Strategic Roadmap for Adopting Data-Driven Proactive Measures in Solar Logistics

Abstract

This study presents a comprehensive overview of the solar industry’s transition towards resilient energy solutions, emphasizing the critical role of data-driven practices in driving this transition through responsible resource management. As continuous technological refinement is essential to optimize solar energy’s potential, the smart use of available data plays a significant part in enhancing the accessibility of solar panels. Building upon prior research investigating the influence of Big Data on solar logistics, this paper proposes a hybrid Multi-Criteria Decision-Making (MCDM) methodology based on expert experience, providing practical support in the implementation of data-driven proactive measures within the solar industry. Specifically, this study focuses on measures aimed at effectively implementing two main logistic strategies, which are Route Optimization (RO) and Warehouse Management (WM). A rigorous analysis of criteria and measures considered to be relevant in the literature is first conducted. Criteria will be screened and weighted to eventually act as drivers toward measure assessment and prioritization. A final sensitivity analysis culminates in the formalization of findings and in the formulation of a pragmatic roadmap tailored for solar industry practitioners, designed to increase operational efficiency while integrating key sustainability principles across supply chain endeavors.

1. Introduction and Goals of Research

The solar industry revolves around leveraging solar radiation to produce electricity, a critical endeavor in today’s context as we seek to transition towards environmentally resilient energy solutions [1]. Central to this objective is the principle of sustainability, which emphasizes the responsible management of resources to safeguard the ecological integrity of our planet. Solar power emerges as a prime exemplar of sustainability owing to its renewable character, affording continuous utilization without depletion. However, continuous innovation in supply chain practices along with advancements in manufacturing technology is indispensable to enhance the management of solar energy utilization [2]. The integration of cutting-edge equipment plays a major role in augmenting technical and logistical efficiency as well as the cost-effectiveness of solar panel systems, therefore enhancing their accessibility and scalability for a broader demographic [3].

In shifting our focus to the supply chain realm, we examine the integration of the manufacturing stage with distribution and eventual consumption by end users to establish an organic supply chain for solar panels, with a primary focus on supply chain optimization. This process is characterized by numerous interdependent components, requiring the coordination of various stakeholders, including manufacturers responsible for production and logistics entities charged with packaging and transporting the panels to their destinations. Throughout this continuum, copious amounts of data are generated related to production volumes, distribution logistics, and key performance indicators. Effectively analyzing this data is crucial to ensuring operational efficiency, something that highlights the significance of supply chain optimization. The strategic deployment of suitable methodologies is required, aiming at expediting the delivery of solar panels to individuals and organizations in need, all while upholding our commitment to environmental stewardship.

This research builds on a previous conference paper [4] that investigated the impact of Big Data on logistics operations in manufacturing. Specifically, Route Optimization (RO) followed by Warehouse Management (WM) were identified as the preferred logistic strategies in the practical case of a solar panel company. The present paper aims to expand this perspective by proposing a flexible hybrid Multi-Criteria Decision-Making (MCDM) approach making use of the DEcision-MAking Trial and Evaluation Laboratory (DEMATEL) and the Elimination et Choix Traduisant la Réalité (ELECTRE) TRI. We begin by examining criteria formalized from the current literature and deemed pertinent to our field of application. These criteria will be weighted using DEMATEL, accounting for their interrelations. Subsequently, the proactive implementation of data-driven measures will be outlined for the preferred strategies mentioned above, namely route optimization and warehouse management. ELECTRE TRI will then assign these measures to classes representing the level of priority suggested for their implementation. Lastly, a sensitivity analysis will be conducted by varying criteria weights, suggesting specific measures based on specific criteria and providing a practical roadmap for practitioners in the solar industry.

The paper is structured as follows. Section 2 provides a thorough literature review, examining key supply chain issues in the solar industry and discussing technology management within the field, along with existing MCDM approaches currently adopted to solve similar problems. Section 3 justifies our methodological choice by explaining the usefulness of applying the mentioned techniques and reporting relevant methodological details. In Section 4, a practical case study in the solar industry is described and addressed, with Section 5 presenting the corresponding findings. Section 6 presents the study’s conclusions and analyzes potential directions of future research. The structure of the paper is also reported in Figure 1 to summarize the logical flow of this research.

2. Literature Review

2.1. Crucial Supply-Chain Issues in the Solar Industry

The solar sector stands as a cornerstone of efficient and sustainable energy solutions, providing a viable alternative to traditional energy sources. Nonetheless, the journey from procurement of raw materials to the delivery of final products encounters several challenges in supply chain management. Such issue as as production disruptions, distribution bottlenecks, and installation obstacles highlight the complex nature of this industry’s logistical landscape. This subsection aims to discuss some of the most significant challenges inherent in the solar energy supply chain.

Chadly et al. [5] led a comprehensive examination of the solar energy supply chain, revealing dependence on the Chinese market across various stages, from sourcing raw materials to delivering final products while managing recycling capacities. The authors discuss the key challenges confronting the solar energy sector, such as trade barriers and potential raw material shortages, by also highlighting opportunities for innovation and improvement within the industry’s dynamic landscape. These may include exploring alternative supply sources, embracing technological advancements, and promoting recycling and circular economy practices. Challenges related to the recycling, refurbishment, and re-certification of photovoltaic (PV) modules were discussed by Tsanakas et al. [6] within the context of solar logistics. They identified key obstacles, including technological and operational complexities in PV recycling, limited repair and refurbishment options for failed PV modules, and the absence of standardized testing protocols. In this context, it is worth mentioning that the industry’s rapid growth and the high cost associated with recycling technologies were the primary reasons identified for the mentioned challenges. This directly influences supply chain dynamics as the mentioned factors can disrupt the smooth flow of materials and resources throughout the supply chain, leading to logistical challenges and increased operational complexities. To address these issues, the authors proposed a circular business model that included more systematic recycling, repair, and refurbishment practices, complemented by the development of new supply chain standards and certifications. In this context, Nyffenegger et al. [7] underlined that, despite the need for change, adopting circular economy practices across the solar supply chain is still in its early stages. The solar industry is indeed facing various challenges, such as resource depletion and increasing waste during the disposal of solar panels. Specifically, the authors conducted interviews with key stakeholders to understand current goals, obstacles, and drivers for this transition. Based on their findings, they developed a roadmap outlining key steps towards a circular economy, from sourcing materials to recycling. They also introduced a framework suggesting strategic approaches for various stages of the supply chain, emphasizing the importance of collaboration to drive change. The authors concluded their work by recommending further research to quantify these findings and explore innovative business models to overcome economic barriers.

2.2. Technology Management in the Field of Study

This subsection focuses on the use of the latest supply chain technologies in the solar industry and aims to address key challenges in the renewable energy sector. By using advanced tools, companies can enhance process efficiency and sustainability across the entire supply chain, transitioning towards a cleaner and more sustainable energy future.

Nguyen et al. [8] explored how Machine Learning (ML) and Artificial Intelligence (AI) algorithms can transform the renewable energy industry, particularly in biomass, biofuels, engines, and solar power. In the solar energy field, ML algorithms analyzed vast amounts of solar data to improve system design, operations, and maintenance, contributing to supporting a more sustainable energy ecosystem. These advanced technologies enable more efficient management of the entire supply chain process for several reasons. First, by optimizing system design and operations, these algorithms can enhance the efficiency of solar panel manufacturing processes, leading to reduced production costs and improved product quality. This directly impacts the supply chain by streamlining the production process and ensuring that the right quantity of solar panels is available when needed. Second, predictive maintenance enabled by these technologies can minimize downtime and ensure the reliability of solar energy systems. This is crucial for supply chain management as it helps to prevent disruptions in the delivery of solar energy products and services to customers. Furthermore, the use of ML and AI-based algorithms can optimize inventory management by accurately forecasting demand and ensuring that inventory levels are optimized to meet customer needs while minimizing excess stock.

