or performance assessment; and (iv) willingness to provide independent and complete judgments. The FDT panel was composed of 33 specialists coming from three knowledge domains: (1) 11 industrial practitioners working in manufacturing or sustainability management, (2) 11 academic researchers specialized in sustainability assessment or industrial management and (3) 11 industrial networks representatives. The total number of 33 experts was selected to ensure balanced representation across the three knowledge domains, with 11 experts from each group. This balanced structure was intended to combine practical manufacturing experience, academic and methodological expertise, and broader industrial network perspectives. The panel size was therefore considered appropriate for FDT-based indicator validation because the purpose was to obtain qualified expert consensus on the contextual relevance of 65 candidate indicators rather than to achieve statistical representativeness of a general population. The anonymized characteristics of the expert panels, including expert group, role in the study, sector or institutional affiliation, qualification criteria, experience criteria, and selection rationale, are summarized in Table S1. This composition was aimed at encompassing theoretical knowledge as well as practical experience related to manufacturing sustainability assessment [19,20,21]. The AHP panel was made up of 21 experts chosen from the larger pool of experts based on their knowledge of sustainability assessment, industrial management, and multi-criteria evaluation. For AHP, a smaller expert panel was used due to the greater cognitive effort involved in pairwise comparison as compared to rating-based validation. Although these panels were relatively small compared with general survey research, they are methodologically appropriate for expert-based studies, where validity depends primarily on expert qualification, domain relevance, and judgment consistency rather than statistical representativeness. Recent expert-input research in the low-carbon hydrogen sector similarly used 20 screened experts and noted that such a modest sample size is consistent with expert opinion-based literature [22]. Therefore, the use of 33 experts for FDT and 21 experts for AHP is consistent with established norms for specialized expert elicitation studies. Expert answers were anonymized and used solely for methodological analysis. The importance of sustainability dimensions, categories and indicators was assessed using Saaty’s pairwise comparison scale [23,24,25] in the AHP questionnaire and a seven-point importance scale in the FDT questionnaire.

2.3. Fuzzy Delphi Technique for Indicator Validation

The Fuzzy Delphi Technique was used to validate the contextual relevance and importance of the 65 candidate indicators. FDT was chosen because expert judgments in sustainability assessment are often expressed in linguistic terms and may contain uncertainty, ambiguity and subjective variation. Fuzzy set theory is a mathematical way to represent such uncertainty. Delphi logic supports the expert-based consensus formation [10,19,20,21]. Each expert rated the importance of each candidate indicator on a seven-level scale. Linguistic ratings were converted into triangular fuzzy numbers (TFNs) as shown in Table S2. TFNs were used for their capability to represent the lower, most likely and upper bounds of expert judgment in a simple and interpretable form [10,19,20,21].

For indicator j evaluated by expert i, the linguistic rating was represented as a triangular fuzzy number, as shown in Equation (1) [19,20,21]:

$${\tilde{x}}_{ij}=\left(l_{ij},m_{ij},u_{ij}\right)$$

(1)

where ${\tilde{x}}_{ij}$ is the fuzzy rating assigned by expert $i$ to indicator $j$; $l_{ij}$, $m_{ij}$, and $u_{ij}$ denote the lower, middle, and upper values of the TFN, respectively.

The group fuzzy opinion for each indicator was then obtained by averaging the lower, middle, and upper fuzzy values across all experts, as shown in Equation (2) [19,20,21]:

$${\tilde{A}}_j=\left(L_j,M_j,U_j\right)\left({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}{l}_{ij},{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}{m}_{ij},{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}{u}_{ij}\right)$$

(2)

where ${\tilde{A}}_j$ is the aggregated TFN for indicator $j$; $L_j$, $M_j$, and $U_j$ are the aggregated lower, middle, and upper fuzzy values; and n is the number of experts.

Expert consensus was evaluated using the vertex distance method between each expert’s TFN and the aggregated group TFN, as shown in Equation (3) [19,20,21]:

$$d_{ij}=\sqrt{{\displaystyle \frac{1}{3}}\left[{\left(l_{ij}-L_j\right)}^2+{\left(m_{ij}-M_j\right)}^2+{\left(u_{ij}-U_j\right)}^2\right]}$$

(3)

where $d_{ij}$ is the fuzzy distance between the assessment of expert $i$ and the aggregated group fuzzy opinion for indicator $j$. A smaller distance indicates stronger agreement between the individual expert and the group judgment.

The average threshold value of each indicator was calculated using Equation (4) [20,21]:

$$D_j={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}{d}_{ij}$$

(4)

where $D_j$ is the average threshold value of indicator $j$. In this study, $D_j\le 0.20$ was used as the threshold for acceptable fuzzy consensus.

The percentage of expert consensus was calculated using Equation (5) [20,21]:

$$\left[P_j={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n}I\left(d_{ij}\le 0.20\right)}{n}}\times 100\right]$$

(5)

where $P_j$ is the expert consensus percentage for indicator $j$, and $I$ is an indicator function equal to 1 when the condition is satisfied and 0 otherwise.

The fuzzy importance score of each indicator was calculated through defuzzification using the simple average method, as shown in Equation (6) [19,20,21]:

$$\left[F_j={\displaystyle \frac{L_j+M_j+U_j}{3}}\right]$$

(6)

where $F_j$ is the defuzzified fuzzy importance score of indicator $j$.

An indicator was accepted when all three conditions were simultaneously satisfied: average threshold value not greater than 0.20, expert consensus not lower than 75%, and fuzzy importance score not lower than 0.50. The acceptance rule is expressed in Equation (7) [20,21]:

$$Accep{t}_j=\{\begin{array}{cc}1,& D_j\le 0.20\wedge P_j\ge 75\%\wedge F_j\ge 0.50\\ 0,& otherwise\end{array}$$

(7)

where $Accep{t}_j=1$ indicates that indicator $j$ was retained for subsequent analysis, while $Accep{t}_j=0$ indicates that the indicator was rejected. The accepted indicators were then advanced to the Pareto screening stage.

2.4. Indicator Screening Using the Pareto 80/20 Principle

After FDT validation, the Pareto principle (80/20) was used to reduce the number of indicators while retaining those with the strongest contribution to expert-assessed sustainability importance. The 80% cumulative contribution threshold was selected as a pragmatic screening point to retain the dominant share of indicator importance while improving analytical manageability. This logic is consistent with the use of Pareto analysis to identify “vital” sustainability indicators in the manufacturing sector before further multi-criteria prioritization [11]. The threshold was also appropriate for this study because sustainability assessment in manufacturing involves multiple indicators, heterogeneous data requirements, and practical implementation challenges [8,9]. Retaining all 64 FDT-validated indicators would have increased the burden of data collection, management complexity, and the workload of pairwise comparison in the subsequent AHP stage. Therefore, the 80/20 screening rule was applied not to exclude theoretically relevant indicators, but to obtain a manageable high-impact indicator set for Group AHP weighting and AI-DSS implementation [11,12].

