Spot light on emissions calculations


With the growing interest in data calculation and disclosure of Scope 3 GHG emissions has cast a spotlight on emissions calculation methodologies. Among the array of methods employed by organizations, the spend-based approach stands out, albeit known for its time-consuming and resource-intensive nature. This article delves into an innovative avenue for streamlining the estimation of Scope 3 GHG emissions, leveraging Artificial Intelligence (AI) and Large Language Models (LLMs) to categorize financial transaction data, aligning it with spend-based emissions factors.

Understanding the Challenge of Scope 3 Emissions Calculation:

Scope 3 emissions, often termed indirect emissions, encompass greenhouse gas emissions (GHG) emanating from an organization’s value chain, beyond its direct operational control or ownership. These emissions, sourced from external entities like suppliers and customers, pose a formidable challenge for accurate quantification. A 2022 CDP study revealed that, for companies reporting to CDP, emissions stemming from their supply chain are on average 11.4 times greater than their operational emissions. Despite this, a staggering 72% of CDP-responding companies only report their operational emissions (Scope 1 and/or 2). Attempts to estimate Scope 3 emissions through manual data collection from suppliers are stymied by obstacles such as extensive supplier bases, complex supply chains, and resource-intensive processes.

Leveraging LLMs for Scope 3 Emissions Estimation:

An emerging approach to estimating Scope 3 emissions involves utilizing financial transaction data, particularly spend data, as a proxy for emissions associated with purchased goods and/or services. Converting this financial data into a GHG emissions inventory necessitates access to information regarding the GHG emissions impact of the products or services procured.

The US Environmentally-Extended Input-Output (USEEIO) framework provides a robust foundation for this endeavor. Utilizing a lifecycle assessment (LCA) framework, USEEIO facilitates the estimation of environmental consequences linked to economic activities by categorizing goods and services into 66 spend categories based on their environmental characteristics.

Similarly, the Eora MRIO dataset furnishes globally recognized spend-based emission factors, mapping inter-sectoral transfers across numerous sectors and countries. This dataset has been tailored to align with USEEIO’s categorization, simplifying the estimation process.

However, despite the potential offered by spend-based commodity-class level data, the manual mapping of vast financial ledger entries to commodity classes remains a daunting task. This is where LLMs emerge as a game-changer. These sophisticated NLP models, bolstered by recent advancements in the field, present a compelling solution for automating the categorization process, offering unparalleled efficiency and accuracy.

Illustrating the Framework:

Figure 1 depicts the framework for Scope 3 emission estimation employing LLMs, comprising four key modules: data preparation, domain adaptation, classification, and emission computation.

Exploring the Potential of LLMs:

Extensive experiments, encompassing cutting-edge LLMs such as roberta-base and bert-base-uncased, underscored the efficacy of fine-tuned LLMs in Scope 3 emissions estimation. Comparative analysis revealed significant improvements over traditional text mining techniques like TF-IDF and Word2Vec, positioning LLMs as a potent tool for grappling with the complexities of Scope 3 emissions accounting.

Integrating AI into IBM Envizi ESG Suite:

Harnessing the power of LLMs, IBM Envizi ESG Suite has incorporated an AI-driven feature, empowering organizations to streamline commodity categorization from spend transaction descriptions. This integration promises to revolutionize Scope 3 emissions calculation, overcoming barriers associated with manual mapping and accelerating the speed to insight.

In summary, the utilization of LLMs for Scope 3 emissions calculation represents a paradigm shift in sustainability data management. The successful integration of advanced LLMs into platforms like IBM Envizi ESG Suite holds the key to expediting GHG footprint assessments while simplifying the process for organizations worldwide.