Ahmad et al. [9] studied the role of AI techniques in the energy sector, focusing on three major aspects: AI in solar and hydrogen power generation, AI in supply and demand management control, and recent advances in AI technology. Their research highlighted that AI techniques widely outperform traditional models in various areas, being crucial for enhancing operational performance in the data-centric energy industry. In this context, the utilization of AI in supply and demand management control underscores its potential to optimize supply chain processes, improving forecasting accuracy, inventory management, and resource allocation. The increasing role of AI was examined by Ahmad et al. [10] also regarding energetic systems. Their study highlighted four key approaches for advancing AI-based energy systems, namely fuzzy logic systems, artificial neural networks, genetic algorithms, and expert systems. This research demonstrates the potential of AI technologies to not only enhance transparency in the utilization of renewable energies in the power industry but also to optimize various aspects of supply chain management, including resource allocation, demand forecasting, and operational decision-making. Bag et al. [11] identified various Industry 4.0 enablers of supply chain sustainability, implementing a framework that offered insights for mitigating negative effects and enhancing benefits in smart production processes. Nia et al. [12] reviewed publications related to energy demand, focusing on the influence of Industry 4.0. By analyzing existing literature, forecasting methods were classified into traditional and intelligent methods. The authors also discussed the advantages and disadvantages of each method, highlighting the potential benefits of using intelligent forecasting methods in reducing errors and costs, as well as increasing profitability. Khanfar et al. [13] discussed how blockchain technology can address challenges faced by manufacturers in implementing best practices along the supply chain. Their research examined the potential contributions of blockchain to the economic, environmental, and social performance of manufacturers and their supply chains.

Considering the growing installation of solar panels, Acharya et al. [14] addressed the need for an effective recycling plan for solar PV modules. They introduced an optimization framework for reverse logistics to collect End-of-Life (EoL) solar PV modules in Delhi. Their study formulated a Convex Mix Integer model with a quadratic objective and constraints, using the McCormick envelope and CPLEX solver for optimization. Results suggested that centralized and optimally decentralized collection plans were more profitable than other approaches. Focusing on addressing sustainability concerns in the supply chain network design of photovoltaic systems, Nili et al. [15] deployed a multi-objective, mixed-integer non-linear optimization model. The aim was to design a supply chain network that minimizes costs, environmental impacts, and social values while also incorporating reverse flow to address environmental and social issues. The augmented $ϵ$-constraint method was used to optimize the model and ensure Pareto optimal solutions. The effectiveness of the study was validated through a real-world case study in Iran.

A decision-making framework for sustainable supply chain development in the renewable energy field, with a PV focus, was developed by Mastrocinque et al. [16]. The proposed framework included the whole energy production supply chain, from raw materials suppliers to disposal, and it was applied to assess the photovoltaic sectors of seven European countries, providing insights for sustainable investment decisions. Lotfi et al. [17] explored the integration of renewable energy into supply chain network design to enhance sustainability. The authors introduced a two-stage robust stochastic model to locate facilities and determine flow quantities within the supply chain. Using a GAMS-CPLEX solver, this study analyzed the effects of conservative coefficients and demand variations on cost functions. Masoomi et al. [18] addressed supplier selection problems in renewable supply chain management by using an integrated Fuzzy Best–Worst Method (FBWM) along with two techniques, which are the Complex Proportional Assessment of Alternatives (COPRAS) and the Weighted Aggregated Sum–Product Assessment (WASPAS). Nine strategic supplier selection criteria were identified, and a real-world example from Iran’s renewable energy supply chain was studied to demonstrate the effectiveness of the proposed framework.

2.3. Existing MCDM Approaches Adopted for Similar Problems

To understand the effectiveness of MCDM in the context of the present research, this subsection will discuss their implementation in the renewable energy sector by also providing a specific focus on the solar energy field.

Sahoo and Goswami [19] reviewed various applications of MCDM methods, examining advancements, strengths, and limitations. Their work provided a comprehensive understanding of the complexity of decision-making processes and guidance on selecting appropriate MCDM techniques according to the specificity of the real scenario, driving sustainable development across different domains. Javaid et al. [20] discussed the importance of adopting modern technologies in the manufacturing field by identifying barriers along with their respective solutions. On the basis of experts’ consultation, the authors first ranked the various identified barriers using the Best–Worst Method (BWM) and second evaluated solutions using the Combined Compromise Solution (CoCoSo) method. Various modeling techniques such as Interpretive Structural Modeling (ISM), Total Interpretive Structural Modeling (TISM), and DEMATEL were also implemented to determine the interrelationships among the identified barriers, developing a thorough analysis of the challenges and opportunities related to modern technology adoption in manufacturing. Paul et al. [21] explored various MCDM methods applied within Sustainable Supply-Chain Management (SSCM). Key methods discussed include DEMATEL and its variants, such as Fuzzy and Grey DEMATEL, the AHP and its fuzzy counterpart for complex decision-making, as well as the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and Fuzzy TOPSIS. Additionally, methods like ELECTRE and the Analytical Network Process (ANP) were highlighted for their roles in addressing conflicting criteria and interdependencies within supply chains.

In the renewable energy sector, Gribiss et al. [22] presented 16 different configurations for energy self-consumption in Renewable Energy Communities (RECs) containing different industrial factories. A mathematical model for each configuration was proposed, and various MCDM-based approaches were employed, namely Weighted Sum (WS), Weighted Product (WP), TOPSIS, and Evaluation based on Distance from Average Solution (EDAS). This study showed the effectiveness of configurations that combine both individual and collective energy self-consumption in RECs. Ramezanzade et al. [23] deployed a new hybrid fuzzy MCDM model aimed at selecting the best renewable energy projects at a sub-national level. The proposed model combined such methods as VIKOR (an acronym for “VIsekriterijumska KOmpromisno Rangiranje” in Serbian, which translates to “Multi-criteria Optimization and Compromise Solution” in English), distance from average solution, and additive ratio assessment. The results identified solar energy as the most suitable renewable energy option, with Jajarm identified as the optimal location for project implementation. In addition, sensitivity analysis demonstrated the robustness of the model. Ali [24] integrated the TOPSIS methodology with triangular neutrosophic sets to rank and select the best source of renewable energy in Egypt.

Saraswat et al. [25] assessed the spatial suitability of solar and wind farm locations in India from technical, economic, and socio-environmental perspectives. They integrated the Geographical Information System (GIS) with MCDM approaches. The proposed model served as a tool for policymakers regarding renewable energy resources and assessed the suitability of existing projects. Almasad et al. [26] integrated the Fuzzy AHP and the Preference Ranking Organization Method for Enrichment Evaluations (PROMETHEE) II to assess the suitability of sites for implementing solar PV projects in Saudi Arabia. The research contemplated 12 factors combined into two main criteria (technical and economic) to minimize construction costs while maximizing power output. The research concluded that Saudi Arabia exhibits substantial potential for solar PV projects, covering approximately 65.1% of the total studied area. Kannan et al. [27] addressed a solar site selection problem in Eastern Iran using MCDM. Specifically, they combined the BWM, the Grey Relational Analysis (GRA), and VIKOR to rank potentially suitable locations. Furthermore, an innovative Monte Carlo Simulation-Based (MCSB) approach was introduced to analyze the sensitivity of results produced from VIKOR and GRA, confirming the robustness of the first one over the second one. Bączkiewicz et al. [28] developed an innovative approach to select solar panels using newly developed MCDM methods, such as Characteristic Objects Method (COMET) combined with TOPSIS and Stable Preference Ordering Towards Ideal Solution (SPOTIS). The methods used in the study were rank-reversal-free and precise, integrating objective entropy weighting to determine the weights of the criteria. The final results demonstrated the effectiveness of the proposed approach in solving the problem of solar panel selection. Another widely used decision-making method is ELECTRE TRI, which is employed to sort alternatives into categories based on multiple criteria. In the solar energy field, focusing on PVs, Thebault et al. [29] applied the ELECTRE TRI methodology to assess urban roofs for their suitability to install PV systems. This evaluation incorporated multiple criteria, such as energy potential, economic viability, and structural conditions. The approach enabled them to identify the roofs most suitable for solar installations, helping in urban planning for solar energy deployment. Singh and Powar [30] implemented a decision-making framework to assess the performance and select the location of solar PV plants in India. They utilized Fuzzy ELECTRE TRI to compare seven PV plants by taking into account uncertainties in the data, such as environmental variability connectivity and accessibility. This approach enhanced the robustness and reliability of their evaluations.

3. Proposed Methodological Approach

As emerged from the previous literature review, MCDM integration is a commonly adopted practice to support decision-making in the renewable energy sector with a particular focus on the solar energy field and is evidently considered a strategic approach in contemporary research. However, to the best of the authors’ knowledge, despite having been successfully implemented in other application fields [31], an integration between DEMATEL and ELECTRE TRI has not been proposed before for the implementation of data-driven measures aimed at optimizing supply chain management in the solar industry. We aim to address this gap by providing a framework capable of offering meaningful advantages when dealing with complex decision-making problems, as explained in the following.

