The SustainabilityPlus model is built on 3 founding pillars:
Transparency
The use of only verified open-source data
Consistent application of scoring principles and parameters on a peer-to-peer basis
SustainabilityPlus provides an industry-leading approach to assessing corporate sustainability performance. Unlike traditional sustainability frameworks, which may rely on self-reported data, SustainabilityPlus employs a combination of artificial intelligence (AI), machine learning, and proprietary methodologies to analyse companies across multiple sustainability dimensions. The methodological framework used in determining sustainability scores references academic principles and follows mathematical and statistical methods.
SustainabilityPlus employs a multi-layered AI-driven methodology to assess companies across 170 sustainability metrics distilled down to 66 individual AI prompts, combining quantitative analysis with qualitative evaluation. Initially, the system integrates three leading large language models (LLMs) to aggregate and qualify publicly available data across the 170 sustainability metrics into a single, company-specific data set. The data set is refined through agentic workflows and embedding techniques, ensuring accuracy, and minimizing hallucinations. The data set is then applied to a scoring framework that utilizes the same three leading LLM’s in order to establish the individual company scores. The scores undergo an automated QA process and are validated through statistical outlier detection, creating a robust framework aligned with established performance indicators aligned with triple bottom line principles.
The model focuses on public companies and uses a combination of data taken primarily, but not exclusively, from the following:
Regulatory submissions
Regulator statements
Corporate filings
Company websites
Analyst or investor data and analysis sources
The model builds a picture of a company over a 3-year period. The datasets are split into primary and secondary frameworks. The primary framework includes data sections which form the PESGO framework as below: PRIMARY PESGO FRAMEWORK
Performance
Earth
Society
Governance
Operations
Three separate large language models (LLMs) are used to process and score the data, ensuring triangulation and minimizing bias. The AI models apply an agentic approach combined with embedding techniques to refine initial outputs.
To enhance reliability, the data is transferred to a proprietary private LLM for further verification. This step includes:
Identifying extreme values that may indicate anomalies.
Ensuring AI-generated insights are evidence-based.
Comparing company performance across industry benchmarks.
The model uses a further 3-layer checks and balances process during the verification process for total transparency:
Triangulation with LLMs
Three LLMs generate a core data set to be used by the same three LLMs to reason independent scores using prompts trained on data subsets.
Embedding Validation: Textual data is converted to vectors to identify inconsistencies. Statistical Validation Outlier Detection: Where $ \tilde{x} $ = median, MAD = median absolute deviation. Scores beyond $ |Z| 3.5 $ trigger manual review.
Agentic Refinement: Human-AI collaboration ensures contextual alignment.
Industry Specific data: SustainabilityPlus utilizes industry specific data points as the model is built to understand areas that are particular sensitive to areas such as sentiment, reputation, stakeholder impact and financial effect.
To enhance reliability, the data is transferred to a proprietary private LLM for further verification. This step includes:
Identifying extreme values that may indicate anomalies.
Ensuring AI-generated insights are evidence-based.
Comparing company performance across industry benchmarks.
The model uses a further 3-layer checks and balances process during the verification process for total transparency:
Cross-LLM concordance (≥85% agreement required)
Embedding-based anomaly detection
Private LLM outlier analysis Error Rate: Hallucinations reduced to 2% through agentic workflows.
SustainabilityPlus follows the principles of academic best practices including:
Statistical Validation:
Where $ \tilde{x} $ = median, MAD = median absolute deviation. Scores beyond $ |Z| 3.5 $ trigger manual review
Sector Benchmarking:
Scores are generated on a consistent peer to peer basis and reviewed periodically (quarterly for subscribers).
Composite Index Theory – Aligning with OECD guidelines for indicator weighting.
Triangulation Principle – Enhancing validity through multi-model consensus.
Statistical Methodology:
SustainabilityPlus maintains a commitment to transparency. The scoring process includes:
Identifying extreme values that may indicate anomalies.
Ensuring AI-generated insights are evidence-based.
companies can review scores through providing verifiable transparent data which can be utilized using the exact same process as described herein. Companies or individuals wishing to proceed with this process can contact SustainabilityPlus.
SustainabilityPlus represents consistent scoring and application based on publicly available information. It is not intended as a financial resource, ratings agency, or investment tool. It provides listings of open-source data to build a picture of any given company with consistency of reporting. Companies do not provide the data and are not consulted regarding it.
Please refer to the Terms and Conditions of the SustainabilityPlus usage and subscription on the website for further details.
The SustainabilityPlus model is built on 3 founding pillars:
SustainabilityPlus provides an industry-leading approach to assessing corporate sustainability performance. Unlike traditional sustainability frameworks, which may rely on self-reported data, SustainabilityPlus employs a combination of artificial intelligence (AI), machine learning, and proprietary methodologies to analyse companies across multiple sustainability dimensions. The methodological framework used in determining sustainability scores references academic principles and follows mathematical and statistical methods.
SustainabilityPlus employs a multi-layered AI-driven methodology to assess companies across 170 sustainability metrics distilled down to 66 individual AI prompts, combining quantitative analysis with qualitative evaluation. Initially, the system integrates three leading large language models (LLMs) to aggregate and qualify publicly available data across the 170 sustainability metrics into a single, company-specific data set. The data set is refined through agentic workflows and embedding techniques, ensuring accuracy, and minimizing hallucinations. The data set is then applied to a scoring framework that utilizes the same three leading LLM’s in order to establish the individual company scores. The scores undergo an automated QA process and are validated through statistical outlier detection, creating a robust framework aligned with established performance indicators aligned with triple bottom line principles.
DATA SOURCES
The model focuses on public companies and uses a combination of data taken primarily, but not exclusively, from the following:
The model builds a picture of a company over a 3-year period. The datasets are split into primary and secondary frameworks.
The primary framework includes data sections which form the PESGO framework as below:
PRIMARY PESGO FRAMEWORK
The model builds protocols around scoring parameters based upon inputs.
All scores are based upon a consistent approach across the peer group
DATA AGGREGATION AND VALIDATION
Three separate large language models (LLMs) are used to process and score the data, ensuring triangulation and minimizing bias. The AI models apply an agentic approach combined with embedding techniques to refine initial outputs.
Triangulation with LLMs
LLM Validation and Refinement
To enhance reliability, the data is transferred to a proprietary private LLM for further verification. This step includes:
The model uses a further 3-layer checks and balances process during the verification process for total transparency:
Layer Verification:
ACADEMIC FOUNDATIONS
SustainabilityPlus follows the principles of academic best practices including:
Where $ \tilde{x} $ = median, MAD = median absolute deviation. Scores beyond $ |Z| 3.5 $ trigger manual review
Scores are generated on a consistent peer to peer basis and reviewed periodically (quarterly for subscribers).
SCORE TRANSPARENCY AND REVIEW
SustainabilityPlus maintains a commitment to transparency. The scoring process includes:
Appeals process – companies can review scores through providing verifiable transparent data which can be utilized using the exact same process as described herein. Companies or individuals wishing to proceed with this process can contact SustainabilityPlus.
DISCLAIMER
SustainabilityPlus represents consistent scoring and application based on publicly available information. It is not intended as a financial resource, ratings agency, or investment tool. It provides listings of open-source data to build a picture of any given company with consistency of reporting. Companies do not provide the data and are not consulted regarding it.
Please refer to the Terms and Conditions of the SustainabilityPlus usage and subscription on the website for further details.
© Copyright 2025 SustainabilityPlus. All rights reserved.
