Sustainability Scores Overview

Updated 28 January 2026

Sustainability scores are composite metrics that quantify environmental, social, and governance performance based on diverse, multi-dimensional data sets.
They employ methods such as normalization, multi-criteria aggregation, and hierarchical weighting to consolidate raw impact metrics from LCA, regulatory disclosures, and real-time data sources.
Their application in finance, procurement, and product certification provides transparent, actionable insights for decision-making and risk assessment.

A sustainability score is a quantitative or categorical evaluation that summarizes the environmental, social, and sometimes economic or governance performance of an entity, product, process, or policy. Sustainability scores serve as decision-support tools by aggregating complex multi-dimensional data—often using methods such as classification, multi-criteria aggregation, normalization, and weighting—into a compact, comparable metric. This article delineates the principal methodologies, taxonomies, normalization strategies, aggregation architectures, and interpretations underlying sustainability scores, providing an academic overview grounded in the primary arXiv literature.

1. Conceptual Foundations and Taxonomies

Sustainability scores encapsulate multiple dimensions of sustainability, with the particular axes determined by application domain and available data. Dominant frameworks include:

Environmental-Social-Governance (ESG) Scores: ESG ratings are pillar-based, typically comprising environmental (E), social (S), and governance (G) (or ESGM with an explicit “missing information” pillar (Sahin et al., 2021)).
Product-Label and Certification-Based Scores: Scores can be derived from third-party (Type I) and private (Type II) product sustainability labels, as in GreenDB’s consumer-goods taxonomy (Jäger et al., 2022).
Task- or System-Oriented Multi-Criteria Scores: These may incorporate technical, economic, social, and environmental dimensions (e.g., the Sustainability Impact Score for software architectures (Fatima et al., 28 Jan 2025); CONFARM’s conflict-mapping ratios for multi-pillar impacts (Chakrabarti, 12 Dec 2025)).
Finance- and Market-Driven Metrics: Market-implied sustainability scores (e.g., SMIS) quantify sustainability based on the revealed preferences of regulated sustainable investment funds (Giacometti et al., 23 Oct 2025), while SDG-alignment scores link company behavior with the UN Sustainable Development Goals via automated pipelines (Hu et al., 2023).
Custom Application Domains: Sustainability scores are developed for supply chains (Stütz et al., 2023), procurement, high-entropy alloy design (Nominé et al., 29 Dec 2025), battery chemistries (Sangsinsorn et al., 2024), ML model reporting (Jouneaux et al., 25 Jul 2025), and decentralized governance (Meneguzzo et al., 21 Jan 2026).

Taxonomic granularity ranges from binary/ordinal (label present or not, A–E grades, etc.) to real-valued continuous scores, and can be tailored to be cross-sectional (fixed time) or spatiotemporal (dynamic clustering on sustainability and spatial metrics (Morelli et al., 2024)).

2. Metrics, Data Sources, and Normalization

Underlying sustainability scores is a diverse set of raw metrics, including:

Raw Impact Metrics: CO₂ eq, water use, land use, waste, energy, supply risk indices, recycling rates, governance attributes, and social indicators, typically collected via LCA datasets, regulatory disclosures, satellite/KG scraping, or direct measurement (Stütz et al., 2023, Sangsinsorn et al., 2024, Nominé et al., 29 Dec 2025, Jouneaux et al., 25 Jul 2025).
Categorical Labels: Type I/II product labels (Jäger et al., 2022), Eco-Score, Nutri-Score (Druschba et al., 2023, Druschba et al., 2023), SDG alignment classes (Hu et al., 2023).
Sentiment/Social Indicators: NLP-derived polarity scores from web, news, and social media for ESG sentiment (Patel et al., 2023).