• DEMATEL helps in understanding the interrelationships among criteria, providing insights into the structure of the decision problem [32]. This understanding enhances the formulation of criteria weights, which in turn improves ELECTRE TRI accuracy.
• The hybrid approach enables a more robust sensitivity analysis by considering both causal relationships among criteria (from DEMATEL) and the trade-offs between alternatives (from ELECTRE TRI) [31]. This comprehensive analysis enhances decision-makers’ confidence in the chosen solution.
• Integrating DEMATEL with ELECTRE TRI enhances decision transparency by offering a structured framework for identifying influential criteria [33] and assessing the performance of a great volume of alternatives [34]. This stimulates stakeholder confidence and facilitates effective communication throughout the decision-making process.
• Compared to integrating other MCDM methods, combining DEMATEL and ELECTRE TRI may offer a deeper coverage of decision aspects. DEMATEL captures the causal relationships among criteria [32], providing a comprehensive understanding of the decision structure, while ELECTRE TRI evaluates alternatives based on outranking relations [31]. This dual perspective ensures a thorough exploration of the decision space, leading to well-informed and robust decisions.

The next subsections briefly provide relevant methodological details.

3.1. DEMATEL

We propose the use of the DEMATEL method to study the interrelations among the arrays of criteria under examination. This choice is primarily rooted in DEMATEL’s unique capability to explore causal patterns and provide decision-makers with a structured comprehension of the problem. Unlike conventional MCDM approaches, DEMATEL enables us to pinpoint the cause-and-effect associations among variables, therefore extracting meaningful insights into the fundamental origins of hidden connections and their ramifications on the overarching system.

By visualizing and measuring these connections, DEMATEL can facilitate more discerning and effective decision-making processes. In contrast to alternative techniques, DEMATEL indeed holds a distinctive edge in its capacity to dissect latent connections within the system, and this capability proves invaluable in formulating well-informed and targeted strategies for enhancement. The primary methodological steps needed for employing DEMATEL are reported next [35].

• Gathering input data from experts regarding the cause-and-effect relationships among n factors, compared pairwise. These relationships are typically described using linguistic variables. The attributed qualitative descriptions of influence will be translated into numerical values using the following scale: 0 (No Influence, NO), 1 (Very Low Influence, VL), 2 (Low Influence, L), 3 (Medium Influence, M), 4 (High Influence, H), 5 (Very High Influence, VH). If multiple experts are involved in the data collection process, each expert should generate a squared $n\times n$ matrix. These matrices should then be combined into a single squared input matrix (referred to as the direct-relation matrix A) before proceeding to the next phase.
• Normalizing the direct-relation matrix, achieving matrix $D=s\times A$. This normalization is performed by computing the value s, as follows:

  $$s=min\left[\frac{1}{ma{x}_{1\le i\le n}{\sum }_{j=1}^n{a}_{ij}},\frac{1}{ma{x}_{1\le j\le n}{\sum }_{i=1}^n{a}_{ij}}\right]$$

  (1)
• Obtaining the total relation matrix T. Such a calculation will take into account the identity matrix I. In particular, it will involve the multiplication between matrix D and the inverse of the difference $I-D$, activating an iteration process for which the generated matrix T will include direct and indirect effects related to the dataset under analysis:

  $$T=D\times {(I-D)}^{-1}.$$

  (2)
• Generating the causal relation map by analyzing the values within matrix T and establishing the most influential elements to be differentiated based on their prominence and relation values. Prominence and relation values are computed, respectively, as $r_i+c_i$ and $r_i-c_i$, where $r_i$ and $c_i$ are $n\times 1$ and $1\times n$ vectors, denoting the sum of rows and columns of matrix T. In the causal relationship map, factors exhibiting higher prominence values are those with the greatest impact on the problem being examined. Moreover, factors with positive relationship values are deemed net contributors, while those with negative relationships are seen as net recipients [36]. The causal relationship map functions as a visual aid to scrutinize and depict the causal connections among the factors, facilitating a clear differentiation between the most influential and significant elements.
• Calculating the final vector of criteria weights by normalizing the corresponding prominence values.

3.2. ELECTRE TRI

After weighting criteria via DEMATEL, we propose the use of the ELECTRE TRI method to sort alternatives (measures) into ordered classes, each class reflecting a different level of priority suggested for the implementation of measures. ELECTRE TRI can be implemented by means of two phases [37]. The first phase aims to build outranking relations between pairs of alternatives on the basis of concordance and discordance criteria. These relations are manipulated during the second phase of the procedure, aiming at sorting alternatives to different classes according to their shared characteristics. The mentioned sorting procedure is developed by means of two different approaches that will be discussed later, which are a pessimistic assignment and an optimistic assignment. We begin by collecting the input data required for the methodological implementation: criteria $B_k$ or aspects mostly relevant to the decision-making problem under analysis; criteria weights $w_k$; reference profiles delimiting classes $P_j$, each profile being defined by two limits $p_h$ and $p_{h+1}$; classes $C_h$ where alternatives will be sorted; alternatives $A_i$; quantitative evaluations of alternatives under criteria $B_k\left(A_i\right)$; threshold value, or cutting level, $\lambda$ ranged within $[0.5,1]$; thresholds of indifference $I_k$, strong preference $S_k$ and veto $V_k$, to be defined to characterize the specific outranking relations among alternatives. These three thresholds need to be set for each criterion by implementing a trial-and-error process, where the analyst makes attempts to determine the most appropriate numerical values [31]. $I_k$ stands for the minimum difference necessary to indicate preference between a pair of alternatives under a given criterion, $S_k$ signifies the minimum difference required to assert strong preference between a pair of alternatives, and $V_k$ denotes the minimum difference highlighting an incompatibility relation between a pair of elements [38].

We report in the following a brief recall about how to implement the steps for carrying out the first phase of the ELECTRE TRI procedure.

• Computing, for each criterion $B_k$, suitable partial concordance indices by pairwise comparing each alternative $A_i$ with reference profiles $P_j$. Concordance indices $C_k(A_i,P_j)$ are obtained as follows:

  $$C_k(A_i,P_j)=\left\{\begin{array}{c}1if[B_k\left(P_j\right)-B_k\left(A_i\right)]\le I_k\hfill \\ 0if[B_k\left(P_j\right)-B_k\left(A_i\right)]\ge S_k.\hfill \\ \frac{B_k\left(A_i\right)-B_k\left(P_j\right)+S_k}{S_k-I_k}\mathrm{otherwise}\hfill \end{array}\right.$$

  (3)
  Once partial concordance indices have been obtained for each criterion, they are merged into the aggregated concordance index $C(A_i,P_j)$ via the following equation:

  $$C(A_i,P_j)=\frac{{\sum }_{k=1}^K{w}_k·C_k(A_i,P_j)}{{\sum }_{k=1}^K{w}_k}.$$

  (4)
• Computing, for each criterion $B_k$, suitable partial discordance indices by pairwise comparing each alternative $A_i$ with reference profiles $P_j$. Discordance indices $D_k(A_i,P_j)$ are obtained as follows:

  $$D_k(A_i,P_j)=\left\{\begin{array}{c}1if[B_k\left(P_j\right)-B_k\left(A_i\right)]>V_k\hfill \\ 0if[B_k\left(P_j\right)-B_k\left(A_i\right)]\le S_k.\hfill \\ \frac{B_k\left(P_j\right)-B_k\left(A_i\right)-S_k}{V_k-S_k}\mathrm{otherwise}\hfill \end{array}\right.$$

  (5)
• Computing the outranking credibility index, which assesses the credibility of outranking relationships between alternatives. This index measures the extent to which an alternative outranks another in terms of meeting the criteria specified in the decision-making process. It can be calculated by implementing the following formula:

  $$\delta (A_i,P_j)=C(A_i,P_j)·\frac{{\prod }_{k\in K^{*}}(1-D_k(A_i,P_j))}{1-C(A_i,P_j)},$$

  (6)

  where $K^{*}$ is the subset of criteria for which $D_k(A_i,P_j)>C(A_i,P_j)$. We specify that $\delta (A_i,P_j)$ is assumed as equal to the aggregated concordance index $C(A_i,P_j)$ When no veto threshold has been established in the stage of input data collection, implying that no specific criterion has been set to denote a level of incompatibility beyond which an alternative is entirely unacceptable. The absence of a veto threshold must be contemplated by the analyst as it means that all the alternatives can be considered potentially viable, even if they do not fully meet certain criteria. This approach allows for a more inclusive consideration of alternatives and can prevent premature elimination of potentially valuable options. It can be useful in various scenarios, e.g., when decision-makers seek flexibility, face uncertainty in preferences, need to conduct exploratory analysis or aim to reduce computational complexity [39].
• Deciding on a particular type of outranking relation using the cutting level $\lambda$. Essentially, $\lambda$ serves as the threshold for $\delta (A_i,P_j)$ to confirm that $A_i$ outranks reference profile $P_j$ towards the final assignment to a specific class. The value of $\lambda$ falls between 0.5 and 1, as specified before, and must exceed the quantity equivalent to $1-(\mathit{highest} \mathit{weight}/\mathit{total} \mathit{weight})$ [40,41]. This step concludes the application of the first phase of ELECTRE TRI.