For each accepted indicator, the defuzzified fuzzy importance score obtained from FDT was normalized into a relative contribution value, as shown in Equation (8) [8,11]:

$$\left[q_j={\displaystyle \frac{F_j}{{\displaystyle {\sum }_{j=1}^m}{F}_j}}\right]$$

(8)

where $q_j$ is the normalized relative contribution of indicator $j$, $F_j$ is the defuzzified fuzzy importance score, and m is the number of FDT-accepted indicators.

The accepted indicators were ranked in descending order according to $q_j$. The cumulative contribution of the ranked indicators was then calculated using Equation (9) [8,11]:

$$\left[C_k={\displaystyle \sum _{r=1}^k}{q}_{\left(r\right)}\right]$$

(9)

where $C_k$ is the cumulative contribution of the top $k$ ranked indicators, and $q_{\left(r\right)}$ denotes the normalized contribution of the indicator ranked in position $r$.

Indicators were retained when their cumulative contribution fell within the Pareto threshold, as shown in Equation (10) [8,11]:

$${Retain}_j=\{\begin{array}{cc}1,& C_k\le 0.80\\ 0,& C_k>0.80\end{array}$$

(10)

where ${\mathrm{Retain}}_j=1$ indicates that the indicator was retained for AHP weighting. When the indicator at the threshold boundary was required to preserve dimensional representation, it was retained to avoid excluding a high-relevance indicator from a TBL dimension.

2.5. Hierarchical Priority Weighting Using Group AHP

The AHP was used to obtain the relative priority weights of the retained indicators. AHP was selected because it breaks down a complex decision problem into levels within a hierarchy and employs pairwise comparisons to produce priority weights with a consistency check. This consistency mechanism is important because expert judgments need to be logically consistent before being used in a weighted sustainability scoring model [23,24,25]. The AHP hierarchy was three levels deep. The first level was the overall goal, namely the industrial sustainability performance. The second level consisted of the three dimensions of TBL: economic, social and environmental sustainability. The third level was constituted by the Pareto-retained sustainability indicators in each dimension. The experts used Saaty’s scale of pairwise comparison to compare elements within the same hierarchical level, where 1 means equal importance and 9 means extreme importance of one element over another [23,24,25]. Since AHP involves multiple experts, the pairwise comparison matrices were aggregated by using the geometric mean method. In the present study, expert judgments were aggregated using equal response weighting as a neutral baseline, meaning that each expert contributed equally to the group AHP matrix. Alternative expert response-weighting schemes, such as experience-based weighting and familiarity-based weighting, were not applied in the current analysis. Previous expert-prioritization research has highlighted the value of examining alternative response-weighting assumptions when assessing the stability of aggregated rankings [26]. Future sensitivity analysis should therefore compare equal, experience-based, and familiarity-based weighting schemes to determine whether the resulting indicator priorities remain stable. The group comparison value for each pairwise comparison between element i and element j was obtained using Equation (11) [24,25]:

$$a_{ij}={\left({\prod }_{k=1}^K{a}_{ij}^{\left(k\right)}\right)}^{{\displaystyle \frac{1}{K}}}$$

(11)

where $a_{ij}$ is the aggregated group comparison value between elements $i$ and $j$; $a_{ij}^{\left(k\right)}$ is the comparison value provided by expert $k$; and $K$ is the number of experts.

The aggregated pairwise comparison matrix was normalized by column, as shown in Equation (12) [23,25]:

$$p_{ij}={\displaystyle \frac{a_{ij}}{{\sum }_{i=1}^n{a}_{ij}}}$$

(12)

where $p_{ij}$ is the normalized value of element i relative to element j, and n is the number of elements in the matrix.

The local priority weight of each element was then calculated using the row average method, as shown in Equation (13) [23,25]:

$$w_i={\displaystyle \frac{1}{n}}{\displaystyle \sum _{j=1}^n}{p}_{ij}$$

(13)

where $w_i$ is the local priority weight of element i.

To evaluate the internal consistency of the pairwise comparison matrix, the maximum eigenvalue was estimated using Equation (14) [23,25]:

$${\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n}{\displaystyle \frac{{\left(\overline{A}w\right)}_i}{w_i}}$$

(14)

where ${\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}$ is the maximum eigenvalue of the aggregated comparison matrix $\overline{A}$, and w is the priority weight vector.

The Consistency Index was then calculated using Equation (15) [23,25]:

$$CI={\displaystyle \frac{{\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}-n}{n-1}}$$

(15)

where CI is the Consistency Index and n is the matrix size.

The Consistency Ratio was calculated using Equation (16) [23,25]:

$$CR={\displaystyle \frac{CI}{RI}}$$

(16)

where CR is the Consistency Ratio and RI is the Random Index corresponding to the matrix size. A CR value below 0.10 was considered acceptable. Matrices with CR values greater than 0.10 were reviewed before final weight calculation.

Finally, the global weight of each indicator was calculated by multiplying the relevant hierarchical weights, as shown in Equation (17) [23,25]:

$$\left[G{W}_j=W_d\times W_{c|d}\times W_{j|c}\right]$$

(17)

where $G{W}_j$ is the global weight of indicator j; $W_d$ is the weight of the TBL dimension; $W_{c|d}$ is the weight of the category within dimension d, if applicable; and $W_{j|c}$ is the local weight of indicator j within its category. The resulting global weights were used in the sustainability scoring model and the AI-DSS computational algorithm.

2.6. Utility Value Analysis and Sustainability Score Calculation

The retained and weighted indicators were diverse in their units, measurement scales and direction of performance. Some indicators were benefit-oriented, where higher values indicated better sustainability performance, for example, return on assets or employee training. Others were cost-based where lower values such as energy intensity, accident rate, waste generation or greenhouse gas emissions showed better performance. Hence, the different indicator values were transformed into standardized utility scores and then aggregated using the Utility Value Analysis. This is consistent with the construction of composite indicators, where normalization, weighting and aggregation are necessary to derive interpretable multidimensional scores [27,28]. Utility Value Analysis was selected because the objective of this stage was to transform heterogeneous sustainability indicators into standardized and interpretable performance scores, rather than only to rank alternatives. In the SIM Model, Group AHP already provides the relative priority weights of indicators, while UVA provides the operational scoring mechanism that converts benefit-oriented and cost-oriented indicator values into a common 0–5 utility scale. This makes the results suitable for indicator-level diagnosis, dimension-level aggregation, overall sustainability scoring, dashboard visualization, and longitudinal comparison. Other MCDM methods, such as TOPSIS, VIKOR, or PROMETHEE, are well established for alternative ranking, but they are less aligned with the need for a transparent, direction-sensitive, and score-based assessment scale for organizational sustainability monitoring.

