A Conceptual Framework for Sustainable AI-ERP Integration in Dark Factories: Synthesising TOE, TAM, and IS Success Models for Autonomous Industrial Environments

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I recommend the publication of the paper, but before its publication, some issues should be completed in order to improve the quality of the manuscript:

1. "The absence of firm-level microdata on AI-ERP adoption led to the creation of a dataset by data mining on Python, available in public reports and scholarly indicators..." - Sources should be cited by citing public reports and scholarly Indicators so that the reader can have a better idea.

2. Referring to Figure 4 - please add the source on which the chart was based. The readability of the chart would also be improved by indicating whether the data concern only the adaptation of AI in enterprises or also other institutions?

3. Does IS Success also take into account the ethical aspect related to the implementation of AI in the organization (responsibility, employee concerns...)? The ethical aspect is a significant limitation when it comes to implementing AI in enterprises.

4. Referring to Figure 5- The source should be supplemented. Does the data refer to a global/country scale?

5. "The hospital industry may choose to implement AI-ERP systems after analyzing regulatory policies, infrastructure capabilities, and financial constraints using a TOE analysis" – It is worth noting the limitations in the case of healthcare related to the handling of medical data by AI.

6. "Following the adoption, the indicators of post-adoption success are measured using the IS Success Model, which reflects the quality of systems, quality of information, quality of services, and satisfaction of the users" - 

   How long after model deployment should success rates be measured? At what intervals are they to be monitored? What are the targets? Shouldn't baseline values be specified here? Are there any recommendations in case of failure to achieve the expected results?

7. After Table 1, you must reformat the text as required.

Author Response

Comments and Suggestions for Authors

I recommend the publication of the paper, but before its publication, some issues should be addressed in order to improve the quality of the manuscript:

1. "The absence of firm-level microdata on AI-ERP adoption led to the creation of a dataset by data mining on Python, available in public reports and scholarly indicators..." - Sources shouldbe cited by citing public reports and scholarly Indicators so that the reader can have a better idea.

Author’s comment- Added on page 9 - 1) Public report cited with source, added on page 10 - 2) A clarify of datasets was built in python.

2. Referring to Figure 4 - please add the source on which the chart was based. The readability of the chart would also be improved by indicating whether the data concern only the adaptation of AI in enterprises or also other institutions?

Author’s comment- Source added on page 11 in below the caption with figure 4. And ddata present AI as enterprise and not governmental adoption.

3. Does IS Success also take into account the ethical aspect related to the implementation of AIin the organization (responsibility, employee concerns...)? The ethical aspect is a significantlimitation when it comes to implementing AI in enterprises.

Author’s comment-The ethical aspect related to the implementation of AI in the organization is added on page 13 first paragraph last line.

4. Referring to Figure 5- The source should be supplemented. Does the data refer to a global/country scale?

Author’s comment- Source added on page 13 for global data below the figure 5.

5. "The hospital industry may choose to implement AI-ERP systems after analyzing regulatory policies, infrastructure capabilities, and financial constraints using a TOE analysis" – It is worth noting the limitations in the case of healthcare related to the handling of medical data by AI.

Author’s comment- After the implementation of AI-ERP systems, such success rates are to be measured on page 14.

6. "Following the adoption, the indicators of post-adoption success are measured using the IS Success Model, which reflects the quality of systems, quality of information, quality of services, and satisfaction of the users"

Author’s comment- Post-adoption success are measured using the IS Success Model added on page 19.

7. How long after model deployment should success rates be measured? At what intervals are they to be monitored? What are the targets? Shouldn't baseline values be specified here? Are there any recommendations in case of failure to achieve the expected results?

Author’s comment- Added on page 14, first after 3 months, 6 months, and 1 years of implementation, and then on a quarterly or biannual basis. These periods would make sure that AI-ERP systems at all times are engaged with business goals and provide quantifiable results. The baselines will provide points of reference to evaluate the gains in respect to set targets. Organisations adopt contingency plans (where root cause analysis would justify the underlying causes of failure, based on questionable results) in the case of failure to reach the expected outcomes, including developing strong feedback, ongoing data collection feedback, improving systems on a systematic basis depending on the feedback and improving training efforts to reduce resistance.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This article addresses the current topic of integrating artificial intelligence with ERP systems for resource planning in highly automated industrial plants, so-called "dark factories." Furthermore, it proposes an integrated implementation model combining three classic theoretical approaches: TOE, TAM, and the IS Success Model. The paper has a satisfactory theoretical foundation and references to research and practical examples. However, after carefully reviewing the article, I believe that corrections and clarifications are required to ensure it meets publication standards. Therefore, I propose the following changes:

1. The authors did not present specific, practical research based on actual AI-ERP implementations in "dark factories" (real-world use cases, case studies, etc.). This should be added to the theoretical section of the article.
2. The model proposed by the authors is based primarily on theoretical analyses, simulations, and secondary data. Results from practical tests demonstrating how the model performs in real industrial settings are lacking. Therefore, I suggest changing the article's title to clearly emphasize the theoretical nature of the model. Furthermore, the abstract and summary should clearly indicate that the model is conceptual in nature and that it serves as a basis for further empirical research. Furthermore, the abstract and introduction should include a single, specific research objective.
3. The introduction should identify the research area and research questions.
4. Research hypotheses should be formulated based on the literature review. Section 2.2: Lack of detailed analysis of potential algorithmic errors that could lead to incorrect decisions in the AI-ERP system. Insufficient discussion of data security threats, including the risk of leakage, unauthorized access, and cyberattacks. Superficial approach to risks related to employee adaptation, such as resistance to change, lack of appropriate competencies, and the need for training. Lack of assessment of the impact of these risks on the stability and effectiveness of the system after implementation. Insufficient discussion of risk management mechanisms and strategies for minimizing the negative effects of AI integration in ERP. Lack of analysis of the impact of legal regulations and ethical issues on the security and acceptance of AI-ERP.
5. Section 2.2. Challenges in AI-ERP Implementations in Industrial Sectors. The authors address the challenges associated with the AI ​​black-box model and ethical issues, highlighting the potential for bias and model accuracy issues, especially in areas with stringent regulations (e.g., China). However, an in-depth discussion of these ethical and social barriers and practical solutions for overcoming them is lacking.
6. The article lacks a more detailed and practically useful elaboration on the areas crucial for the effective use and evaluation of an integrated AI-ERP model in the industrial sector. Sections 2.4.4 and 4.4 should detail key success indicators (e.g., system quality, user satisfaction, operational efficiency). In sections 2.4.4 and 4.1, the authors should thoroughly characterize the implementation and describe feedback mechanisms for continuous system improvement. In the following sections, 4.1 and 4.4, the tools and methods for measuring aspects such as "AI autonomy" and "human-machine collaboration" (e.g., metrics, evaluation methods) should be specified. Sections 4.4 and 4.5 (in the context of regression studies and KPIs) should refine the description of methods for implementing and monitoring key components of the framework at the operational level. Sections 4.1 and 2.5 (Gaps in Existing Research) should include a specification of tools for measuring the analyzed mechanisms in terms of the specifics of the "dark industry."
7. In section 2.5. Gaps in Existing Research, the authors mention the need to examine regional and sectoral differences, but this topic remains superficial. The impact of specific cultural, regulatory, and infrastructural conditions on the AI-ERP adaptation and implementation process would require further discussion. Furthermore, this section highlights ethical challenges and the need to ensure algorithm transparency, but there is no detailed description of how these issues can be incorporated into the proposed implementation model or what mechanisms might address them.
8. Chapter 3, Methodology, lacks primary data from empirical studies, which may limit the reliability and generalizability of the results. A justification for variable selection and measurement is needed. It is worthwhile to more precisely describe how specific variables, such as "system autonomy" or "behavioral intention," are measured or modeled in simulated data to increase the transparency and reproducibility of the study. Advanced data analysis methods are lacking. OLS regression is the primary method. Moderating variables are not included. Elements such as environmental pressures, employee skills, and ethical aspects can influence AI-ERP implementation outcomes and are worth considering or proposing in future research.The analysis is limited to 25% of the variance in user satisfaction.The model explains approximately 25% of the variance in user satisfaction, which clearly indicates that important factors may not be included.The authors should analyze potential additional variables or influencing factors.
9. Section 4.4.Integrating TOE, TAM, and IS Success Model in Business Settings discusses model integration, but it lacks a comprehensive description of how to adapt the model to diverse industries and different regional contexts.Adding such information would increase the universality and effectiveness of the proposed framework.Furthermore, there is no elaboration on how an integrated AI-ERP model can systematically identify, monitor, and address ethical and social issues during the implementation process, and how to promote transparency and user trust.
10. In Chapter 5. Discussion: The formulations regarding regional and sectoral differences are rather general.The model could be enriched with specific guidelines or adaptation mechanisms that would allow organizations to adapt AI-ERP implementations to different regulatory, cultural, and market environments.
11. In Chapter 6, Conclusions, despite highlighting the need to study local conditions, the authors do not present detailed recommendations or implementation steps that take these differences into account.The model would be more practical if it included clear adaptation guidelines and considered the dynamics of diverse markets.Although the authors acknowledge the importance of AI ethics, there is no comprehensive discussion of possible social barriers (such as employee resistance or impact on workplaces) or a framework for addressing them in implementation practice.The limitations of the research are also not provided.
12. There are a few minor errors or ambiguities in the text that could be corrected to improve the flow and clarity of the message. Language improvements may include: improving the consistency and precision of technical terms and phrases, avoiding repetition and simplifying certain sentences for readability, clarifying and removing unusual grammatical or punctuation structures that may hinder comprehension, and using more natural idiomatic expressions where the style might seem too stiff or artificial. Overall, the language used in the article is appropriate and professional, but before final publication, it's recommended to have a native speaker or professional language editor review the text to ensure the highest linguistic and stylistic quality.