Once the first phase is concluded, we will briefly explain how to carry out the second phase of the ELECTRE TRI procedure. This phase aims to produce the final assignment of alternatives to classes, which can be performed by means of two distinct procedures, namely the pessimistic and the optimistic procedures.

The pessimistic (or conjunctive) procedure determines the classification of alternative $A_i$ by verifying the condition where $A_i$ outranks $P_j$ to assign it to class $C_h$. This procedure consists of two steps:

• successively comparing each alternative $A_i$ with the profile limits of classes until the condition that $A_i$ outrank $P_j$ is met;
• assigning alternative $A_i$ to class $C_{h+1}$.

The pessimistic procedure is generally preferred over the optimistic one, as it is grounded in the desire for more conservative results. By favoring results obtained via the pessimistic procedure, the method tends to assign alternatives to classes defined by lower profiles. This means that alternatives are categorized based on stricter criteria, which increases the likelihood of selecting options that perform well across all criteria. This approach is particularly valuable in decision-making contexts where risk aversion or the desire to avoid overly optimistic assessments is important [42].

The optimistic (or disjunctive) procedure determines the classification of alternative $A_i$ by verifying the condition where $P_j$ outranks $A_i$ to assign it to class $C_h$. This procedure consists of two steps:

• successively comparing each alternative $A_i$ with the profile limits of classes until the condition that $P_j$ outrank $A_i$ is satisfied;
• assigning alternative $A_i$ to class $C_h$.

The optimistic procedure might be favored over the pessimistic one when decision-makers prioritize exploring the upper bounds of potential outcomes or when they are more risk-tolerant. Additionally, in contexts where the consequences of errors are relatively low, the optimistic approach allows for a broader consideration of alternatives based on their best possible performance [43].

For the above-mentioned reasons, in the next case study, we will prioritize results obtained using the pessimistic procedure.

4. Case Study

4.1. Problem Setting

This case study provides an in-depth analysis of a company previously analyzed in [4], with the goal of tailoring effective recommendations on specific measures to be implemented within the main logistic strategies suggested in the previous work. The company operates in the renewable energy sector, specializing in the production and marketing of high-quality solar panels. The manufacturing facility employs cutting-edge technologies, including ribbon-less technology, aiming at surpassing traditional manufacturing methods. This innovative approach allows the company to enhance process efficiency while ensuring greater reliability in the product portfolio. Apart from technological advancements, a significant commitment to meeting market demands emerges by analyzing the adaptable approach to product development. Indeed, continuously assessing industry trends and customer preferences makes it easier for the company to tailor the product offer accordingly. This customer-centric focus has been crucial in establishing and maintaining an effective competitive edge.

Moreover, the company invested significantly in refining its traditional approaches to production and logistics. By prioritizing the use of superior-grade materials and maintaining a heavily automated manufacturing plant, high-quality products need to be consistently delivered within designated timelines. However, logistical challenges can arise unexpectedly, requiring the implementation of proactive measures. Consequently, the company invests significant resources in fine-tuning its logistical strategies to ensure a consistent and adaptable flow of modules over time. This can be achieved by planning and coordinating to manage inventory effectively as well as responding promptly to fluctuations in demand.

Several strategies were analyzed in [4], aiming at optimizing data-driven supply chain processes. Among these, two key strategies emerged: route optimization and warehouse management. Route optimization refers to analyzing extensive datasets, including traffic patterns, weather conditions, and delivery schedules. By processing these data, transportation fleets optimize their routes, therefore reducing transportation costs and enhancing delivery efficiency. Essentially, this strategy aims to streamline the transportation process, ensuring goods reach their destination in a timely and cost-effective manner. On the other hand, warehouse management mainly focuses on evaluating warehouse layout and order volumes. By assessing these factors, the company optimizes various aspects of warehouse operations, including storage, picking, and packing processes. This optimization leads to improved management of procedures within the warehouse, resulting in reduced costs and smoother operations overall. It appears evident that prioritizing route optimization and warehouse management can significantly enhance logistic operations on the whole, benefiting the company’s bottom line and customer satisfaction.

In the present case study, we propose a roadmap for the gradual adoption of these two strategies through the prioritization of related proactive data-driven measures. This prioritization process will be grounded in meaningful criteria derived from aspects considered as pertinent to the business sector of reference. These criteria were established through a comprehensive analysis of existing literature, resulting in the identification of ten key aspects, comprehensively described in

Table 1. Relevant criteria emerged from the literature.

ID	Criteria	Description
${\mathrm{B}}_1$	Optimization of lead-time	The ability to reduce the duration between placing an order for a product and its delivery to the destination [44].
${\mathrm{B}}_2$	Satisfaction level of Customer	This refers to how satisfied customers are with logistics services, including whether they receive their deliveries on time, the quality of the service, and how well it performs overall [45].
${\mathrm{B}}_3$	Dependability	Consistently delivering logistics services, making sure that deliveries happen on time and keeping the promises made to customers [46].
${\mathrm{B}}_4$	Responsiveness	Being prepared and capable of promptly handling and resolving customer questions or complaints [44].
${\mathrm{B}}_5$	Affordability	The ability to offer logistics services at a fair price compared to the quality of service provided [47].
${\mathrm{B}}_6$	Quality	Ensuring that customers’ goods stay safe and do not become damaged, spoiled, or lost while being transported and handled [48].
${\mathrm{B}}_7$	Sustainability	Reducing the impact of logistics operations on the environment, including factors like carbon emissions, waste, water use, air pollution [49].
${\mathrm{B}}_8$	Supply-Chain Visibility	The extent to which people within the supply chain have access to the timely and accurate information that they consider to be key or useful to their operations [50].
${\mathrm{B}}_9$	Supplier Performance Monitoring	The ability to access supplier performance metrics that will help in supplier selection decisions and facilitate strategic partnerships with reliable suppliers [51].
${\mathrm{B}}_{10}$	Reverse Logistics Optimization	Efficiently implementing reverse logistics, including product returns, repairs, and recycling to analyze return patterns and customer feedback [52].

. As specified before, this research is a considerable extension of a conference paper [4]. While the previous work analyzed seven out of ten criteria outlined in Table 1, our current research has substantially developed upon this foundation. We have updated the nomenclature and descriptions of these criteria but have also led a more exhaustive literature review tailored to our specific application domain. Of particular significance is the incorporation of additional key aspects, such as supply chain visibility, supplier performance monitoring, and reverse logistics optimization, which were absent in the previous iteration.

Criteria are interdependent with each other, and understanding their varying importance is crucial in the prioritization process of data-driven measures. Each criterion may exert a different level of influence on the effectiveness of the measures being implemented and, ultimately, on the success of the corresponding strategies on the whole. This aspect will be evaluated in the following subsection, where the DEMATEL procedure will be applied to achieve the vector of criteria weights.

4.2. DEMATEL Weighting Criteria

As previously touched, certain criteria may directly impact the feasibility and practicality of implementing measures connected to specific strategies. For instance, criteria such as “Optimization of lead-time” (${\mathrm{B}}_1$) and “Responsiveness” (${\mathrm{B}}_4$) directly address the agility of logistics operations. These factors are critical for ensuring timely deliveries and responsive customer service, which are essential for maintaining competitiveness in the market. On the other hand, criteria like “Sustainability” (${\mathrm{B}}_7$) and “Quality” (${\mathrm{B}}_6$) emphasize broader considerations beyond immediate operational concerns. While ensuring the reliability and safety of logistics operations is of paramount importance, the company also aims to address environmental impacts and uphold product quality standards to maintain long-term viability and reputation. Furthermore, criteria such as “Supply-Chain Visibility” (${\mathrm{B}}_8$) and “Supplier Performance Monitoring” (${\mathrm{B}}_9$) underline the importance of transparency and collaboration within the supply chain. Access to timely and accurate information, as well as the ability to evaluate supplier performance, are essential for promoting strategic partnerships and optimizing supply chain efficiency. By evaluating the mutual weights of these criteria via DEMATEL, the company can rely on a structured approach based on expert opinions in prioritizing data-driven measures addressing the most pressing needs and challenges within logistical operations.

