cs.AIApr 5, 2026
SaveAI Agents for Sustainable SMEs: A Green ESG Assessment Framework
Organizations: Faculty of Information Systems, University of Economics and Law · Vietnam National University, Ho Chi Minh, Vietnam
Abstract
This study presents a novel, AI-driven framework for assessing Environmental, Social, and Governance (ESG) performance in European small and medium-sized enterprises (SMEs). An initial phase established expert-validated ESG baseline scores from a subset of the Flash Eurobarometer FL549 survey data. In the second phase, a scalable AI agent system, built on the n8n automation platform, applied these baselines to perform automated ESG classification and generate contextual recommendations using large language models (LLMs). The results demonstrate the AI system's high consistency with human-derived outputs, thereby supporting more effective monitoring and intervention strategies aligned with the European Green Deal.
Figures & tables
Hình 1: A Human-AI workflow for establishing an ESG baselines and performing automated ESG analytics.
Hình 2: An end-to-end AI-agent workflow for ESG analytics.
| Item | Specification |
|---|---|
| Task | Predict tier label (Weak/Average/Good/Excellent) per country–group pair. |
| Features | Questionnaire-derived normalized signals (question-level scores prior to aggregation). |
| Target | Tier labels induced by baseline quartiles (group-specific). |
| Model | Multinomial logistic regression (L2 regularization) as minimal supervised baseline. |
| Train/Test | Train on baseline subset; test on held-out countries under the same split regime. |
| Validation | RRSSV with 40% baseline resampling over seeds (protocol in Methodology). |
Bảng I: Planned learning-based baseline for ESG tier prediction (design specification).
Hình 3: Baseline score distributions for four groups: (a) Group 1 - Governance, (b) Group 2 - Energy and Circular Economy, (c) Group 3 - Biodiversity, (d) Group 4 - Climate Strategy. (Source: Authors)
| Group | Test | p-value |
|---|---|---|
| Governance | Shapiro-Wilk | 0.0000 |
| D’Agostino-Pearson | 0.0005 | |
| Energy & Circular Economy | Shapiro-Wilk | 0.0000 |
| D’Agostino-Pearson | 0.0045 | |
| Biodiversity | Shapiro-Wilk | 0.0001 |
| D’Agostino-Pearson | 0.0000 |
Bảng II: The statistical distribution tests of four groups
Hình 4: Classified ESG score distributions for four groups: (a) Group 1 – Governance, (b) Group 2 – Energy and Circular Economy, (c) Group 3 – Biodiversity, (d) Group 4 - Climate Strategy. (Source: Authors)
| Group | Classification | Baseline | Workflow | Diff. |
|---|---|---|---|---|
| Governance | Weak | 2.033 | 1.997 | -0.036 |
| Average | 3.279 | 3.194 | -0.085 | |
| Good | 5.112 | 4.945 | -0.167 | |
| Excellent | 7.504 | 7.584 | 0.080 | |
| Energy & Circular | Weak | 1.353 | 1.288 | -0.065 |
| Average | 1.976 | 2.074 | 0.098 |
Bảng III: Score differences between AI-Agent workflow and baseline for all groups
| Criterion | Anchors (1 / 3 / 5) | Scale |
|---|---|---|
| Relevance | 1: Generic; weak linkage to diagnosed gaps. 3: Mostly on-domain; partially tied to gaps. 5: Precisely targets evidenced weaknesses (score/tier + salient signals). | 1–5 |
| Actionability | 1: Not actionable; lacks implementable steps. 3: Some steps but missing feasibility/prioritization. 5: Specific, feasible actions with clear levers and prioritization. | 1–5 |
| Faithfulness | 1: Adds unsupported facts or contradicts inputs. 3: Minor extrapolation; mostly consistent with inputs. 5: Strictly evidence-bounded; no unverifiable claims; consistent with inputs. | 1–5 |
Bảng IV: Expert rubric for evaluating LLM recommendations (1–5 ordinal).
| Nation | Score | Feedback | Suggestion |
|---|---|---|---|
| BG | 6.751 | Scored only 6.75/10. Fair but can improve. | Bulgaria should focus on enhancing incentives for renewable energy investment and improving energy efficiency in industrial sectors. The government needs to develop financial and technical support policies to promote green transition and strengthen international cooperation to learn and attract resources. |
| LV | 7.280 | Scored only 7.28/10. Fair but can improve. | Latvia should enhance policies supporting green businesses and promote investment in renewable energy. The government should consider providing financial and technical incentives to encourage the shift to more sustainable technologies and practices. |
| RS | 6.466 | Scored only 6.47/10. Fair but can improve. | Russia needs to improve the effectiveness of current policies and increase investment in green infrastructure. The government should promote renewable energy projects and implement energy-saving programs for households and businesses. |
| SK | 7.279 | Scored only 7.28/10. Fair but can improve. | Slovakia should strengthen financial support policies for renewable energy projects and improve energy efficiency in buildings. Also, SK should enhance international cooperation to learn from best practices and attract investment in advanced green technologies. |
| US | 6.996 | Scored only 7.00/10. Fair but can improve. | US is recommended to focus on increased investment in renewable energy projects and green infrastructure development. Stronger incentive policies are also needed to encourage businesses and individuals to adopt sustainable solutions, thereby enhancing the nation’s position in the global green market. |
Bảng V: LLM-generated responses for selected countries in Group 2