For benefit-oriented indicators, the utility score was calculated using Equation (18) [27,28]:

$$U_{ij}=5\times {\displaystyle \frac{x_{ij}-x_j^{\mathrm{m}\mathrm{i}\mathrm{n}}}{x_j^{\mathrm{m}\mathrm{a}\mathrm{x}}-x_j^{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(18)

where $U_{ij}$ is the utility score of organization i for indicator j; $x_{ij}$ is the observed value; $x_j^{\mathrm{m}\mathrm{i}\mathrm{n}}$ is the minimum benchmark value; and $x_j^{\mathrm{m}\mathrm{a}\mathrm{x}}$ is the maximum benchmark value. The score was scaled from 0 to 5.

For cost-oriented indicators, the utility score was calculated using Equation (19) [27,28]:

$$U_{ij}=5\times {\displaystyle \frac{x_j^{\mathrm{m}\mathrm{a}\mathrm{x}}-x_{ij}}{x_j^{\mathrm{m}\mathrm{a}\mathrm{x}}-x_j^{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(19)

where a lower observed value produces a higher utility score. The minimum and maximum benchmark values were selected using a hierarchical protocol to ensure consistent implementation. Regulatory or compliance thresholds were prioritized first, followed by sector-specific industry references, historical organizational data, and expert-defined ranges only when other benchmarks were unavailable. The selected benchmark source, reference year, rationale, indicator direction, and minimum–maximum values were documented in the system database to support transparency, comparability, and score traceability. For comparative or longitudinal assessment, the same benchmark set should be applied across assessed units and periods, with any benchmark update recorded as a new version. When $x_j^{\mathrm{m}\mathrm{a}\mathrm{x}}=x_j^{\mathrm{m}\mathrm{i}\mathrm{n}}$, the indicator was treated as non-discriminating for the assessed dataset and was reviewed before aggregation.

The dimension-level sustainability score was calculated using Equation (20) [27,28]:

$$S_{id}={\displaystyle \frac{\sum _{j\in d}G{W}_j{U}_{ij}}{\sum _{j\in d}G{W}_j}}$$

(20)

where $S_{id}$ is the sustainability score of organization i in dimension d, and $j\in d$ denotes the set of indicators belonging to that dimension.

The overall SIM sustainability score was calculated using Equation (21) [27,28]:

$$S_i={\displaystyle \sum _{j=1}^K}G{W}_j{U}_{ij}$$

(21)

where $S_i$ is the overall sustainability score of organization i, K is the total number of weighted indicators, $G{W}_j$ is the global weight of indicator j, and $U_{ij}$ is the normalized utility score. Because the global weights were normalized to sum to 1, the overall score ranged from 0 to 5.

To support managerial interpretation, the overall and dimension-level scores were classified into five performance levels: Poor/High Risk (0–1.00), Needs Improvement (>1.00–2.00), Average/Moderate (>2.00–3.00), Very Good (>3.00–4.00), and Excellent/Best Practice (>4.00–5.00). This classification was used to translate numerical sustainability scores into interpretable managerial categories.

To identify improvement priorities, a weighted performance gap score was calculated using Equation (22) [27,28]:

$$G_{ij}=G{W}_j\left(5-U_{ij}\right)$$

(22)

where $G_{ij}$ is the weighted gap score of organization i for indicator j. A high $G_{ij}$ value indicates an indicator with both high strategic importance and weak performance. This value was used by the AI-assisted interpretation module to rank priority improvement areas.

2.7. Development of the Web-Based AI-Enabled Decision Support System

The weighted sustainability assessment model was developed as a web-based decision support system enabled by AI—see Figure 2. The AI-DSS aimed to transfer the validated and weighted indicator framework into an operational platform to support data entry, automated scoring, visualization, benchmarking, and AI-assisted interpretation. Decision support systems are especially helpful in sustainable manufacturing because of the multiple criteria, trade-offs, uncertainty and need for actionable recommendations that are involved in sustainability decisions [29]. The architecture of the system consisted of five major modules, namely user management, input of sustainability data, score calculation, visualization and reporting, and AI-assisted recommendation. The data input module enables users to input organizational data for each sustainability indicator. The score calculation module used the normalization and weighted aggregation equations described in Section 2.6. The visualization module provided dashboards, tables, radar charts and comparative summaries of the indicator-level, dimension-level and overall sustainability performance. The reporting module produced structured outputs of sustainability assessments for organizational review.

The backend database was designed to store semi-structured sustainability data such as organization profile, indicator values, benchmark values, AHP weights, utility score and AI-generated interpretation records. A document-oriented database structure was used, as sustainability data may vary across organizations, sectors and assessment cycles. Such database architectures are suitable for flexible and evolving data models, especially if semi-structured records and changing schema requirements have to be handled [30]. To support web access and system stability, the system was deployed in a web server environment supported by traffic balancing and reverse-proxy logic, which can improve availability, scalability, and request distribution in server-based applications [31]. The AI component was integrated via an API-based integration to enable structured sustainability interpretation. The AI module did not compute indicator weights or modify sustainability scores. Instead, it used the validated and weighted score profile produced by the SIM Model. We created structured input objects for the AI module, which include indicator names, TBL dimensions, normalized utility scores, global weights calculated from the AHP, weighted gap scores, and performance classes. This design makes AI-generated outputs based on traceable quantitative evidence and structured schema-based input instead of unstructured raw data [32,33].

The AI module was implemented using the Gemini 3.5 Flash model through the Gemini API and functioned only as an interpretive layer above the deterministic scoring model. The AI did not calculate AHP weights or modify sustainability scores; instead, it interpreted structured JSON inputs containing the selected year, ECN/ENV/SCL indicator data, system-calculated scores, and related performance information. The prompt structure consisted of four controlled components: role assignment, structured data injection, JSON-only output constraints, and field-specific instructions for improvements, best practices, summary, and output language. To reduce generic or unsupported outputs, the AI was instructed to generate responses only from the supplied sustainability data and to return outputs in a predefined JSON schema. Hallucination risk was further controlled through output validation and fallback responses when the API failed, returned invalid JSON, or produced empty recommendation fields. AI-generated outputs were presented as advisory recommendations only. Final managerial decisions remained under human responsibility. To support traceability and reproducibility, the system recorded the model identifier, prompt template, structured input data, generated output, output language, assessment year, form type, and anonymized organization identifier for each request. Only structured indicator values and system-calculated scores were transmitted to the API, while factory names and other directly identifying information were excluded. Access to stored assessment records and AI-generated outputs was restricted through authenticated user roles and private database permissions.