A complex brainstorming process was organized with the company management, aiming at collecting input values reflecting how the criteria reported in Table 1 influence each other. This enabled us to formalize the input matrix reported in

Table 2. Input matrix reporting linguistic evaluations for DEMATEL application.

	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$
${\mathrm{B}}_1$	NO	H	M	VL	M	NO	VL	L	M	VL
${\mathrm{B}}_2$	L	NO	VL	M	H	H	L	NO	VL	L
${\mathrm{B}}_3$	H	VH	NO	M	M	H	L	VL	NO	H
${\mathrm{B}}_4$	VL	VH	M	NO	L	H	VL	M	L	M
${\mathrm{B}}_5$	H	H	M	L	NO	M	M	VL	NO	VL
${\mathrm{B}}_6$	M	VH	H	M	M	NO	L	L	L	M
${\mathrm{B}}_7$	L	VL	M	VL	M	L	NO	NO	VL	M
${\mathrm{B}}_8$	VH	H	M	M	VL	L	M	NO	H	M
${\mathrm{B}}_9$	M	M	L	H	VL	M	L	H	NO	VL
${\mathrm{B}}_{10}$	NO	H	L	L	NO	L	VL	M	M	NO

, which collects linguistic evaluations of influence, successively quantitatively translated as described in the first point of Section 3.1.

Table 3. Total relation matrix, values of prominence ($r_i+c_i$), relation ($r_i-c_i$) and criteria weights.

	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$	${\mathit{r}}_{\mathit{i}}+{\mathit{c}}_{\mathit{i}}$	${\mathit{r}}_{\mathit{i}}-{\mathit{c}}_{\mathit{i}}$	Weights
${\mathrm{B}}_1$	0.0991	0.2422	0.1707	0.1204	0.1683	0.1059	0.0999	0.1144	0.1395	0.1113	3.1420	−0.3986	0.0952
${\mathrm{B}}_2$	0.1509	0.1486	0.1314	0.1713	0.1986	0.2099	0.1264	0.0674	0.0911	0.1425	4.0353	−1.1593	0.1222
${\mathrm{B}}_3$	0.2300	0.3249	0.1316	0.2008	0.2045	0.2423	0.1480	0.1118	0.0904	0.2231	3.6817	0.1334	0.1114
${\mathrm{B}}_4$	0.1564	0.3186	0.2050	0.1246	0.1716	0.2426	0.1226	0.1628	0.1396	0.1966	3.5095	0.1715	0.1062
${\mathrm{B}}_5$	0.2122	0.2625	0.1883	0.1522	0.1076	0.1915	0.1577	0.0922	0.0713	0.1271	3.1668	−0.0416	0.0959
${\mathrm{B}}_6$	0.2167	0.3339	0.2409	0.2108	0.2094	0.1488	0.1550	0.1430	0.1441	0.2050	3.8555	0.1597	0.1167
${\mathrm{B}}_7$	0.1359	0.1533	0.1637	0.1052	0.1561	0.1412	0.0603	0.0564	0.0801	0.1549	2.4986	−0.0842	0.0756
${\mathrm{B}}_8$	0.2690	0.3112	0.2199	0.2138	0.1596	0.2011	0.1809	0.0970	0.2039	0.2061	3.2390	0.8862	0.0980
${\mathrm{B}}_9$	0.2041	0.2615	0.1786	0.2221	0.1420	0.2081	0.1437	0.1881	0.0873	0.1416	2.9692	0.5850	0.0899
${\mathrm{B}}_{10}$	0.0961	0.2406	0.1440	0.1477	0.0865	0.1565	0.0969	0.1433	0.1447	0.0862	2.9372	−0.2521	0.0889

reports the total relation matrix computed in Python, along with values of prominence and relation for each criterion. Values of prominence have been normalized to obtain criteria weights displayed in the last column of Table 3.

Results are graphically visualized in Figure 2, from which we can immediately notice as DEMATEL assigns higher prominence to criterion ${\mathrm{B}}_2$ (Satisfaction Level of Customer). From a practical standpoint, this underlines the fundamental importance perceived by the company in meeting customer expectations and fostering positive relationships. Attributing higher weight to this criterion ensures that logistical strategies are designed and executed with the primary goal of enhancing customer satisfaction. Following closely behind, the criterion of Quality (${\mathrm{B}}_6$) appears to be crucial for maintaining customer satisfaction, as neglecting quality considerations can lead to damaged or lost goods, resulting in potential business losses. Dependability (${\mathrm{B}}_3$) ranks next in importance due to its direct impact on customer trust and confidence. Consistently delivering on promises and meeting delivery deadlines strengthen reliability and credibility, promoting positive customer experiences and long-term relationships. Responsiveness (${\mathrm{B}}_4$) is also significant as it contemplates the ability to address and resolve customer inquiries or complaints promptly.

Figure 3 reports the histogram of prominence values showing that, according to the company management, two criteria are associated with approximately the same level of importance, i.e., ${\mathrm{B}}_1$ (Optimization of lead-time) and ${\mathrm{B}}_5$ (Affordability).

The weights reported in the last column of Table 3 will be used within the set of input data for the following ELECTRE TRI application.

4.3. ELECTRE TRI Prioritizing RO and WM Measures

In the present subsection, we are going to lead two separate iterations of the ELECTRE TRI procedure to assess various data-driven measures related to the two strategies to be implemented by the company according to [4], which we recall being route optimization and warehouse management. We have first led a literature review to formalize a set of relevant measures that can be implemented for each strategy. Ten measures related to route-optimization strategy (${\mathrm{RO}}_1$, ${\mathrm{RO}}_2$, …, ${\mathrm{RO}}_{10}$) are reported in

Table 4. Data-Driven Proactive Measures for Route-Optimization (RO) Strategy.

ID	Measure	Description
${\mathrm{RO}}_1$	Geographic Information System (GIS) and Weather-Adaptive Routing	The GIS web service facilitates collaborative workspace for transportation departments, streamlining coordination. It utilizes GIS to plan optimal routes for site assessments and installations, adapting to weather forecasts for efficient planning in weather-dependent industries like the Solar Industry [53].
${\mathrm{RO}}_2$	Dynamic Scheduling	Implementing systems that dynamically adjust schedules and routes based on real-time information, such as weather conditions, which are critical in solar installations [54].
${\mathrm{RO}}_3$	Advanced Mapping for Remote Locations	Utilizing high-resolution maps that provide detailed views of remote or difficult-to-access locations to plan the safest and most efficient access routes [55].
${\mathrm{RO}}_4$	Optimized Material Delivery	Scheduling deliveries of heavy equipment like solar panels and storage batteries in coordination with installation progress to avoid on-site congestion [56].
${\mathrm{RO}}_5$	Real-Time Communication Networks	Establishing robust communication links between the field teams and central control to facilitate immediate updates and rerouting based on site conditions [56].
${\mathrm{RO}}_6$	Modular Route Planning	Creating flexible, modular routing plans that can be easily adjusted as project scopes expand or contract, typical in large-scale solar projects. This measure is focused more on the planning stage [57].
${\mathrm{RO}}_7$	Inventory Management Integration	Linking route plans with inventory levels to ensure that all necessary materials are available on-site when needed, reducing return trips [58].
${\mathrm{RO}}_8$	Multi-Modal Transport Solutions	Utilizing a combination of transport methods (road, rail, air) for equipment delivery, especially to remote or hard-to-reach locations [59].
${\mathrm{RO}}_9$	Risk Assessment	Identifying disruptions like traffic congestion or adverse weather, assessing their likelihood and impact on delivery schedules, which allows for proactive strategies to mitigate risks [60].
${\mathrm{RO}}_{10}$	AI-Enhanced Optimization Tools	implementing AI tools and methods to continually refine and improve routing strategies based on outcomes, learning from each project to optimize future routing [61].

, while ten measures related to warehouse-management strategy (${\mathrm{WM}}_1$, ${\mathrm{WM}}_2$, …, ${\mathrm{WM}}_{10}$) are formalized in

Table 5. Data-Driven Proactive Measures for Warehouse-Management (WM) Strategy.