Normalization strategies are highly context dependent:

Min-Max/Decile/Percentile Scaling: Scores are frequently mapped to unit intervals or deciles to enable aggregation across indicators with disparate units (e.g., supply risk, carbon footprint) (Nominé et al., 29 Dec 2025, Sangsinsorn et al., 2024).
Z-Score Standardization: Widely used in financial ESG (e.g., Refinitiv Asset4) for cross-company comparability (Chen, 2023).
Hybrid/Hierarchical Normalization: When multi-level raw data exist (700+ KPIs aggregated to 10 categories, then to overall ESG as in Refinitiv), normalization is recursive (Chen, 2023).
Imputation for Missing Data: ESGM introduces an explicit missing-data pillar, with scores assigned based on the empirical percentile of disclosure completeness (Sahin et al., 2021).

3. Aggregation Architectures and Formulae

Aggregation mechanisms for sustainability scores range from simple means to highly structured multi-level schemes. Notable examples include:

Approach	Formula/Procedure	Domain Example
Unweighted Mean (w/ tie-breaker)	$S = \frac{S_{nut} + S_{env}}{2}$ ; tie favors nutrition	Scale-Score food label (Druschba et al., 2023, Druschba et al., 2023)
Fixed Weighted Sum	$S_{overall} = \sum w_i S_i$ , sector-specific w	Refinitiv ESG (Chen, 2023); SMIS weights (Giacometti et al., 23 Oct 2025)
Multi-Pillar Convex Combination	$S = \sum w_p S_p$	ESGM w/ optimized pillar weights (Sahin et al., 2021)
Decision-Conflict Ratio	$R_j = N_j/(P_j + N_j)$	CONFARM multi-criteria score (Chakrabarti, 12 Dec 2025)
Multi-KPI Summation	$SIS_{d1,d2} = \sum_{i=1}^n \sum_{j=1}^m (P_{d1,i} + P_{d2,j}) \cdot Impact_{ij}$	Software architecture (Fatima et al., 28 Jan 2025)
Model Score Vector Ensemble	$p(c\|x;\theta)$ softmax over $c$ labels	Product label classification (Jäger et al., 2022); SDG alignment (Hu et al., 2023)
Market-Implied Differential	$SMIS_i = p_{9,i} - p_{U\backslash 9,i}$	Fund flow-based sustainability (Giacometti et al., 23 Oct 2025)

Weighting schemes may be fixed, sector-optimized (as in ESGM), or adaptively selected for maximum risk correlation (Sahin et al., 2021). Some methodologies, especially procurement and multi-dimensional risk, advocate “highest-precision” data first, falling back to coarser estimates where unavailable (Stütz et al., 2023).

4. Empirical Performance and Use-Cases

Sustainability scores are typically validated and applied in the following scenarios:

Classification and Prediction: ML models (ensemble trees, GCNs, R-GCNs) are used to predict categorical or ordinal sustainability labels, achieving test F₁ scores up to ≈0.96 for consumer products (GreenDB (Jäger et al., 2022)), or micro average F₁ = 0.89 for SDG alignment (Hu et al., 2023).
Portfolio and Procurement Decision-Making: Composite scores are used in supplier selection and procurement optimization, with scoring pipelines automating LCA through to dashboard integration (Stütz et al., 2023).
Financial Asset Screening and Backtesting: Stratified tilting to high-sustainability portfolios (SMIS or ESGM) can yield higher risk-adjusted returns than naive ESG weighting (Giacometti et al., 23 Oct 2025, Sahin et al., 2021).
Design and Governance Evaluation: In software architecture, Sustainability Impact Scores structure trade-off analysis among technical, environmental, social, and economic QAs (Fatima et al., 28 Jan 2025). DAO sustainability is operationalized via 0–3 scoring of participation, funds, efficiency, and decentralization, summed to a 0–12 composite (Meneguzzo et al., 21 Jan 2026).
Spatial and Spatiotemporal Clustering: Sustainability scores are leveraged for regional, sectoral, and temporal profiling through multi-matrix clustering, revealing dynamic ESG performance landscapes (Morelli et al., 2024).