Comments on the Quality of English Language

12. There are a few minor errors or ambiguities in the text that could be corrected to improve the flow and clarity of the message. Language improvements may include: improving the consistency and precision of technical terms and phrases, avoiding repetition and simplifying certain sentences for readability, clarifying and removing unusual grammatical or punctuation structures that may hinder comprehension, and using more natural idiomatic expressions where the style might seem too stiff or artificial. Overall, the language used in the article is appropriate and professional, but before final publication, it's recommended to have a native speaker or professional language editor review the text to ensure the highest linguistic and stylistic quality.

Author Response

Please  see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper comprehensively reviews the integration of artificial intelligence (AI) and enterprise resource planning (ERP) systems, focusing specifically on "dark factories"—highly automated, unmanned manufacturing environments that minimize human intervention through robotics, AI, and the Internet of Things. Drawing on secondary data and an extensive literature review, the authors identify key gaps in existing AI-ERP research, including the lack of a unified adoption framework, insufficient attention to AI-specific challenges, and limited post-adoption evaluation metrics. To address these issues, they propose a novel integrated conceptual framework that synthesizes the technology-organization-environment (TOE) framework, the technology acceptance model (TAM), and the information systems (IS) success model. This model incorporates factors specific to dark factories, such as AI autonomy, human-machine collaboration, operational agility, and sustainability elements.

Key contributions include:

By integrating theoretical models with real-world industrial challenges, this paper advances the discussion on sustainable AI-ERP adoption, providing value to researchers, technology leaders, and policymakers in the manufacturing industry and related sectors.

1) This paper provides a comprehensive theoretical foundation for the entire AI-ERP lifecycle, from pre-adoption readiness (TOE) to user acceptance (TAM) to post-adoption success (the IS Success Model), tailored for sustainable operations in autonomous industrial environments.

2) It provides practical insights for organizations, identifying actionable strategies across the pre-, mid-, and post-adoption phases, and provides a table outlining the synergies between the model and industry applications.

3) It emphasizes sustainability, linking AI-ERP to environmental, economic, and social benefits, aligned with broader goals such as the circular economy and resilient supply chains.

4) Through simulated regression analysis of enterprise-level data, empirical results demonstrate a positive correlation between system usage and user satisfaction, thus partially validating the framework.

 

Detailed Evaluation of Methodology, Analysis, and Conclusions

This research methodology primarily utilizes qualitative and review methods, supplemented by quantitative analysis using simulated data. The authors construct a case study based on secondary sources, conduct a systematic review to identify gaps, and integrate the TOE, TAM, and IS Success Models. In terms of quantitative analysis, they simulated a dataset based on real-world indicators from public reports and academic data, focusing on observational data from 200 companies in technology-intensive industries. They used ordinary least squares (OLS) regression in Python to analyze the relationships between AI-ERP system usage, cost reduction, productivity improvement, efficiency gains, and user satisfaction. This aligns with the study's goal of constructing a model by linking three models.

The analysis is clearly structured and logically rigorous. The literature review effectively highlights gaps, such as insufficient attention to the dark industry context and the fragmented use of adoption models. The proposed framework is presented visually and in detail in a table format, clearly demonstrating the mapping of variables and applications across industries. Regression analysis revealed that system usage had a significant positive impact on user satisfaction (β = 0.679, p < 0.001, adjusted R² = 0.248), and correlation diagnostics supported the model's validity. However, other variables were not significant, which could be attributed to collinearity or the dominance of system usage—a plausible explanation that would benefit from further research. Data trends reinforce the narrative of increasing automation but fluctuating due to challenges.

The analysis strongly supports these conclusions, highlighting the framework's role in guiding AI-ERP integration to achieve sustainability in "dark factories." They reiterate the lifecycle approach and call for future empirical validation, while acknowledging limitations such as its reliance on secondary/simulated data. However, these conclusions could be more explicitly linked to sustainability metrics. Trends in figures reinforce the narrative of growing automation but fluctuating adoption due to challenges. The conclusions are well-supported by the analyses, emphasizing the framework's role in guiding AI-ERP integration for sustainability in dark factories. They reiterate the lifecycle approach and call for future empirical validation, acknowledging limitations like reliance on secondary/simulated data. However, the conclusions could more explicitly tie back to sustainability metrics and address potential biases in the simulated data. Overall, the methodology is appropriate for a conceptual review paper, though the simulated quantitative element adds novelty but lacks the rigor of primary empirical data, potentially limiting generalizability.