ID	Measure	Description
${\mathrm{WM}}_1$	Real-Time Inventory Tracking	These systems provide real-time data on inventory levels and locations within the warehouse. Help in locating items quickly with fewer errors and improve order fulfillment [62].
${\mathrm{WM}}_2$	Automated Replenishment Systems and Multi-Channel Fulfillment	Use of automated systems in monitoring inventory and automatically reordering products. It ensures a continuous supply when stock levels fall below a fixed threshold level and manages inventories efficiently across all sales platforms, reducing stockouts and overstock situations [63].
${\mathrm{WM}}_3$	Cross-Docking	Implementing cross-docking where received products at a warehouse can be processed immediately rather than storing them. Suitable for products with short shelf life or high demand [64].
${\mathrm{WM}}_4$	Integrated Warehouse-Management System (WMS)	Utilizing a WMS to integrate with other business systems like ERP and CRM to streamline warehouse operations. This will be helpful for decision-making and to enhance the accuracy of the data [65].
${\mathrm{WM}}_5$	Quality Control Procedures	Establishing robust quality control procedures throughout the warehouse process to ensure standards of particular products or items. This reduces returns, increases customer satisfaction, and maintains product integrity [66].
${\mathrm{WM}}_6$	Sustainable Warehousing Practices	Introducing green initiatives such as energy-efficient lighting, solar power installations, and waste reduction programs. It will reduce total environmental impact and lower operational costs [67].
${\mathrm{WM}}_7$	Safety and Security Enhancements	Enhancing warehouse safety and security with modern tools, surveillance systems, access controls, and cybersecurity measures for IT systems [68].
${\mathrm{WM}}_8$	Continuous Improvement and Workplace Management	Developing a culture of continuous improvement by reviewing warehouse operations and requesting feedback from employees [69].
${\mathrm{WM}}_9$	Collaborative Robotics (Cobots)	Introducing robots that work alongside humans to perform repetitive or physically demanding tasks, enhancing productivity and safety [70].
${\mathrm{WM}}_{10}$	Lean Inventory Techniques	Applying lean principles to minimize waste in inventory management, focusing on maintaining just enough inventory to meet demand. Using Lean, Six Sigma, and other methodologies to identify inefficiencies [71,72].

.

The literature review on implementation measures provides a strong foundation for this research. As previously done to identify criteria to be weighted with DEMATEL, this approach ensures that the measures assessed in the next two ELECTRE TRI iterations are comprehensive and informed by established practices and theories, providing the company with reliable recommendations.

We now proceed by formalizing the input data required for leading each of the two ELECTRE TRI iterations, which will be separately carried out for measures related to RO and WM strategies. Input data have been formalized according to the explanations provided in [31] and collected in

Table 6. ELECTRE TRI input data for measures related to RO strategy.

Input	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$
$P_2$	7.67	6.67	5.00	4.00	3.33	5.33	5.00	4.67	3.33	5.67
$P_1$	5.33	4.33	3.00	2.00	1.67	3.67	3.00	2.33	1.67	4.33
$I_k$	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
$S_k$	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00
$V_k$	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00
$w_k$	0.0952	0.1222	0.1114	0.1062	0.0959	0.1167	0.0756	0.0980	0.0899	0.0889

and

Table 7. ELECTRE TRI input data for measures related to WM strategy.

Input	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$
$P_2$	7.00	6.00	6.00	6.33	5.00	6.67	6.33	7.67	5.33	3.33
$P_1$	5.00	4.00	4.00	4.67	3.00	4.33	3.67	5.33	2.67	1.67
$I_k$	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
$S_k$	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00	2.00
$V_k$	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00	4.00
$w_k$	0.0952	0.1222	0.1114	0.1062	0.0959	0.1167	0.0756	0.0980	0.0899	0.0889

. Specifically, measures reported in Table 4 and Table 5 were numerically assessed with the support of an external technical consultant with proven experience in the logistics field. We asked the decision-maker to provide a numerical evaluation on a 10-point scale by assessing measures under the criteria (Table 1) previously weighted through the DEMATEL application.

Quantitative evaluations of measures under criteria are reported in

Table 8. Assignment of RO measures to classes via pessimistic (PP) and optimistic (OP) procedures.

ID	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$	PP	OP
${\mathrm{RO}}_1$	8.00	9.00	7.00	2.00	3.00	5.00	4.00	5.00	2.00	7.00	A	A
${\mathrm{RO}}_2$	10.00	6.00	7.00	3.00	5.00	6.00	5.00	3.00	4.00	6.00	A	A
${\mathrm{RO}}_3$	5.00	4.00	3.00	2.00	3.00	4.00	2.00	4.00	0.00	4.00	C	B
${\mathrm{RO}}_4$	8.00	8.00	4.00	4.00	5.00	6.00	3.00	4.00	3.00	7.00	B	A
${\mathrm{RO}}_5$	7.00	6.00	3.00	6.00	4.00	5.00	2.00	4.00	2.00	6.00	B	B
${\mathrm{RO}}_6$	5.00	4.00	2.00	3.00	5.00	6.00	1.00	0.00	1.00	4.00	B	B
${\mathrm{RO}}_7$	3.00	2.00	1.00	2.00	0.00	3.00	3.00	4.00	1.00	3.00	C	C
${\mathrm{RO}}_8$	8.00	7.00	5.00	3.00	5.00	7.00	7.00	4.00	0.00	3.00	A	A
${\mathrm{RO}}_9$	7.00	4.00	2.00	1.00	0.00	5.00	3.00	5.00	4.00	5.00	B	B
${\mathrm{RO}}_{10}$	10.00	8.00	5.00	0.00	3.00	2.00	6.00	7.00	5.00	6.00	B	A

and

Table 9. Assignment of WM measures to classes via pessimistic (PP) and optimistic (OP) procedures.

ID	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$	PP	OP
${\mathrm{WM}}_1$	6.00	5.00	6.00	8.00	6.00	6.00	2.00	9.00	7.00	3.00	B	B
${\mathrm{WM}}_2$	8.00	6.00	7.00	7.00	5.00	5.00	4.00	10.00	8.00	3.00	B	A
${\mathrm{WM}}_3$	7.00	7.00	6.00	6.00	7.00	5.00	5.00	8.00	6.00	1.00	B	A
${\mathrm{WM}}_4$	8.00	4.00	2.00	3.00	1.00	4.00	3.00	8.00	7.00	5.00	C	B
${\mathrm{WM}}_5$	6.00	8.00	8.00	7.00	3.00	8.00	2.00	9.00	3.00	4.00	B	A
${\mathrm{WM}}_6$	4.00	6.00	4.00	8.00	2.00	2.00	9.00	3.00	2.00	0.00	C	B
${\mathrm{WM}}_7$	3.00	2.00	4.00	7.00	5.00	5.00	1.00	5.00	1.00	1.00	C	B
${\mathrm{WM}}_8$	6.00	5.00	5.00	6.00	4.00	9.00	4.00	7.00	1.00	3.00	B	B
${\mathrm{WM}}_9$	9.00	6.00	6.00	5.00	3.00	8.00	5.00	6.00	0.00	0.00	B	B
${\mathrm{WM}}_{10}$	7.00	5.00	6.00	6.00	4.00	8.00	4.00	6.00	2.00	2.00	B	B

, along with ELECTRE TRI results obtained in each iteration in terms of assignment to three different priority classes (A, B, C) according to the pessimistic (PP) and optimistic (OP) procedures. In particular, class A suggests high priority in implementing measures, class B includes measures associated with medium priority recommended for implementation, while class C groups measures that are revealed to not be particularly urgent and whose implementation can be postponed according to budget availability. Results reported in Table 8 and Table 9 have been obtained by fixing the cutting value to $\lambda =0.8$.

As explained before in the methodological section, results derived from the ELECTRE TRI pessimistic procedure should be preferred over the optimistic assignment in order to guarantee robustness and pursue risk mitigation. The company should take this aspect into account when formalizing the final decision on specific data-driven measures to be implemented with priority for adopting route optimization and warehouse-management strategies. The pessimistic procedure ensures indeed that decisions are made considering the worst-case scenarios, leading to more conservative and reliable outcomes. On the contrary, the optimistic procedure may overlook potential risks or uncertainties, leading to less robust decisions that could be more vulnerable to unexpected events.

5. Discussion of Findings

Significant insights emerge from our real-world application, offering crucial considerations for solar energy management, and the integration of data-driven measures plays a key role in enhancing supply chain management processes within this sector. As the company progressively embraces proactive measures, the systematic approach herein proposed supports the integration of the desired strategies into solar energy management, ensuring optimized decision-making and resource utilization.

Initially, on the basis of the recommendations provided in [4], the company decided to plan the implementation of a route-optimization strategy, followed by a warehouse-management strategy. This decision required further support in defining a method for prioritizing the implementation of measures that align with the specific needs and capabilities of the mentioned strategies. For this reason, the DEMATEL-ELECTRE TRI integration herein proposed aims to enable a progressive and incremental approach for gradual adjustments and improvements, minimizing disruptions and maximizing effectiveness while taking into account budget constraints.

