Not all Artificial Intelligence can be relied on to produce meaningful sustainability analysis, writes Chandini Jain
ESG and sustainability research is a time-consuming endeavor that requires significant effort to gather, process and summarize large amounts of unstructured data. To date, there is no standardised way to capture and communicate a company's ESG performance, adding to the complexity that ESG teams must manage.
The use of ESG scores, which simplify a company's sustainability record into a single metric, has fallen out of favour over concerns about inconsistent methodologies, a lack of transparency, and the potential for greenwashing. Just this past August, S&P Global discontinued the use of ESG scores for assessing company credit quality in favour of dedicated analytical narratives.
As more firms end their reliance on ESG scores, they're faced with the daunting task of conducting their own research and monitoring to evaluate company ESG performance and to ensure their decisions align with their unique and publicly stated ESG criteria — and any relevant regulations.
That means highly compensated finance industry professionals are sifting through volumes of company-produced statements, biodiversity and deforestation data, regulatory and legal documents, supplier information, and local media coverage to assess a company's sustainability record and detect any greenwashing. Information overload — and the ability to source hard-to-get data — are universal ESG challenges.
Until recently, this was the choice that financial industry firms faced: Continue using flawed scores that meet no firm's unique requirements, or invest significant time and resources to conduct manual research in-house.
But generative AI presents us with a new and better way of conducting ESG research, one that's faster, more efficient and more comprehensive than the others. Because generative AI's strength is in its ability to summarise vast amounts of unstructured data, it promises to do the heavy lifting of finding relevant information, cleaning it, and generating ESG intelligence for us. The researcher's starting point would be custom-generated insights — rather than a meaningless score — so they can focus on high-level analysis and making decisions faster, such as whether or not to invest, engage or lend.
Unfortunately, generative AI has failed to deliver for ESG use cases in financial services for the same reasons it has failed for enterprise use cases across all industries: It has three fatal flaws.
The first flaw with generative AI is that it lacks comprehensiveness and timeliness. The large language models (LLMs) used by generative AI tools don't possess the domain-specific knowledge and access to the kinds of niche datasets required for a thorough ESG evaluation. This is a well-known issue as generative AI models cannot fully comprehend complex sustainability-related regulations or nuanced ESG issues and scenarios.
And because they lack access to real-time data, generative AI output is always out of date. This means recent developments, including news coverage, lawsuits or regulatory actions, can't be factored into any results. Here, generative AI shares a weakness with ESG scores: They're hopelessly backward-looking.
The second flaw with the LLMs that power generative AI tools is that they can't provide sources for the information they generate. This lack of transparency creates a trust gap and introduces investment, regulatory and reputational risks. For ESG reporting, it is crucial to link back to the source of the data. Similarly with company engagement, it is important to be able to back up any claims against the company with verified information.
Finally, generative AI has a well-known issue of producing misleading or outright fabricated responses. In any regulatory filing, data accuracy is extremely important. If the ESG team preparing a Sustainable Finance Disclosure Regulation report for an Article 9 fund needs to spend time verifying every response generated by AI — or worse, ends up using 'hallucinated' output — the benefits are lost and the risk of fines is unacceptably high.
The good news for finance industry ESG teams is innovation in AI is moving fast, and the flaws with LLM-based generative AI tools that have made them unsuitable for enterprise use cases have been addressed by a new AI technique called retrieval augmented generation, or RAG AI.
First developed by Meta, RAG AI is purpose-built for knowledge-intensive use cases that require sifting through large volumes of noisy data and demand a high degree of accuracy and trust. It combines the power of retrieval-based models that can access real-time information and industry-specific datasets such as sustainability reports, scope 2 supplier data or regulatory fines with generative models that are able to produce natural language responses. This ensures decisions are based on real-time data, allowing for faster decision-making without wasting time augmenting the output of generative AI tools with recent news and other developments.
Just as importantly, RAG AI-based systems can cite their sources — a critical requirement for regulatory reporting. The ability to maintain a trail of information sources and include them in reports means that the output can be used directly in regulatory filings, saving countless hours of analyst time. In addition, research teams can back up their ESG analysis with a clear understanding of which data sources contributed to the assessment and why, building trust in the process at a time when ESG investing faces intense scrutiny.
Finally, RAG AI can cross-reference information across multiple — and continuously updated — sources, which means research teams can easily verify the ESG claims made by companies against industry benchmarks, third-party reports, and actual reality on the ground. This can be a huge step forward in identifying attempts at greenwashing.
RAG AI offers ESG teams the opportunity to conduct their research in an efficient, comprehensive and accurate way so they can focus their time on high-level analysis and making decisions faster than they ever could before.
Chandini Jain is chief executive officer at data company Auquan, which uses AI to perform ESG analysis. This month, Auquan raised $3.5 million in a seed fundraising round. Click here to read Auquan's white paper: The Advantages of RAG AI (Retrieval Augmented Generation) Over Generative AI for Financial Services.