2.8. System Usability and Practical Applicability Assessment

The usability and practical applicability of the developed AI-DSS were assessed using the System Usability Scale. SUS was selected because it is a standardized, widely used, and efficient instrument for evaluating perceived usability of interactive systems [34,35,36]. The assessment involved 30 practitioners from 30 industrial organizations. Participants used the AI-DSS prototype and then completed a 10-item SUS questionnaire using a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

For each respondent, the SUS score was calculated using Equation (23) [34,35,36]:

$$SU{S}_i=2.5\left[\sum _{q\in \mathrm{1,3},\mathrm{5,7},9}\left(R_{iq}-1\right)+\sum _{q\in \mathrm{2,4},\mathrm{6,8},10}\left(5-R_{iq}\right)\right]$$

(23)

where $SU{S}_i$ is the SUS score of respondent i, and $R_{iq}$ is the rating provided by respondent i for item q. Odd-numbered items are positively worded, while even-numbered items are negatively worded. The multiplication factor of 2.5 converts the total score to a 0–100 scale.

The mean SUS score was calculated using Equation (24) [34,35,36]:

$$\overline{SUS}={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}SU{S}_i$$

(24)

where $\overline{SUS}$ is the average usability score and N is the number of respondents. The resulting SUS score was interpreted using established usability benchmarks, where higher scores indicate stronger perceived usability, ease of interaction, and practical suitability [35,36].

In addition to SUS, a Satisfaction Index was calculated to summarize user satisfaction with the AI-DSS as a practical decision support tool. The Satisfaction Index was calculated as the percentage of the maximum possible Likert score, as shown in Equation (25) [37]:

$$SI={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N}{\displaystyle {\sum }_{q=1}^Q}{R}_{iq}}{N\times Q\times R_{\mathrm{m}\mathrm{a}\mathrm{x}}}}\times 100$$

(25)

where SI is the Satisfaction Index, Q is the number of satisfaction items, $R_{iq}$ is the rating provided by respondent i for item q, and $R_{\mathrm{m}\mathrm{a}\mathrm{x}}$ is the maximum Likert score. The SI values were interpreted as follows: 80–100% = very satisfied, 60–80% = satisfied, 40–60% = moderately satisfied, 20–40% = less satisfied, and 0–20% = not satisfied.

3. Results and Discussion

3.1. Sustainability Indicator Synthesis

A systematic literature synthesis identified 65 candidate sustainability indicators for manufacturing organizations within the Triple Bottom Line (TBL) framework. The complete list of the 65 candidate sustainability indicators is presented in Table S3, including 13 economic indicators, 28 social indicators, and 24 environmental indicators. This table represents the full indicator catalog used as the input for FDT validation. The key implication is that the initial indicator pool captured the multidimensional nature of manufacturing sustainability. The relatively large number of social and environmental indicators reflects the broader operational responsibilities of manufacturing organizations beyond financial performance, including employee welfare, occupational safety, resource efficiency, emissions, waste, and environmental governance.

The broad distribution of indicators shows that the manufacturing sustainability cannot be assessed using environmental indicators alone. The results indicate that industrial sustainability has a very visible component in environmental performance, but that economic and social measures are also necessary to reflect the operational reality of manufacturing organizations. This is in agreement with Neri et al. [7], who stressed the necessity of a balanced set of TBL indicators for industrial supply chains, and Saad et al. [6], who argued that sustainability assessment frameworks should include the identification, normalization, weighting, aggregation and interpretation of indicators. Similarly, Contini and Peruzzini [38] demonstrated that sustainability KPIs for manufacturing companies should include the environmental, economic, and social dimensions across the value chain rather than a single performance dimension. It also displays the complexity of manufacturing sustainability, as evidenced by the relatively large number of social and environmental indicators in the initial pool. Social sustainability in manufacturing is linked to workplace safety, labor standards, employee well-being, training, stakeholder relations and customer-related outcomes. Environmental sustainability is energy, emissions, water, material use, waste, and regulatory responsibility. These results are in line with Trianni et al. [4], who found that manufacturing companies have difficulties with the selection and the consistent application of sustainability indicators due to the multiple dimensions and heterogeneous data requirements of sustainability performance. Therefore, the 65-indicator pool provided a comprehensive but preliminary structure that required expert validation before adoption in the SIM Model.

3.2. Fuzzy Delphi Validation of Candidate Indicators

The 65 candidate indicators were validated by experts’ judgments of 33 specialists using the Fuzzy Delphi Technique (FDT). Acceptance criteria were expert consensus ≥ 75%, threshold value ≤ 0.20 and fuzzy importance score ≥ 0.50. The detailed FDT results for each TBL dimension are presented in Tables S4–S6. The results showed an acceptance rate of 98.5% with 64 of the 65 indicators meeting all three criteria. The only indicator that did not meet the consensus criterion was SCL 1.10 Employment Opportunities for Vulnerable Groups and was therefore excluded from further analysis. Kendall’s coefficient of concordance was statistically significant and showed agreement among the experts (p < 0.05). This result indicates that the expert panel size was sufficient to produce consistent collective judgments for indicator validation, supporting the methodological credibility of the FDT results despite the purposive and specialized nature of the sample. The high acceptance rate indicates that the literature-derived indicator pool was mainly relevant for manufacturing sustainability assessment. But the rejection of one social indicator is important methodologically. The indicator that was excluded was not necessarily unimportant in normative terms, but the expert panel did not reach sufficient agreement on its general applicability across manufacturing contexts. This suggests that some social sustainability indicators may be highly dependent on the size of the firm, its labor policy, type of production, regulatory environment and organizational strategy. This result supports the argument that sustainability indicators should not be transferred from the literature directly to the operational assessment systems without contextual validation [4,7,9].

The FDT results are also in line with Lin et al. [8] who proved that FDT is appropriate to develop sustainability indicators when expert judgments involve linguistic uncertainty. Unlike the conventional rating methods, FDT converts qualitative evaluations into triangular fuzzy numbers, which allows the quantification of subjective expert judgments without ignoring the uncertainty. In this research, the use of FDT helped to improve the construct validity of the SIM Model by providing an indication that retained indicators are not only of theoretical significance but are also accepted by experts from manufacturing-related domains. The FDT stage provides a more robust methodological basis than research based solely on literature review or pre-established lists of indicators. Hourneaux et al. [39] proposed a minimum set of TBL indicators for industrial companies. Mengistu and Panizzolo [40] empirically analyzed useful and applicable sustainability metrics for SMEs. The studies provide important evidence that the practical applicability of an indicator must be considered in its selection. In this paper, the logic is extended with the application of fuzzy expert consensus in order to validate the indicator set before further reduction and weighting.

Although FDT provides a systematic mechanism for representing linguistic uncertainty and aggregating multiple expert judgments, it does not eliminate the subjective origin of those judgments. The observed consensus indicates agreement within the participating panel rather than universal acceptance across all manufacturing contexts. The retained indicator structure may therefore be influenced by the professional, institutional, and geographical composition of the panel.