The initial DEMATEL application was based on the preliminary identification of a set of suitable criteria considered to be relevant to the problem of interest in the existing literature (Table 1). We were able to attribute different degrees of importance to criteria, and results led to higher weights attributed to the aspects of Satisfaction Level of Customer (${\mathrm{B}}_2$), Quality (${\mathrm{B}}_6$), Dependability (${\mathrm{B}}_3$), and Responsiveness (${\mathrm{B}}_4$). These results were obtained by mathematically treating input evaluations of mutual influence between pairs of criteria that were provided by the company management. On the basis of the composition of the resulting weights, we carried out two separate iterations of ELECTRE TRI aimed at assigning measures to three different classes (A, B, C) and expressing the priority of implementation.

The first iteration was performed on ten data-driven measures specifically aimed at implementing the road optimization strategy. Labeled as ${\mathrm{RO}}_1$, ${\mathrm{RO}}_2$, …, ${\mathrm{RO}}_{10}$, these measures were once again formalized by means of a thorough literature review (Table 4). Measures whose implementation is suggested with high-priority fall in class A (PP), and they are GIS and Weather-Adaptive Routing (${\mathrm{RO}}_1$), which utilizes geographic information systems and real-time weather data to optimize route planning based on weather conditions; Dynamic Scheduling (${\mathrm{RO}}_2$), enabling real-time adjustments to schedules and routes, enhancing responsiveness to changing circumstances; and Multi-Modal Transport Solutions (${\mathrm{RO}}_8$), integrating multiple modes of transportation to optimize logistics and reduce costs. Implementing these measures would drive the company towards efficient allocation of resources to initiatives that offer the most immediate and impactful improvements in road optimization, laying a solid foundation for further data-driven enhancements in solar energy management.

The second ELECTRE TRI iteration was performed on ten data-driven measures specifically aimed at implementing the warehouse-management strategy. Labeled as ${\mathrm{WM}}_1$, ${\mathrm{WM}}_2$, …, ${\mathrm{WM}}_{10}$, these measures were once again formalized by means of a thorough literature review (Table 5). Measures whose implementation is suggested with high-priority fall in class A (OP), and they are Automated Replenishment Systems and Multi-Channel Fulfillment (${\mathrm{WM}}_2$), which streamline inventory replenishment processes and enhance order fulfillment across multiple channels; Cross-Docking (${\mathrm{WM}}_3$), facilitating efficient transfer of goods from inbound to outbound shipments, minimizing storage time and costs; and Quality Control Procedures (${\mathrm{WM}}_5$), ensuring product quality standards are met before distribution. These measures would support the company in enhancing warehouse management, reducing costs, and further improving customer satisfaction.

From a methodological perspective, it can be observed that, for the second group of measures (WM measures), ${\mathrm{WM}}_2$, ${\mathrm{WM}}_3$ and ${\mathrm{WM}}_5$ have been sorted to class A through the optimistic procedure (OP), while the pessimistic procedure (PP) suggests a medium priority for implementation. This discrepancy suggests the usefulness of prioritizing measures aimed at optimizing the route-optimization strategy. As such, we strongly recommend the company to invest in implementing ${\mathrm{RO}}_1$, ${\mathrm{RO}}_2$ and/or ${\mathrm{RO}}_8$ to maximize effectiveness in logistic operations. By restricting our recommendation to these specific measures, the company can make more informed decisions on how to allocate resources efficiently while achieving tangible benefits, building momentum for future adoption of WM measures and other measures assigned to the B class. This step-by-step implementation process aims to ensure a strategic and cost-effective transition to data-driven practices, enabling the company to derive maximum value from its solar energy assets while effectively managing budget constraints. The results of the proposed decision-making approach prioritize measures, leaving the final implementation to be tailored by the company based on its specific budgetary and operational considerations. We further specify that the focus of this research lies in providing recommendations for the implementation of strategic measures at a managerial level rather than quantifying their outcomes upon their practical implementation.

To narrow down our analysis even more, we now proceed by evaluating the robustness of results related to the route-optimization strategy by implementing a sensitivity analysis on criteria weights. The objective is two-fold: first, to identify if certain measures consistently dominate across different scenarios, indicating their significance relative to others. Second, we aim to discern if adjustments in the weight of a particular criterion prompt recommendations for prioritizing different measures from the ones previously discussed and recommended. By comprehensively analyzing these factors, we aim to gain a deeper understanding of the stability and adaptability of measures for the route-optimization strategy in various contexts.

In addition to our baseline scenario (BS) considering the weights originally calculated via DEMATEL, we have developed ten additional scenarios, labeled as ${\mathrm{S}}_1$, ${\mathrm{S}}_2$, …, ${\mathrm{S}}_{10}$ (see

Table 10. Weight scenarios elaborated for sensitivity analysis.

ID	${\mathbf{B}}_1$	${\mathbf{B}}_2$	${\mathbf{B}}_3$	${\mathbf{B}}_4$	${\mathbf{B}}_5$	${\mathbf{B}}_6$	${\mathbf{B}}_7$	${\mathbf{B}}_8$	${\mathbf{B}}_9$	${\mathbf{B}}_{10}$
BS	0.0952	0.1222	0.1114	0.1062	0.0959	0.1167	0.0756	0.0980	0.0899	0.0889
${\mathrm{S}}_1$	0.3952	0.0889	0.0781	0.0729	0.0626	0.0834	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_2$	0.0619	0.4222	0.0781	0.0729	0.0626	0.0834	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_3$	0.0619	0.0889	0.4114	0.0729	0.0626	0.0834	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_4$	0.0619	0.0889	0.0781	0.4062	0.0626	0.0834	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_5$	0.0619	0.0889	0.0781	0.0729	0.3959	0.0834	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_6$	0.0619	0.0889	0.0781	0.0729	0.0626	0.4167	0.0423	0.0647	0.0566	0.0556
${\mathrm{S}}_7$	0.0619	0.0889	0.0781	0.0729	0.0626	0.0834	0.3756	0.0647	0.0566	0.0556
${\mathrm{S}}_8$	0.0619	0.0889	0.0781	0.0729	0.0626	0.0834	0.0423	0.3980	0.0566	0.0556
${\mathrm{S}}_9$	0.0619	0.0889	0.0781	0.0729	0.0626	0.0834	0.0423	0.0647	0.3899	0.0556
${\mathrm{S}}_{10}$	0.0619	0.0889	0.0781	0.0729	0.0626	0.0834	0.0423	0.0647	0.0566	0.3889

). Each scenario was developed incrementally, increasing the weight of a specific criterion, and numbered in correspondence with that criterion (${\mathrm{B}}_1$, ${\mathrm{B}}_2$, …, ${\mathrm{B}}_{10}$; see Table 1). For example, scenario ${\mathrm{S}}_1$ emphasizes the weight originally attributed to criterion ${\mathrm{B}}_1$, scenario ${\mathrm{S}}_2$ emphasizes the weight originally attributed to criterion ${\mathrm{B}}_2$, and so on, until scenario ${\mathrm{S}}_{10}$, which emphasizes the weight originally attributed to criterion ${\mathrm{B}}_{10}$. This means that as we move from ${\mathrm{S}}_1$ to ${\mathrm{S}}_{10}$, each successive scenario incrementally focuses on attributing more importance to the corresponding criterion ${\mathrm{B}}_1$ to ${\mathrm{B}}_{10}$. We specify that the weights of the remaining nine criteria are proportionally decreased to maintain a total weight of 1 in each scenario. Specifically, we have introduced an incremental increase of 30% to the weight of the corresponding criterion from its original value in each scenario. It is important to note that this is an additional increase in the criterion’s original weight, effectively augmenting its importance in the decision-making process. The elaborated scenarios provide a comprehensive understanding of the interplay between different criteria and their influence on the robustness of our findings. By systematically adjusting the weighting of each criterion, we can clearly discern how changes in priority affect the prioritization of measures to be recommended to the company, providing a final overview in

Table 11. Assignments to classes obtained in diverse weight scenarios via pessimistic procedure.