5. Transparency, Explainability, and Critique

Transparency is paramount in sustainability scoring. Most frameworks aim for:

Algorithmic Transparency: Publishable formulas, vector weights, and mapping rules are standard (e.g., Refinitiv’s sectoral weights can be empirically recovered and validated, with R² > 0.99 (Chen, 2023)).
Explainability Layers: LIME (feature importance), GNNExplainer (graph rationales), and sub-score breakdowns for each dimension (e.g., DAO Portal (Meneguzzo et al., 21 Jan 2026), SDG scoring dashboard (Hu et al., 2023)).
Missing Data Handling: The “Missing” (M) pillar is mathematically explicit (percentile of missingness), prevents low-disclosure firms from being misclassified as low merit, and supports optimization for ESG–risk alignment (Sahin et al., 2021).

Critiques and open issues include:

Potential to Obscure Individual-Indicator Weaknesses: Simple means or sum aggregation may dilute extreme negative performance in a single pillar (Chen, 2023).
Reliance on Third-Party or Self-Declared Data: Green-washing risk is significant in Type II labels, and in sustainability reports scraped for ML scoring (Hu et al., 2023).
Class Imbalance and Score Volatility: Rare extreme classes depress macro-F₁ even as micro-F₁ remains high (Hu et al., 2023).
Market vs. Agency Divergence: SMIS (fund manager flow) shows low correlation with static agency ESG (cross-plots “elephant ears”) (Giacometti et al., 23 Oct 2025).

6. Advanced and Emerging Methodologies

Recent research expands sustainability scoring into new domains and methodological spaces:

Spatiotemporal and Hybrid Clustering: Multi-matrix hierarchical clustering leverages both spatial (geographical) and temporal (multi-year ESG time series) matrices for cluster formation, parameterized by convex combination weights (Morelli et al., 2024).
Ordinal and Multi-Level Scoring: Conflict-mapping frameworks such as CONFARM map each design decision and its cross-pillar impacts to a sustainability ratio in [0,1], facilitating convergence checks for system-scale evaluation and benchmarking across sectors (Chakrabarti, 12 Dec 2025).
High-Dimensional Material Screening: Systematic sustainability rankings for high-entropy alloy design employ multi-criteria (LCA, supply risk, ESG, companionability, reserves) aggregation and shortlist only the resilient top 5% for further investigation (Nominé et al., 29 Dec 2025).
ML Model Sustainability Reporting: YAML-based model cards formalize energy, water, and carbon footprints per ML model and task, establishing the foundation for future composite scoring and SLA integration, though with no immediate aggregation/reduction to a single index (Jouneaux et al., 25 Jul 2025).

7. Future Directions and Best Practices

Active debates persist around:

Weight Optimization and Non-Linearity: There is movement towards data-driven or risk-maximizing weight vectors (ESGM), potential use of geometric means to penalize very low sub-scores, and adaptive, sector-specific overlays (Sahin et al., 2021, Chen, 2023).
Integration with Decision Processes: The trend is towards modular, auditable pipelines (with open scripts/configs), real-time dashboards for operational domains (procurement, DAO governance), and MCDM frameworks (e.g., TOPSIS with LLM-generated score tables) for policy support (Bina et al., 13 Feb 2025).
Mitigation of Green-Washing and Opaque Disclosures: Explicit detection strategies, adversarial text classifiers, and robust imputation of missing data are recurrent recommendations (Sahin et al., 2021, Hu et al., 2023).
User Customization and Explainability: Progressive interfaces let users adjust pillar or label weightings, reveal sub-score panels, and access textual/graph evidence used in scoring (Druschba et al., 2023, Hu et al., 2023, Meneguzzo et al., 21 Jan 2026).
Standardization and Interoperability: YAML/JSON-LD schemas for sustainability-card integration (in ML and e-commerce) are gaining traction as vehicles for automated model comparison and ecosystem-level analytics (Jouneaux et al., 25 Jul 2025, Jäger et al., 2022).

In summary, sustainability scores represent a rapidly diversifying class of composite metrics architected for multidimensional assessment, high-throughput evaluation, and actionable transparency in domains ranging from financial ESG to consumer products, supply chains, battery chemistries, software architectures, and automated policy support. Ongoing innovation in metric definition, normalization, explainability, aggregation, and risk alignment will be crucial to their evolving role in regulatory compliance, investment, operational optimization, and scientific discovery.