AI is a powerful tool for assessing ESG risk in portfolios but it could also lead to an incomplete view of a company’s true impact if not implemented properly, says Cordelia Dower-Tylee at EdenTree Investment Management.
As demand has grown to incorporate ESG risk into investment research, the excitement around artificial intelligence (AI) has increased exponentially. AI will revolutionise investment, offering new avenues into data collection, information processing, and financial analysis. Naturally, therefore, attention has turned to the role it plays in sustainable portfolios, and whilst it undoubtedly will feature, we believe investors should be cautious of placing too much reliance on AI to conduct ESG analysis.
Invaluable in helping standardise ESG information
AI certainly has a role to play within ESG analysis, as its computational power offers significant solutions to an array of challenges within the industry. Firstly, analysts are tasked with gathering huge quantities of data to aid their ESG assessments. Due to the breadth of ESG topics this involves numerous information sources, including company accounts, news flow, and academic research.
AI can be used to gather this non-linear data at greater speeds, permitting analysts to spend more time interpreting information, a more valuable use of their expertise.
AI may also prove invaluable in helping to standardise ESG information. Unlike financial accounts, ESG data is not uniform and is often unaudited, which makes comparing companies and sectors a highly manual process. AI also has a pivotal role in providing data estimation where there are gaps, and improving the consistency between datasets. For example, AI driven insights into site-based water scarcity, amongst other biodiversity indicators where the data is poor and the risk can be high, is highly valuable in enabling investors to appropriately assess the risk and opportunity.
Errors within carbon footprinting are a good illustration as it is common for companies to get mapped to the wrong parent company.
Finally, AI can open up new insights for analysts to consider. One such example is sentiment analysis, a method which looks at the emotional tone of a document to determine if it is positive, negative or neutral. This has powerful applications – for example, it could be used to draw conclusions on a company’s dedication to the low carbon transition by assessing tone, highlighting exaggeration, and analysing trends across years of disclosures.
Limitations of AI in ESG research
Despite these benefits, we caution that relying too heavily on AI to run ESG portfolios will likely lead to an incomplete picture of a company’s true impact. Aside from the well-publicised concerns of ‘hallucinations’, data quality, and privacy, a major limitation of AI in ESG research is that its outputs are often binary and decontextualised.
Analysis completed by AI can often lack nuance and lead to black and white conclusions which do not reflect a company’s true impact on ESG topics. ESG topics are often intangible and don’t easily conform to binary outcomes, and this is particularly true for the ‘S’ of ESG. Whilst carbon emissions and diversity may be easy to measure, it is much more difficult to assign a value to topics like employee wellbeing and company culture. Using AI to analyse such nuanced and qualitative areas may not capture the subtleties that can distinguish good risk management from poor risk management.
Perhaps above all else, outputs generated by AI still require human interpretation and oversight, as they are often inaccurate and may over- or understate the level of risk or impact. Errors within carbon footprinting are a good illustration as it is common for companies to get mapped to the wrong parent company. Without human correction of these errors, it could significantly alter the numerical outputs – a concern given carbon footprint data is frequently used as proxy for climate risk within funds. Put simply, human oversight and interpretation is critical to ensure AI outputs are accurate and applied appropriately.
Furthermore, ESG investment decisions do not take place in a vacuum. Rather, they are analysed as part of a wider context, something which may be overlooked by an AI-driven approach. A recent example is Enel, a utility with a highly credible transition plan, who recently missed their 2024 scope 1 GHG reduction target due to geopolitical influences and adverse weather conditions. Our human-driven approach allowed us to capture and consider these extenuating factors, and maintain conviction in a company which, despite a small fluctuation in their trajectory, remains highly committed to the low-carbon transition. An AI-led analysis is less likely to capture this nuance.
Finally, an AI-driven approach overlooks the role of engagement, a tool frequently used by ESG analysts to consolidate investment recommendations. Person-to-person dialogue is often most revealing about a company’s intentions, culture and progress, offering a rich assessment which AI can find harder replicate.
AI’s role: complementary not core to ESG analysis
It is clear that AI has a role to play within ESG research. Its proficiencies in synthesising and collecting data, and highlighting patterns is a powerful tool for any ESG investors’ toolkit. However, we remain cognisant of the risks that come with an investment approach that relies heavily on AI-driven recommendations. As such we view AI’s role to be complimentary to sustainable research rather than core to investment strategies. This is because ESG at its core is human; to unpack the complex and layered topics that form the ‘ESG’ acronym a human, a qualitative human lens is required.
*Cordelia Dower-Tylee is a responsible investment analyst at EdenTree Investment Management.