ID	BS	${\mathbf{S}}_1$	${\mathbf{S}}_2$	${\mathbf{S}}_3$	${\mathbf{S}}_4$	${\mathbf{S}}_5$	${\mathbf{S}}_6$	${\mathbf{S}}_7$	${\mathbf{S}}_8$	${\mathbf{S}}_9$	${\mathbf{S}}_{10}$
${\mathrm{RO}}_1$	A	B	A	A	B	B	B	B	B	B	A
${\mathrm{RO}}_2$	A	A	B	A	B	A	B	B	B	B	B
${\mathrm{RO}}_3$	C	C	C	C	C	C	C	C	B	C	C
${\mathrm{RO}}_4$	B	B	B	B	B	A	B	B	B	B	B
${\mathrm{RO}}_5$	B	B	B	B	A	B	B	B	B	B	B
${\mathrm{RO}}_6$	B	C	C	C	C	B	B	C	C	C	C
${\mathrm{RO}}_7$	C	C	C	C	C	C	C	C	C	C	C
${\mathrm{RO}}_8$	A	B	B	B	B	A	A	A	B	C	B
${\mathrm{RO}}_9$	B	B	C	C	C	C	B	C	B	B	C
${\mathrm{RO}}_{10}$	B	B	B	B	C	B	C	B	B	B	B

.

The following practical considerations can be formulated.

• ${\mathrm{RO}}_1$ (GIS and Weather-Adaptive Routing) is assigned to high-priority class A under scenarios ${\mathrm{S}}_2$, ${\mathrm{S}}_3$, and ${\mathrm{S}}_{10}$, respectively, associated with higher weights attributed to criteria ${\mathrm{B}}_2$ (Satisfaction level of Customer), ${\mathrm{B}}_3$ (Dependability), and ${\mathrm{B}}_{10}$ (Reverse Logistics Optimization). At a practical level, this indicates that the company should prioritize the implementation of measure ${\mathrm{RO}}_1$ if enhancing customer satisfaction, ensuring reliability, or optimizing reverse logistics processes are paramount objectives. By incorporating GIS and Weather-Adaptive Routing, the company can indeed retrieve real-time data and use predictive analytics to optimize routing decisions, leading to improved on-time deliveries, reduced transportation costs, and enhanced customer experiences. Additionally, by proactively adapting to weather conditions, the company can mitigate risks associated with delays and disruptions, enhancing overall operational efficiency and competitiveness.
• ${\mathrm{RO}}_2$ (Dynamic Scheduling) is assigned to high-priority class A under scenarios ${\mathrm{S}}_1$, ${\mathrm{S}}_3$, ${\mathrm{S}}_5$, and ${\mathrm{S}}_{10}$, respectively, associated to higher weights attributed to criteria ${\mathrm{B}}_1$ (Optimization of lead-time), ${\mathrm{B}}_3$ (Dependability), ${\mathrm{B}}_5$ (Affordability), and ${\mathrm{B}}_{10}$ (Reverse Logistics Optimization). At a practical level, this suggests that the company should prioritize the implementation of measure ${\mathrm{RO}}_2$ if optimizing lead-time, ensuring reliability, achieving cost-effectiveness, or optimizing reverse logistics processes are primary objectives. By implementing dynamic scheduling algorithms, the company can indeed optimize production schedules in real time, ensuring efficient resource allocation, minimizing idle time, and reducing lead times. This enables the company to promptly respond to changing demand patterns, enhance production, and meet customer expectations with timely deliveries. Additionally, dynamic scheduling contributes to cost-effectiveness by minimizing overtime costs, reducing inventory holding costs, and optimizing resource utilization.
• ${\mathrm{RO}}_8$ (Multi-Modal Transport Solutions), is assigned to high-priority class A under scenarios ${\mathrm{S}}_5$, ${\mathrm{S}}_6$, and ${\mathrm{S}}_7$, respectively, associated to higher weights attributed to criteria ${\mathrm{B}}_5$ (Affordability), ${\mathrm{B}}_6$ (Quality), and ${\mathrm{B}}_7$ (Sustainability). At a practical level, this suggests that the company should prioritize the implementation of measure ${\mathrm{RO}}_8$ if achieving affordability, ensuring quality, or promoting sustainability are key objectives. By considering multi-modal transport solutions, the company can diversify its transportation options, optimize routes, and minimize transportation costs. In such a way, competitive pricing can be offered to customers while maintaining profitability. Additionally, multi-modal transport solutions enhance the quality of transportation services by providing greater responsiveness to customer demands. By reducing reliance on single-mode transportation and incorporating environmentally friendly modes such as rail or sea freight, the company can significantly minimize its carbon footprint and contribute to sustainability efforts.

Further considerations can be elaborated on the possibility of implementing measure ${\mathrm{RO}}_4$ (Optimized Material Delivery), assigned to high-priority class A under scenario ${\mathrm{S}}_5$, either measure ${\mathrm{RO}}_5$ (Real-Time Communication Network), assigned to high-priority class A under scenario ${\mathrm{S}}_4$. At a practical level, this suggests that the company may evaluate the opportunity of implementing measure ${\mathrm{RO}}_4$ for affordability purposes (${\mathrm{B}}_5$) or measure ${\mathrm{RO}}_5$ for increased responsiveness (${\mathrm{B}}_4$). We lastly specify that, under scenarios ${\mathrm{S}}_8$ and ${\mathrm{S}}_9$, respectively, corresponding to the maximization of criteria related to Supply-Chain Visibility (${\mathrm{B}}_8$) and Supplier Performance Monitoring (${\mathrm{B}}_9$), no measure is assigned to high-priority class for implementation. In such a case, we would recommend making a final decision based on the results obtained and discussed for the remaining scenarios.

Results derived from the sensitivity analysis are graphically displayed in Figure 4.

6. Conclusions

This paper offers a comprehensive examination of the solar industry’s progression towards resilient energy solutions, stressing the importance of optimizing supply chain processes and resource management. Building on previous research focused on the impact of Big Data on solar logistics, we now propose a hybrid MCDM method rooted in expert opinions to support proactive, data-driven measures within the solar sector. Specifically, we focus on the progressive implementation of two key strategies, i.e., route optimization and warehouse management, conducting a detailed literature analysis of significant criteria along with measures suitable to each strategy to be prioritized.

The company analyzed in our case study needed a reliable methodological approach to prioritize measure implementation in alignment with the two mentioned strategies. The proposed MCDM integration aims to develop a gradual and budget-conscious approach for implementing proactive, data-driven measures formalized from the existing literature. We identified critical criteria, with an emphasis on such aspects as customer satisfaction, quality, dependability, and responsiveness, as primary considerations. Subsequently, measures were sorted into ordered classes, suggesting the priority of implementation. Initially, our focus was on measures related to road optimization strategy, revealing that GIS and weather-adaptive routing, dynamic scheduling, and multi-modal transport should be implemented with priority. These measures would offer immediate enhancements, laying a solid foundation for broader improvements in solar energy management. Subsequently, our attention turned to measures related to warehouse-management strategy, where automated replenishment systems, cross-docking, and quality control emerged as priorities, improving customer satisfaction. By concentrating on measures with substantial impact, the company can allocate resources efficiently, facilitating the future adoption of measures classified as medium-priority class. From a methodological perspective, we observed that results related to the route-optimization strategy exhibit higher consistency. Building upon this, we focused on this specific strategy and elaborated a final sensitivity analysis on criteria weights in order to discern which measures among GIS and weather-adaptive routing, dynamic scheduling, and multi-modal transport better match specific criteria. By focusing on specific criteria, this flexible approach enables the company to strategically align implementation efforts with its overarching objectives.

The proposed approach emphasizes prioritizing measures, allowing the company to customize the final implementation based on individual budgetary and operational factors while easing the transition to data-driven practices. It is important to note that this research primarily focuses on offering managerial-level recommendations for strategic measure implementation rather than quantifying their outcomes post-implementation. Future developments of this research could focus on working closely with the company to quantify the outcomes achieved through the application of proposed measures. Additionally, there is potential to expand the scope to include optimization of panel manufacturing technology, extending the current focus on supply chain process optimization for other companies involved in solar panel production and marketing. One limitation of this approach, centered on subjective input evaluations, is the inherent potential for bias in the assessment process. To address this, future research could incorporate multiple perspectives for an even more comprehensive analysis. However, it is also crucial to acknowledge the importance of expert opinions provided by the company management and the technical consultant in aligning with the practical context of solar energy management, as they provided valuable insights that may not be captured through purely quantitative methods. Moreover, the proposed method serves as a versatile framework that can be adopted by companies across various sectors. By integrating their own input evaluations, organizations can tailor the approach to suit their specific needs and industry requirements. This flexibility allows for broader applicability, empowering businesses beyond the solar energy sector to leverage the benefits of optimized route planning and data-driven resource allocation.