This is a timely and well-organized paper that effectively integrates existing models into a practical framework for AI-ERP in a sustainable automation environment. The focus on "dark factories" and sustainability is particularly innovative and relevant to the Industry 4.0/5.0 transition. However, to enhance the paper's publication potential, please consider the following suggestions:

1. The simulated regression is a good start, but the reliance on secondary/synthetic data may raise validity concerns. Consider incorporating original or real datasets in future revisions. If simulated data are retained, please provide more details on data generation and robustness checks.
2. While sustainability is a core theme, it could be more quantitatively linked to the framework. For example, the model could be expanded to include specific outcomes such as carbon reduction or circular economy processes, perhaps incorporating hypothetical KPIs or case studies. This would better align with the journal's focus and distinguish it from general AI-ERP reviews.
3. The paper identifies gaps in the literature but could explicitly discuss its own limitations, such as its Western-centric references and a possible overreliance on recent, unverified sources. Furthermore, cultural and regional differences in the adoption process should be explored to broaden its applicability.
4. The paper claims to employ quantitative methods, including regression analysis based on simulated data, but the methodology section only briefly describes the process of creating a simulated dataset using Python. The data generation assumptions, distribution model, specific variable sources, or code implementation details are not detailed. This results in poor reproducibility of the method and prevents readers from verifying the reliability of the simulated data.
5. The entire methodology relies heavily on theoretical constructs based on the TOE, TAM, and IS Success models, with simulated data used only as a supplement and no actual empirical validation. This makes the paper more like a literature review of the conceptual framework than a rigorous quantitative study, lacking methodological innovation.
6. The paper uses an OLS model in the regression analysis to predict user satisfaction, but the adjusted R² is only 0.248, explaining only 25% of the variance. This indicates that the model has weak explanatory power and may have omitted key variables such as ethical factors or cultural differences. The coefficients of other independent variables, such as cost reduction and productivity increase, are positive but insignificant, which the paper attributes to multicollinearity, but no detailed diagnostics of the VIF (variance inflation factor) or correlation matrix are provided.
7. Diagnostic statistics show no autocorrelation and normal residuals, but no heteroskedasticity tests or exogeneity validation are reported. This could lead to biased coefficient estimates, especially in simulated data.
8. The paper lacks actual experiments, comparisons with field tests or control groups, and relies solely on simulated regressions. This lacks external validity. The results are based solely on simulations and may not reflect real-world dark factory scenarios.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

I would like to congratulate you for the effort and the relevance of the topic you address. After reviewing your manuscript, I have identified several aspects that require improvement in order to strengthen the paper and make it more accessible for both scholars and practitioners. Please find below my detailed comments.

1. Originality and Relevance of the Topic

The subject is timely and well aligned with the objectives of Sustainability, considering the intersection of AI, ERP, and sustainability in the context of “dark factories.” The paper proposes an integrated conceptual framework (TOE, TAM, and IS), which represents a notable theoretical contribution. However, the practical applicability remains limited due to the lack of robust empirical data.

2. Introduction

The introduction presents an adequate conceptual framework by referring to various works that have employed TOE, TAM, and IS in the adoption of AI-ERP, and it also outlines the objectives of the study. However, a clear reminder of the study’s contribution is missing, and the references are not consistent with the IEEE style required by the journal.

3. Literature Review

The review is generally appropriate in my opinion, except for the references, which should be reformatted to follow IEEE style.

4. Methodology

I have identified the following issues:

The authors claim to have used secondary data from the literature and public reports, data mining from open sources, and a simulated dataset of 200 observations for the period 2020–2023. While integrating the TOE–TAM–IS models is a justified choice, the methodology raises several concerns:

• the dataset is simulated, not empirically collected;
• there are no details on the specific sources or validation process;
• the regression analysis (OLS) has limited explanatory power (R² ~ 0.25);
• the analysis period is very short—only three years.

Quality and Relevance of the Data

The data used for 2020–2023 appear more illustrative than representative. No major statistical sources are mentioned (e.g., IFR, Eurostat, World Bank). Many values (e.g., 30% cost reduction, 15–35% improvement in forecasting accuracy) seem to be taken generically from the literature, and no case studies or validations with real firms are provided. In other words, the paper tends to overgeneralize and claims applicability across multiple industries without sufficient evidence.

5. Conclusions

The conclusions should include a comparative discussion of the study with other research in the field, and the contribution of this work to the scientific context should be emphasized more strongly.

Final Recommendation

The paper has potential and makes an interesting theoretical contribution, but it requires a stronger empirical basis, clearer methodology, and improved presentation in order to be considered for publication in Sustainability.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf