This paper outlines a project vision and proposal submitted to a Romanian national grant competition on sustainability. The objective is to engineer an ESG (Environmental, Social and Governance) Knowledge Management capability that leverages multi-modal knowledge representation - BPMN diagrams enriched by an ESG-specific semantic layer, converted to knowledge graphs and exposed to LLM services through a Graph RAG architecture. A proof-of-concept of technological readiness level 4 is planned to be developed, to demonstrate the streamlining of the different knowledge representation modes and to enable evaluative experimentation. The research is framed as a Design Science research project due to its artifact-building nature and practical problem orientation. The problem pertains to a need for a Knowledge Management capability identified through several regional workshops discussing ESG requirements and challenges for local small and medium enterprises. Data analytics tools and reporting templates are typically invoked in such contexts, but most are repurposed legacy tools, lacking a granular mapping on enterprise architecture layers or business operations. Such mappings are relevant because ESG injects new architecture elements (new ESG-oriented roles, processes, documents), new semantics (new task types, event types, flows), new attributes (e.g. carbon footprint) which are all in the scope of enterprise modeling - if not at architecture level, at the very least at business process management level. LLM agents must become aware of such semantics to tailor their content or recommendation generation for enterprise knowledge already available in BPMN and other types of enterprise models. However, legacy enterprise modeling methods are semantically insufficient to capture crosscutting ESG concerns, so a knowledge engineering effort is also included in this project vision, to extend BPMN with the domain-specificity of ESG and to achieve a “meet-in-the-middle” point between ESG accounting requirements and Generative AI content.
Participation in ESG-focused workshops organized by industry clusters in the Cluj-Napoca, Romania
[
The regional workshops highlighted the growing demand for advanced ESG management tools that go beyond the scope of data collection and reporting, towards managing the intricate dependencies involved in ESG data provisioning through cross-cutting concerns across layers of enterprise and business process architectures.
ESG reporting (also known as sustainability reporting) is becoming a mandatory requirement for
companies. The reporting effort is complex, requiring the orchestration of data sources, new work
procedures, rules and multi-criterial decision-making. ESG reporting is defined at international level
(through standards and regulations such as International Accounting Standards IAS/IFRS, Directive
2014/95/EU (NFRD), Directive 2464/2022 – CSRD, the Paris Agreement, B Corp Certification, etc.)
and at national level it involves the incorporation of European ESG directives into national laws. In
Romania, this adaptation includes the Ministry of Public Finance Order no. 85 (January 12, 2024) [
This national framework triggered many regional workshops with local IT service providers
raising both awareness and requirements. In our previous reports [
In the traditional practice of Business Process Management (BPM), organizations consider factors
like time, cost, efficiency and quality. However, these factors do not directly ensure sustainability –
business processes must be augmented with dedicated task types, execution roles or data object
taxonomies to serve ESG goals that are additional to the main process goal. A new dimension has been
added to BPM, leading to the emergence of Green Business Process Management (Green BPM) [
For technological platforms, Generative AI can play a key role in helping organizations to design and execute a structured, content-driven ESG roadmap enabling better decision-making, enhanced reporting, and alignment with sustainability goals [4]. Therefore, our project vision addresses the need to integrate ESG capabilities into business process descriptions, while at the same time exposing such enriched descriptions to Generative AI assistance - to better support process analysis, reporting and generally process-centric Knowledge Management. There seems to be a practitioner experience gap between ESG intentions and measurable outcomes due to fragmented knowledge, lack of traceability, and the complexity of ESG metrics. Our project vision seeks to bridge that gap by proposing the development of a hybrid AI platform that will leverage BPMN as a process description standard, ESG as a semantic enrichment layer (on metamodel level) and knowledge graphs as mediators to inform Large Language Model (LLM) services with process-centric ESG design decisions and semantic annotations.
The paper is structured as follows: this introductory unit is followed by related work comments focusing on the BPM lifecycle integrated with LLMs. The problem statement and research methodology are depicted in Sections 3 and 4, followed by an outline of preliminary efforts in combining BPM and a hybrid LLM-KG approach. The research vision is presented in Section 6, making explicit the relevance to CAiSE in Section 7, followed by conclusions.
ESG helps to assess how an organization manages risks and opportunities created by cross-cutting factors along three dimensions. The Environmental dimension measures a company’s impact on the natural ecosystem (carbon emissions, efficient resource utilization in production processes, pollution, waste, eco-friendly efforts and sustainable services). This dimension is influenced by the other two – Social, referring to the company's social relationships, both external (along supply chains or activities with local communities) and internal (with employees); and Governance, which concerns transparency and responsibility within company's operations.
Given the complexity of semantic intricacies between ESG factors, organizations face significant challenges in integrating and tracing ESG considerations into business processes. GenAI solutions are valuable in facilitating ESG data measurement, e.g. in areas like emissions tracking or assessing social aspects like gender diversity [4], however we advocate the use of GenAI beyond data analytics scenarios, since it can also handle conceptual (process) analysis [5].
The use of GenAI as a tool to assist business analysts and automation is demonstrated in [6].
Starting from diagrammatic visualizations (such as ER and BPMN diagrams), GenAI tools can provide
support throughout the software development lifecycle (SDLC). Our proposal does not focus on
SDLC, but rather on the BPM lifecycle [7]. The potential of integrating LLMs into each stage of the
BPM lifecycle was suggested by the agenda of Vidgof et al. [8]. Going beyond that call to action, this
research also employs metamodeling to enrich business process modeling with ESG aspects currently
not covered by any graphical modeling standard — consequently, a key contribution of this proposal
will be the extension of BPMN with distinctive ESG constructs. Early stage propositions are based
on the Bee-Up modeling tool2 as reported in [
Figure 1 describes the traditional BPM lifecycle and maps it to potential contributions of LLMs in the ESG context. In the Process identification phase, LLMs can analyze regulations and policies with the goal of identifying business processes into which ESG factors can be integrated. Grohs et al. [9] use LLMs in process identification and discovery from textual descriptions, their evaluation indicating promising performance. Including LLMs in the Process discovery phase helps on mapping ESG initiatives within existing business processes; at this phase, LLMs can be viewed as Process Mining tools [10] focused on discovering processes based on historical data.
Process analysis with the help of LLM can help with identification of ESG risks or greenwashing [11]. Process redesign with LLM support includes ESG policy recommendations (for example, reducing energy and water consumption in certain processes). Process implementation integrated with LLM may include an automatic generation of ESG reporting or guiding employees in executing manual work procedures through LLM-generated recommendations. Process monitoring with LLM provides suggestions based on ESG data analysis, monitoring compliance with current regulations, as well as evaluating ESG impact or generating predictions.
Another converging research stream investigates how GenAI models can parse BPMN diagrams either in graphical form or serialized through various export formats [5], which leads to our suggestion of employing semantic graphs as mediator and interchange format between a BPMN modeling tool and an LLM service. By extending existing standards like BPMN with ESG concepts, we also face the challenge of how well metamodel augmentations are received by LLMs since no training corpus is available for such narrow scope domain-specific modeling languages (DSMLs). We aim to experiment on this challenge by informing the LLM with various RDF-based process 2 https://bee-up.omilab.org/activities/bee-up/ vocabularies extended with ESG-specific semantics – e.g. new taxonomies of tasks, events, flows, data objects.
ESG evaluations must be primarily facilitated internally, providing organizations with tools for assessing and designing their own ESG policies tightly coupled to business operations. This motivates the objective of this research – the development of an innovative AI and BPMN-based platform for Knowledge Management for ESG reporting. Integration with GenAI will answer gaps such as what Minkkinen et al. highlight as a lack of responsible AI integration in ESG investment analysis [12].
The idea of engineering a platform that leverages BPMN and GenAI for ESG Knowledge
Management relates to the current state of widely adopted ESG tools [
Therefore, our project vision aims to repurpose and extend BPMN to complement the data-centric quantitative ESG practices with an interpretative platform based on hybrid AI interpreting ESGenriched business process models. Identifying and managing the interplay between ESG pillars, business processes and enterprise systems can lead to the Knowledge Management capability targeted by this project proposal: • •
The ESG-process dependencies are chains of relationships for which As-Is analysis and ToBe prescriptive process redesigns are required – therefore, we use the BPMN standard as a starting conceptual structure, to be extended with new taxonomies of tasks, events, decisions, roles, information flows and dependencies that capture the ESG perspective; GenAI tools are increasingly recognized as advisory services (recommendations) or for analysis (of private content). Hybridizing LLMs and knowledge graphs to optimize an LLM for an organization’s private content [18] can be further enhanced with diagrams inherited from traditional BPM practices [5,19,20].
By combining these two technological facets, this project proposal adopts an updated form of “Model-driven Engineering” (MDE) - promoted as “Semantics-driven Engineering” [21] due to the focus on knowledge streamlining between diverse knowledge processing environments. Instead of focusing on diagrams to generate software components or automate workflows, we extend the BPMN standard to inform GenAI assistants with metamodel-enriched process semantics that incorporate ESG concerns in the process description vocabulary.
The objective of the project proposal is to develop an intelligent platform based on AI and BPMN as support in Knowledge Management on the organization and tracing of ESG-related activities. The targeted requirements are to facilitate the design, documentation, analysis and optimization of business processes from the perspective of ESG concerns. The proposed innovative platform will consist of the following components: • • • • a graphical language (DSML) and the corresponding modeling tool that extend the BPMN standard [22] with ESG-specific concepts, dependencies and taxonomies, automated mechanisms for analysis, reporting and decision support through the interoperability of the extended BPMN+ESG diagrams with GenAI services (OpenAI services to be used for a demonstrator), interoperability mechanisms based on RDF graphs as a mediator between the modeling tool and the GenAI services, an integrated operating procedure to support a managerial capability that can be evaluated using Knowledge Management maturity standards such as APQC’s maturity model [23].
The work methodology is tailored for an artifact-building engineering effort that involves both "method engineering" in the sense of [24] and software development. The appropriate methodology for this type of scientific research is Design Science Research (DSR) [25]. Considering the specific task of extending an existing modeling language, the AMME methodology (Agile Modeling Method Engineering) [26] is also needed and will rely on the ADOxx metamodeling platform. Therefore, the project methodology reflects the hybridization between DSR and modeling method engineering, as suggested in Fig. 2.
The DSR methodology structures the project's approach into well-defined stages that also inform the work plan, which is inherently iterative, allowing for the gradual expansion of the developed artifact based on intermediate feedback. This requires two engineering-evaluation cycle iterations (WP3 and WP4) as the core of the work plan detailed below: • • • • • • • • • •
WP2. Consolidate solution requirements WP3.1. Design and development of the BPMN-ESG modeling tool WP3.2. Design and development of the model-driven AI-based KM platform WP3.3. Demonstration and evaluation of the integrated prototype Milestone 1 (Month 10) – Evaluated prototype on Technological Readiness Level 3 WP4.1. Design and development of the BPMN-ESG modeling tool WP4.2. Design and development of the model-driven AI-based KM platform WP4.3. Demonstration and evaluation of the integrated prototype
Milestone 2 (Month 20) – Working prototype on Technological Readiness Level 4
The formulation of this project proposal was preceded by several converging activities: As reported
in [
The KM4ESG project aligns to several topics of interest for the CAiSE community: • • • •
Artificial Intelligence including generative AI and Machine Learning: i.e. the application of AI techniques to enhance information systems; Knowledge Graphs: the use of Knowledge Graph as mediator to improve LLM interaction highlights the conference's interest in decision models and AI integration; Business process modeling, analysis and improvement: focusing on traditional BPMN and later incorporating ESG-specific elements (extending BPMN for ESG integration) ensures that business process descriptions are the core procedural knowledge for ESG concerns; Sustainability and social responsibility management: the integration of ESG into business process supports decision-making aligned with sustainability goals.
Additionally, the proposal connects directly to CAiSE’s call for the development and evaluation of artifacts that advance both theory and practice in information systems engineering. 3 https://www.ontotext.com/knowledgehub/fundamentals/what-is-graph-rag/ 4 https://adoxx.org/ 5 https://docs.ragas.io/en/stable/
AI can play a crucial role in ESG reporting, offering significant benefits for both companies and stakeholders [29,30]. GenAI automation not only reduces the volume of manual labor but also ensures the accuracy of reporting in compliance with current regulations [31].
This project vision proposes the development of a model-driven, process-centric, platform for Knowledge Management and an updated view of traditional knowledge flows (e.g. Nonaka’s conversion cycle [32]). The envisioned contributions target two areas: enriching domain-specific modeling by integrating ESG considerations into BPMN and improving sustainability management with the help of hybrid AI informed through a model-driven approach.
[4] EY, How Generative AI can build an organization’s ESG roadmap, 2025. URL: https://www.ey.com/en_in/insights/ai/how-generative-ai-can-build-an-organization-s-esg-roadmap/. [5] Dolha, D.N., Buchmann, R.A., Generative AI for BPMN Process Analysis: Experiments with Multi-Modal Process Representations. In: Perspectives in Business Informatics Research. BIR 2024. Lecture Notes in Business Information Processing, vol 529. Springer, Cham, 2024, pp. 1935. doi: https://doi.org/10.1007/978-3-031-71333-0_2. [6] Niculescu, V., Chisăliță-Crețu, M.C., Osman, C.C., Sterca, A., Model-Driven Development Using LLMs: The Case of ChatGPT. In: Proceedong of the 20th International Conference Evaluation of Novel Approaches to Software Engineering, Science and Technology Publications, Lda, Setúbal, Science and Technology Publications, Lda, Setúbal, 2025, pp. 328-339. doi: 10.5220/0013484400003928. [7] Dumas, M., La Rosa, M., Mendling, J., Reijers, H.A., Fundamentals of Business Process Management, Second edition. Springer, Heidelberg, 2018. doi: https://doi.org/10.1007/978-3-662-56509-4. [8] Vidgof, M., Bachhofner, S., Mendling, J., Large Language Models for Business Process Management: Opportunities and Challenges. In: Business Process Management Forum. BPM 2023. Lecture Notes in Business Information Processing, vol 490, Springer, Cham. pp. 107–123, 2023, doi: https://doi.org/10.1007/978-3-031-41623-1_7. [9] Grohs, M., Abb, L., Elsayed, N., Rehse, J.R., Large Language Models Can Accomplish Business Process Management Tasks. In: Business Process Management Workshops. BPM 2023. Lecture Notes in Business Information Processing, vol 492. Springer, Cham, 2023, pp. 453–465. doi: https://doi.org/10.1007/978-3-031-50974-2_34. [10] Van Der Aalst, W., Data Science in Action. Springer, Second Edition, Berlin Heidelberg, 2016.
doi: https://doi.org/10.1007/978-3-662-49851-4. [11] Gregory, R.P., When is Greenwashing an Easy Fix?, Journal of Sustainable Finance & Investment, 13(2), 2021, pp. 919–942, doi: https://doi.org/10.1080/20430795.2021.1907091. [12] Minkkinen, M., Niukkanen, A., Mäntymäki, M., What About Investors? ESG Analyses as Tools for Ethics-Based AI Auditing. AI & Society 39(1), 329–343, 2024. doi: https://doi.org/10.1007/s00146-022-01415-0. [13] Morning Sustainalytics, Official Website, 2025. URL: https://www.sustainalytics.com/. [14] LSEG, Environmental, Social and Governance Scores from LSEG, 2025. URL: https://www.lseg.com/content/dam/data-analytics/en_us/documents/methodology/lseg-esgscores-methodology.pdf. [15] Global Reporting Initiative (GRI), Official Website, 2025. URL: https://www.globalreporting.org/. [16] Sustainability Accounting Standards Board (SASB), Official Website, 2025. URL: https://sasb.org/. [17] World Economic Forum (WEF), Official Website, 2025. URL: https://www.weforum.org/. [18] Microsoft, Project GraphRAG – LLM-Derived Knowledge Graphs, 2025. URL: https://www.microsoft.com/en-us/research/project/graphrag/. [19] Chiş, A., Stoica, O.I., Ghiran, A.M., Buchmann, R.A., A Knowledge Graph Approach to Cyber Threat Mitigation Derived from Data Flow Diagrams. In: 2024 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR), pp. 1–6, IEEE, 2024. doi: 10.1109/AQTR61889.2024.10554074 [20] Niculescu, V., Chisalita-Cretu, C., Osman, C.C., Sterca, A., Vector Based Modeling of Business Processes. In: Proceedings of 20th International Conference Evaluation of Novel Approaches to Software Engineering, Science and Technology Publications, Lda, Setúbal, 2024, pp. 735–742. doi: 10.5220/0012739100003687. [21] Buchmann, R., Eder, J., Fill, H.G., Frank, U., Karagiannis, D., Laurenzi, E., Mylopoulos, J., Plexousakis, D., Santos, M.Y., Large language models: Expectations for semantics-driven systems engineering. Data & Knowledge Engineering 152, 102324, 2024, pp. 1-13. doi: https://doi.org/10.1016/j.datak.2024.102324. [22] Object Management Group (OMG), Business Process Model and Notation (BPMN) Specification.
Version 2.0.2, 2024. URL: https://www.omg.org/spec/BPMN/2.0.2/PDF. [23] APQC, APQC's Levels of Knowledge Management Maturity, 2025. URL: https://www.apqc.org/resource-library/resource-listing/apqcs-levels-knowledge-managementmaturity. [24] Brinkkemper, S., Saeki, M., Harmsen, F., Assembly Techniques for Method Engineering. In: Advanced Information Systems Engineering. CAiSE 1998. Lecture Notes in Computer Science, vol 1413. Springer, Berlin, Heidelberg, pp. 381–400. doi: https://doi.org/10.1007/BFb0054236. [25] Peffers, K., Tuunanen, T., Rothenberger, M.A., Chatterjee, S., A Design Science Research Methodology for Information Systems Research. Journal of Management Information Systems, 24(3), 45–77, 2007. doi: https://doi.org/10.2753/MIS0742-1222240302. [26] Karagiannis, D., Agile Modeling Method Engineering. In: Proceedings of 19th Panhellenic Conference on Informatics, pp. 5–10, 2015. doi: https://doi.org/10.1145/2801948.2802040. [27] Ramautar, V., España, S., The OpenESEA Modeling Language and Tool for Ethical, Social, and Environmental Accounting. Complex Systems Informatics & Modeling Quarterly, 34 (2023) 1-29. doi: https://doi.org/10.7250/csimq.2023-34.01. [28] Karmaker, S.K., Feng, D., TELeR: A General Taxonomy of LLM Prompts for Benchmarking Complex Tasks. In: Findings of the Association for Computational Linguistics: Empirical Methods in Natural Language Processing 2023, pp. 14197–14203. ACL, 2023. [29] European Institute of Management and Finance (EIMF), The Impact of Artificial Intelligence in Environmental, Social and Governance Reporting, 2025. URL: https://eimf.eu/the-impact-of-artificial-intelligence-in-environmental-social-and-governance-reporting/. [30] Dydon.ai, Agent AI: A New Era in Sustainable Finance for ESG and EU Taxonomy Reporting, 2025. URL: https://dydon.ai/agent-ai-a-new-era-in-sustainable-finance-for-esg-and-eu-taxonomy-reporting/. [31] Daněk, J., Pospíšil, J., How Generative AI Autonomous Agents Shape the Current Business Trends and Smart Learning, Trendy v podnikání - Business Trends 2024/2, pp. 35-46, 2024. doi: https://doi.org/10.24132/jbt.2024.14.2.35_46. [32] Nonaka, I., von Krogh, G., Tacit knowledge and knowledge conversion: controversy and advancement in organisational knowledge creation theory. Organization Science 20 (2009) 635–652. https://doi.org/10.1287/orsc.1080.0412.