3.3. Pareto Screening and Indicator Reduction

After validating FDT, the Pareto 80/20 principle was used to reduce the validated indicator set from 64 to 50 high-impact indicators. The final selected set consisted of 10 economic indicators, 22 social indicators and 18 environmental indicators—see Figure 3. This resulted in an approximately 21.9% reduction in the number of indicators while maintaining the TBL structure. To examine whether the final indicator set was dependent on a single cut-off rule, alternative cumulative contribution thresholds were also considered. A 70% threshold would have retained 43 indicators, producing a more compact set but with lower TBL coverage, whereas the 85% and 90% thresholds would have retained 54 and 56 indicators, respectively, thereby increasing coverage but also increasing the analytical burden for data collection and AHP pairwise comparison. Therefore, the 80% threshold was retained because it provided a practical balance between sustainability coverage and operational manageability.

The aim of Pareto screening was not to arbitrarily exclude theoretically relevant indicators but to reduce analytical complexity before the AHP stage. This is important since pairwise comparison becomes more and more onerous with the increasing number of indicators. If all 64 FDT-validated indicators for AHP were to be retained, the expert panel would have had to make a much larger number of comparisons, increasing the risk of fatigue, inconsistent judgment and lower quality data. Those economic indicators that were retained were related to cost efficiency and financial return like technology cost per production value, production cost per unit, return on assets, return on investment, value added from production processes, and payback period. The social indicators continued to focus on occupational health and safety, labor rights, employee development, stakeholder engagement, supplier safety, ethics and customer satisfaction. The retained environmental indicators concerned energy and climate, water and wastewater, material efficiency, circularity, waste utilization, environmental compliance, and environmental risk assessment. The findings are in agreement with Gani et al. [11], who identified key environmental sustainability indicators in the manufacturing industry using a Pareto analysis. However, the present study extends that approach by screening the entire TBL framework with Pareto screening, not just the environmental indicators. The results also support the work of Swarnakar et al. [8] who pointed out that sustainability indicator prioritization requires a structured approach since manufacturing firms cannot monitor all possible sustainability indicators equally with practical attention. The Pareto stage thus served as a bridge between indicator validation and indicator weighting. The FDT identified the relevant indicators, which were accepted by the experts. The Pareto screening identified the accepted indicators contributing most strongly to the total importance assessed by experts. This sequence logic reduces redundancy and enhances operational manageability of the SIM Model. Indicators excluded by the Pareto stage were not classified as invalid or unimportant. They remain within the comprehensive master catalog and may be reinstated when required by sector-specific materiality, legal obligations, stakeholder priorities, or organizational strategy. This consideration is particularly important for social and environmental indicators whose importance may be high in a specific context despite a lower aggregate expert score.

3.4. AHP-Based Priority Structure of Sustainability Indicators

Group AHP was used to assign weights to the 50 Pareto-retained indicators based on 21 experts’ judgments. The AHP hierarchical structure encompassing all three TBL dimensions and their constituent indicators is presented in Figure S2. All Consistency Ratios were below the acceptable limit of 0.10, which means that the expert judgments were logically consistent. For the economic dimension, the CR was 0.0000 at the category level, 0.0719 for the social dimension, and 0.0548 for the environmental dimension. These results show that the pairwise comparison matrices were internally consistent and appropriate for deriving the priority weights. The TBL dimension-wise aggregation of the global weights revealed that the environmental dimension had the highest share of total weight with approx. 43.8%, the economic dimension with approx. 39.0%, and the social dimension with approx. 17.2%. This result shows that experts strongly emphasized environmental and economic performance strategically. The aggregate level environmental dominance reflects the increasing pressure on manufacturing firms to manage energy use, emissions, resource consumption, water intensity, waste and environmental compliance. This result can be explained by the operational characteristics of manufacturing organizations, where environmental performance is directly connected to energy consumption, raw material use, water demand, waste generation, emissions, and regulatory compliance. These issues are usually measurable, externally visible, and closely linked to both environmental risk and operating cost. Therefore, experts may have assigned higher aggregate weight to environmental indicators because they represent sustainability issues with immediate managerial, regulatory, and resource-efficiency implications. The high environmental weight should not be interpreted as reducing the importance of economic or social sustainability; rather, it indicates that environmental performance was perceived as the most urgent and quantifiable sustainability priority at the dimensional level. This is consistent with the results of Yip et al. [3] and Kumar and Mani [9] who stated that sustainability assessment in the manufacturing industry is increasingly driven by environmental responsibility, regulatory pressure and the need for quantifiable operational improvements.

However, the most dominant individual indicators were economic indicators. Return on assets (ECN 1.10) was first with a global weight of 0.0831, followed by value added from production processes (ECN 1.11) with a global weight of 0.0630. Thus, although the environmental dimension was the most influential dimension overall, experts still considered economic performance to be a crucial enabling condition for the implementation of sustainability. Manufacturing organizations need financial strength, value creation and investment capacity to support environmental technologies, safety systems, employee development and digital sustainability platforms. The third and fourth highest ranked indicators were environmental indicators: Material Efficiency Relative to Waste Generation (ENV 1.12; 0.0492) and water intensity (ENV 1.9; 0.0472). Other most weighted environmental indicators were energy intensity, recyclable waste, renewable and alternative energy utilization, reusable waste, wastewater treatment, and annual greenhouse gas emissions. The results show that the priorities of environmental sustainability in manufacturing are related to resource productivity and circularity and not only to compliance with regulations. In correlation with the statement of Contini and Peruzzini [36] that manufacturing sustainability KPIs should allow companies to measure performance along the value chain and that digitalization and Industry 4.0 technologies can support real-time sustainability monitoring in industrial environments [41], this interpretation is presented.

The lower aggregate weight of the social dimension should not be interpreted to mean that social sustainability is less important. Rather, it suggests that experts perceived social performance to be more context-sensitive and less directly quantifiable than economic and environmental performance. In the social aspect, occupational health and safety had the highest category weight, which means that safety at the workplace is still the most important social sustainability issue in manufacturing. This finding is consistent with the pragmatic nature of industrial operations, where accidents, lost time injuries, occupational diseases and safety training are directly related to worker well-being, productivity loss, regulatory exposure and organizational reputation. Thus, the AHP results show that the SIM Model is not assigning equal or subjective weights to sustainability indicators. Instead, it generates a differentiated priority structure based on expert judgment and validated by consistency testing. This is a major improvement over sustainability assessment approaches that aggregate indicators irrespective of their relative importance [6,27]. The weighted structure allows the SIM Model to detect not only if an organization performs well or not, but also if the weak performance is in strategically important indicators. The complete global weight ranking is provided in Table S7, while matrix-level consistency results are reported in Tables S8–S11.

Group AHP structures expert judgments and evaluates their internal consistency, but acceptable CR do not establish objective or universal priority weights. The geometric mean reduces the influence of individual responses, while CR evaluates logical coherence within each matrix. A different panel composition may nevertheless produce a different priority structure.

3.5. Integrated Interpretation of the Final Indicator Architecture

The final SIM indicator architecture combines three methodological logics: acceptance based on FDT, reduction based on pareto, and weighting based on AHP. This sequential structure of the process gives a lot of credibility to the assessment model since each indicator that remains in the final framework has passed three steps of filtering: contextual relevance, contribution to the total importance and hierarchical priority. This finding tackles the methodological fragmentation identified in previous sustainability assessment research. Some studies have proposed indicator sets for manufacturing sustainability [6,7,38,40] and others have applied MCDM methods such as AHP, BWM or fuzzy approaches for prioritization [8,11,12,25]. But many frameworks only rank the indicators or do not operationalize the weighted indicators in a decision support system. The SIM Model overcomes this limitation by linking indicator governance with digital implementation and AI-assisted interpretation. The integration of Pareto screening between FDT and AHP is especially important. Gani et al. [11] have demonstrated the utility of Pareto analysis for prioritizing the critical environmental indicators. Swarnakar et al. [8] have emphasized the need for a systematic prioritization of the sustainability indicators in manufacturing. In this study, Pareto screening has been used not as a separate prioritization method but as a complexity reduction step before AHP. This design reduces the cognitive load of the experts and makes the group-based pairwise comparison more practicable. Thus, the final 50-indicator architecture is comprehensive enough to represent TBL sustainability and concise enough for organizational application. This balance is important, since large sustainability indicator systems may be difficult for firms to implement, particularly small and medium-sized manufacturers with limited data collection capacity. Similarly, Mengistu and Panizzolo [40] argued that the sustainability metrics should be meaningful and applicable to firms without overwhelming them with too much or uncertain information. This need is endorsed by the SIM Model that offers a reduced, weighted and system ready indicator framework.

3.6. AI-DSS Implementation and Sustainability Intelligence

The validated and weighted indicator structure was built into a web-enabled AI-based decision support system (AI-DSS). The system comprises modules for user management, data input, sustainability scoring, visualization, reporting and AI-assisted recommendation. Figures S2 and S3 present examples of sustainability assessment results from the industrial case study dataset, demonstrating the system’s capacity to generate comprehensive, indicator-level performance profiles. Figure 4 illustrates radar chart visualization for comparative sustainability analysis and the automated summary report output, demonstrating the multi-level analytical communication capability of the platform. The platform provides assessment at the indicator, dimension and organizational levels and presents results via dashboards, radar charts, comparative summaries and sustainability reports. The AI-DSS implementation is a significant contribution as it transforms sustainability assessment from a static measurement exercise into an operational decision support process. Many sustainability assessment studies set up indicators and weights but do not show how the results can be integrated into a user-friendly system for practitioners. Unlike the static nature of the indicator architecture, the SIM Model operationalizes the indicator architecture in a web-based platform that can support repeated assessment, comparison across factories and managerial interpretation. However, the empirical evidence in this study should be interpreted as demonstrating methodological validity, system implementation, and perceived usability rather than direct improvement in decision quality, sustainability planning outcomes, reporting accuracy, or resource allocation effectiveness. These outcomes remain proposed decision support applications that require future longitudinal and outcome-based validation.

This finding is in agreement with Zarte et al. [29] who argued that decision support systems can aid sustainable manufacturing by integrating product and production lifecycle decisions with sustainability assessment. It also supports Ramanujan et al. [13] who argued that Industry 4.0 technologies can contribute to manufacturing sustainability assessment not only in data acquisition but also in performance measurement, results interpretation, decision-making, intervention and validation. Similarly, Raffaeli et al. [42] showed that Industry 4.0 solutions can enhance sustainability performance by providing better control of production systems, waste reduction, reduction of raw material consumption and increased digital capability. The AI-assisted portion adds to the system’s role beyond dashboard visualization. The SIM Model does not employ AI to develop the indicators set, weight indicators, or substitute expert judgment. AI is not the scoring model but an interpretive layer above the validated scoring model. The model takes structured inputs such as normalized scores, global weights derived from AHP, weighted gap values and performance categories. In the implemented prototype, the AI module produces structured advisory outputs in JSON format, including improvement suggestions, best practices, and summary statements based on the supplied sustainability profile.

The SIM Model also has implications for Industry 5.0-oriented manufacturing. While Industry 4.0 emphasizes digitalization, data acquisition, automation, and performance monitoring [13,41], Industry 5.0 places stronger emphasis on human-centric, sustainable, and resilient industrial systems, where AI supports rather than replaces human decision-making [15]. The SIM Model supports this direction by keeping expert judgment and managerial responsibility at the center of sustainability assessment while using AI only as an advisory interpretation layer. FDT and AHP preserve human expertise in indicator validation and weighting, UVA translates heterogeneous data into interpretable sustainability scores, and the AI-DSS supports managers through structured visualization, gap identification, and advisory recommendations [14,16]. Therefore, the model contributes to Industry 5.0 by linking digital intelligence with human decision-making, sustainability performance management, and repeated organizational learning. However, these Industry 5.0 implications should be interpreted as proposed applications of the framework, because the present study evaluated perceived usability rather than direct improvements in resilience, human-centered decision quality, or sustainability outcomes.

This design is in accordance with recent findings on AI and sustainable manufacturing. Gholami [14] showed the capability of large language models (LLMs) to support decision-making applications in sustainable reconfigurable manufacturing systems. Ghobakhloo et al. [15] argued that Generative AI can be used to achieve the Industry 5.0 sustainability goals through increased analytical insight, knowledge accessibility, and decision support. Aljohani [16] showed that the AI-based ESG strategy evaluation can be enhanced by applying fuzzy multi-criteria decision-making. The SIM Model adds to this emerging literature by framing GenAI not as an autonomous decision-maker, but as a structured interpretation engine based on transparent FDT, Pareto, AHP, and utility-score outputs. That difference matters, because AI-based sustainability recommendations can be generic or difficult to verify if they rely solely on unstructured data. The SIM Model reduces this risk by providing AI with a weighted, validated sustainability profile. Therefore, the AI output is more defensible because it is constrained by indicator weights, performance scores and defined gap logic. In this context, the contribution of AI in the SIM Model is not only automation but sustainability intelligence with priorities.

3.7. Usability and Practical Applicability

From the usability evaluation with 30 industrial practitioners representing 30 manufacturing organizations, an overall System Usability Scale score of 86.0 out of 100 and an overall Satisfaction Index of 86.0% were obtained—see

Table 1. Results of System Usability Scale (SUS) evaluation for the Sustainable Industrial Measurement (SIM) Model.

Evaluation Criteria	Mean ($\overline{\mathit{x}}$)	SD	SI (%)	Interpretation
1. The system is attractive and suitable for regular use.	4.20	0.55	84.00	Very Satisfied
2. The system is simple and not complicated to use.	4.20	0.76	84.00	Very Satisfied
3. The system is easy to use and clearly understandable.	4.40	0.50	88.00	Very Satisfied
4. Users can operate the system without requiring technical support personnel.	4.00	0.64	80.00	Very Satisfied
5. The various functions of the system are well integrated and appropriately connected.	4.40	0.50	88.00	Very Satisfied
6. The system operates consistently and systematically.	4.60	0.50	92.00	Very Satisfied
7. Users can learn how to use the system very quickly.	4.20	0.41	84.00	Very Satisfied
8. The system is convenient and not cumbersome to use.	4.00	0.64	80.00	Very Satisfied
9. Users feel confident when using the system.	4.40	0.50	88.00	Very Satisfied
10. Users can easily learn and get started with the system.	4.60	0.50	92.00	Very Satisfied
Overall	4.30	0.55	86.00	Very Satisfied

. This corresponds to an excellent level of usability and indicates that the practitioners were very satisfied with the system. All ten criteria of usability were rated as “Very Satisfied”. The top two ranked criteria were system consistency and ease of learning with a mean of 4.60 and Satisfaction Index of 92.00%. The results show that users perceived the system to be predictable, coherent and easy to learn. Ease of understanding, functional integration, user confidence and ease of getting started also scored well. The lowest values were obtained for the criteria related to the functioning of the system without technical support and the convenience of use, but both remained at the level of “Very Satisfied” with the Satisfaction Index values of 80.00%.

These findings are practically relevant, as sustainability assessment systems are often methodologically rigorous but difficult for practitioners to use and thus often fail. The excellent SUS result demonstrates that complex analytical procedures such as FDT, Pareto screening, AHP weighting, utility scoring, and AI-assisted interpretation can be transferred into a usable system interface. This supports the practical feasibility of the SIM Model for manufacturing organizations. The result for usability also supports the claim that AI-DSS platforms can reduce the interpretation burden for users. Practitioners are not required to interpret all 50 indicators and their global weights manually. Instead, the system scores, visualizes, finds priority gaps and generates AI-assisted recommendations. This is in correlation with the path proposed by Ramanujan et al. [13] in which they emphasized that sustainability assessment should shift from measurement to interpretation, decision-making and intervention support. However, the SUS result should be interpreted as evidence of perceived usability and practitioner acceptance rather than direct evidence of improved decision quality. The prototype provides functions for standardized scoring, visualization, weighted gap identification, and AI-assisted recommendations; however, the effectiveness of these functions in improving sustainability planning, reporting accuracy, resource allocation, or managerial prioritization was not empirically evaluated. Accordingly, the SUS findings demonstrate perceived usability, while the actual decision support value requires future outcome-based validation.

3.8. Theoretical and Managerial Implications

The SIM Model contributes theoretically to manufacturing sustainability assessment by demonstrating how TBL indicators can be transformed into a validated, reduced, weighted, and AI-assisted assessment architecture. As summarized in

Table 2. Comparative analysis of the SIM Model against prior sustainability assessment frameworks (Yes = Full implementation; Partial = Limited implementation; No = Absent).

Evaluation Dimension	SIM Model (This Study)	Swarnakar et al. (2022) [8]	Dewi et al. (2023) [43]	Singh et al. (2016) [44]	ESG Platforms in 2026 (Persefoni/ Watershed)	Zarte et al. (2019) [29]
1. Sequential Validation Architecture (FDT + Pareto + AHP)	Yes	Partial (no Pareto)	Partial (no Pareto)	No	No	No
2. Evidence-Based Indicator Reduction (64 to 50)	Yes	No	No	No	No	No
3. Group AHP with CR Verification	Yes	Yes	Yes	No	No	Partial
4. AI-assisted interpretation of expert-validated results	Yes	No	No	No	No	No
5. Operational Web-Based DSS	Yes	No	No	Expert system only	Yes	Prototype only
6. Formal Usability Validation (SUS = 86.0)	Yes	No	No	No	No	No
7. Balanced TBL Integration (Economic + Social + Environmental)	Yes	Yes	Yes	Environmental focus	No	Partial
8. SME-Oriented Scalability and Context	Yes	No	No	Yes	Large organizations only	No

, the proposed model differs from previous sustainability assessment frameworks because it integrates all major methodological stages into a single sequential workflow, including indicator validation, evidence-based reduction, priority weighting, score calculation, system implementation, AI-assisted interpretation, and usability evaluation. This integrated structure addresses an important limitation in prior studies, where indicator identification, weighting, decision support implementation, and system validation are often treated as separate or incomplete components.

Compared with Swarnakar et al. [8], which provided an integrated approach for prioritizing sustainability indicators in manufacturing, the SIM Model extends indicator prioritization by adding Pareto-based reduction, operational web-based implementation, Generative AI interpretation, and formal usability testing. Dewi et al. [43] integrated Delphi, AHP, Sustainable Value Stream Mapping, and Traffic Light System to assess sustainable manufacturing performance in a specific industrial case. However, their framework was primarily case-based and did not incorporate an AI-assisted interpretation engine or formal usability validation. Singh et al. [44] developed a fuzzy rule-based expert system for SME sustainability assessment, showing the usefulness of expert-system logic in manufacturing contexts. Nevertheless, that approach was limited by a smaller indicator structure and did not include sequential FDT, Pareto screening, Group AHP weighting, or Generative AI-based sustainability intelligence. Zarte et al. [29] reviewed decision support systems for sustainable manufacturing and emphasized the importance of integrating economic, social, and environmental dimensions into decision-making. However, they also noted that many existing DSS approaches remain more developed at the strategic level than at the operational level. The SIM Model responds to this gap by operationalizing sustainability assessment through a web-based AI-DSS that supports scoring, visualization, diagnosis, and managerial recommendations.

The comparison also highlights the novelty of the SIM Model in the context of existing digital ESG and sustainability reporting approaches. Recently, digital technologies such as artificial intelligence, machine learning, cloud-based systems, blockchain, and structured electronic reporting formats are increasingly being used to support corporate sustainability reporting, ESG disclosure analysis, and information management [45,46]. However, the current literature still focuses more on reporting digitalization, disclosure quality, data processing and ESG communication than on a full academic sequence of expert-validated TBL indicator synthesis, Pareto-based indicator reduction, Group AHP weighting with consistency verification and SUS-based usability validation. The novelty of the SIM Model, therefore, does not lie solely in the use of AI but in how the AI is integrated with a transparent and auditable sustainability assessment methodology. In this model, AI does not replace expert judgment or weighting but interprets validated and weighted sustainability scores to highlight strengths, diagnose gaps in priorities and generate suggestions for improvement. Boonkanit and Suthiluck [47] also support the suitability of structured expert-based DSSs for sustainability decisions. They integrated the Delphi method and fuzzy AHP to select concrete waste management alternatives for a transition to a circular economy.

The finding that environmental indicators received the highest aggregate dimensional weight, while economic indicators occupied the top individual ranks, provides a nuanced interpretation of manufacturing sustainability. Environmental performance reflects strong sustainability pressure from resource consumption, emissions, waste, water use, and regulatory expectations. At the same time, economic performance remains a critical implementation enabler because firms require financial capacity, value-added production, and investment returns to support long-term sustainability improvements. This supports the TBL perspective that economic, social, and environmental dimensions are interrelated rather than independent. Similarly, Nogueira et al. [48] emphasized that the TBL paradigm integrates economic, social, and environmental dimensions in explaining business performance. Accordingly, the SIM Model should be viewed as a weighted and interdependent assessment system rather than a simple checklist of independent indicators.

From a managerial perspective, the SIM Model helps organizations prioritize sustainability issues more systematically. A low-performing indicator with a high global weight should be treated as a strategic improvement priority, whereas a low-performing indicator with a low global weight may require monitoring but not immediate resource allocation. This is more useful than unweighted dashboards, where all indicators may appear equally important. The model can support resource allocation, sustainability report preparation, internal benchmarking, factory comparison, and improvement planning. It also has implications for policy and industrial support programs. Industrial agencies and manufacturing associations can use the weighted indicator structure to design sustainability capacity-building programs that focus first on high-priority areas such as material efficiency, water intensity, energy intensity, waste circularity, return on assets, and value-added production. This is especially relevant for manufacturing organizations in developing economies, where firms often operate under budget constraints, uneven data readiness, and increasing pressure from customers, regulators, and supply-chain partners.

4. Conclusions

This study developed the Sustainable Industrial Measurement (SIM) Model as an integrated sustainability assessment architecture for manufacturing organizations under the Triple Bottom Line framework. The main scientific contribution of the study lies in linking expert-validated indicator governance, evidence-based indicator reduction, consistency-verified priority weighting, utility-based sustainability scoring, web-based DSS implementation, and AI-assisted interpretation within a single sequential framework. This integration addresses a key gap in the literature, where indicator selection, weighting, scoring, system implementation, and managerial interpretation are often treated as separate components. The main findings are summarized as follows:

(1) The proposed SIM Model provides a transparent and sequential framework for transforming broad sustainability concepts into measurable, validated, and weighted indicators. Unlike conventional assessment approaches that rely on predefined or equally weighted indicators, the SIM Model applies FDT to manage expert uncertainty, Pareto screening to reduce indicator complexity, and Group AHP to derive consistency-verified priority weights.

(2) The final indicator structure confirms that manufacturing sustainability requires an integrated Triple Bottom Line perspective. Environmental indicators received the highest aggregate dimensional weight, reflecting the importance of energy, emissions, water use, material efficiency, waste circularity, and environmental compliance. At the same time, economic indicators such as return on assets and value-added production ranked highly at the individual indicator level, indicating that economic performance remains a key enabler of sustainability implementation.

(3) The integration of the weighted indicator framework into a web-based AI-DSS strengthens the operational value of the model. The system enables organizations to convert sustainability data into standardized scores, visualize performance across dimensions, compare factories or assessment periods, and generate structured reports. This shifts sustainability assessment from static performance documentation toward decision-oriented sustainability intelligence.

(4) The Generative AI component was designed to extend the prototype beyond automated reporting by interpreting the validated and weighted sustainability score profile, identifying priority gaps, and generating structured advisory recommendations. It interprets the validated and weighted sustainability score profile to identify strengths, diagnose priority gaps, detect cross-dimensional patterns, and generate improvement recommendations. Because the AI module operates on structured inputs derived from FDT, Pareto screening, AHP weighting, and utility scoring, its outputs are grounded in transparent assessment logic rather than generic text generation.

(5) The usability evaluation demonstrated excellent perceived usability of the SIM Model. Testing with 30 industrial practitioners produced a System Usability Scale score of 86.0. This finding indicates that the prototype was perceived as user-accessible but does not confirm its effectiveness in improving decision quality, sustainability planning, reporting accuracy, resource allocation, or organizational outcomes.

Therefore, the main scientific contribution of this study lies in the development of an integrated methodological and operational architecture for sustainable manufacturing assessment. The SIM Model advances the literature by linking expert-validated TBL indicator construction, evidence-based indicator reduction, consistency-verified AHP weighting, Utility Value Analysis, web-based DSS implementation, and AI-assisted interpretation within a single sequential framework. This integration addresses key limitations of prior sustainability assessment approaches that often treat indicator selection, weighting, scoring, system implementation, and managerial interpretation as separate components. The present evidence directly supports the methodological validity of the indicator framework, the feasibility of system implementation, and the perceived usability of the platform among practitioners. However, the model’s effects on decision quality, recommendation accuracy, resource allocation, and managerial outcomes were not empirically tested and should be examined in future outcome-based studies.

There are limitations of this study. First, the expert judgments and usability evaluations were performed in a specific manufacturing and institutional context. The resulting indicator weights may therefore reflect the priorities of the participating expert group and may need further validation in other industries, countries and firm sizes. In addition, the current Group AHP procedure used equal response weighting when aggregating expert judgments. Therefore, the sensitivity of criteria rankings to alternative stakeholder assumptions was not examined. Future research should conduct sensitivity analysis using three expert response-weighting schemes: (i) equal weighting, (ii) experience-based weighting based on experts’ professional or research experience, and (iii) familiarity-based weighting based on experts’ familiarity with the evaluated sustainability topics [26]. Such analysis would help assess whether the indicator rankings remain stable under different expert-weighting assumptions and would strengthen the robustness and generalizability of the SIM framework. Second, the AHP structure presupposes hierarchical relationships among indicators, while in practice some sustainability indicators may be interdependent. Third, Pareto screening was applied to produce a manageable operational indicator set. Although representation across all three TBL dimensions was retained and alternative thresholds were discussed, the selected cut-off may affect context-sensitive lower-ranked indicators. The complete indicator catalog should remain available for sector-specific materiality review. Finally, the system evaluation focused mainly on perceived usability rather than demonstrated decision support outcomes. Although the SUS score indicates strong practitioner acceptance, this study did not directly measure whether the SIM Model improves decision quality, sustainability planning effectiveness, reporting accuracy, resource allocation, or managerial prioritization. Future research should conduct longitudinal multi-factory implementation, expert assessment of generated reports and recommendations, and before–after comparison of managerial decisions to validate the actual decision support value of the system.