Daniele Cat Berro, Managing Director at MainStreet Partners

Artificial Intelligence (AI) has already been used across many sectors and industries  and in some cases has innovated radically the way data-crunching is executed. But its application in ESG data screening has been limited and only recently is emerging as a strong trend.

ChatGPT Catalyst

Regulators and practitioners readily agree that AI has the potential to modernise areas such as portfolio management, risk management and trading, enabling asset managers to analyse broader data sets more deeply.

ChatGPT, an open-AI, stole headlines in January for garnering a $10 billion investment from Microsoft, changing how it competes with Apple, Google and other tech giants. As the most advanced AI that can very effectively answer questions, write articles, summarize information, and more, all at the fingertips of anyone with an internet connection, investors have already been testing ChatGPT’s capabilities as a stock picker and data analysis tool.

For independent ESG advisory and portfolio analytics providers like MainStreet Partner, we see how this technological wave can advance the types of ESG data gathered and analysed, as well as potentially democratising the asset management sector.

Like all AI tools, the machine-learning is only as robust as the teaching inputs  its built upon. That said, with careful training AI could level the playing field for smaller asset managers and ESG data providers whose resources for ESG data gathering and analysis would otherwise be dwarfed by larger investment firms. This levelling could transform the competitiveness of ESG fund providers. At the same time is also true that several data providers, which all have been recruiting data engineers and data scientists in this space, are investing in it as a way to reduce the cost of data sourcing which will eventually replace a big proportion of their outsourced data gathering teams.

The European Securities and Markets Association (ESMA) has suggested that transparency should be increased in the ESG ecosystem, increasing ESG disclosures and market data quality. This can be achieved mainly thanks to technological solutions, in particular AI.

AI can make a great contribution in helping ESG data retrieval and verifying the quality of that data in the quickest, most cost-effective way.  Currently, a simplified process for ESG rating mainly works as follows:

1. Source raw data/KPIs from companies’ reports and/or surveys, which is partially automated.
   1. Input those data into a proprietary model.
   2. Qualitative overlay from analysts (this does not always happen since there are a lot of pure quantitative models out there that seek to restrict human bias)
2. controversies and reputational problems related to the companies are taken into consideration as part of the ESG analysis or/and as an overlay.

In terms of data retrieval from companies, AI offers great potential in the first part of the process, drawing on ‘domain specific data’, while its ability to organise and prioritise information from ‘unstructured data’ can be applied to the reputational element at the end of the process.

So, what are the two key types of AI data retrieval?

• Domain specific ESG data retrieval, in which relevant ESG information is gathered from companies’ reports. Data within those are integral parts of a company’s ESG assessment, as required by the EU Sustainability Regulation (e.g. PAIs). Currently, most essential ESG information within companies is held in a fragmented, de-centralised way that is not conducive, at this point, to being collated and reported by AI.
• Unstructured ESG data retrieval, in which information from can signal companies’ ethical, environmental and governance behaviour. There are already a few providers, , which provide ESG evaluations based on data sourced via unstructured data.The ‘unstructured data’ part is already quite developed, while the domain specific data part is a little more complex, though many players and investing in it. AI can make a great contribution in helping data retrieval and checking data quality, increasing credibility in the industry and helping to identify companies that are really going in the right direction.

However, an inherent risk of this is that AI-based data decision making could drive ESG asset managers to invest in the same companies, creating a monopoly or oligopoly. This would drive up company valuations and create a structural bias in the portfolio allocation. This would ultimately lead to a highly concentrated ESG environment, increasing volatility and penalising the Sharpe ratio. Should this happen, then the industry will have failed to harness AI technology in a way that ESMA and others hopes, revitalising criticism of ESG screening.

 Conclusion

The industry is not yet in the position to outsource ESG analysis to an AI, but this is an inevitable direction of travel. Many asset managers are still in early stages—exploration and prototyping—with many emerging technologies, including AI. A study by Accenture found that 95% [1] of respondents said that an asset manager’s technology, data and digital capabilities will be differentiators in 2025, 72% [2] of asset managers do not view themselves as leading firms regarding their digital maturity.

There is a clear need from the industry for high-quality and high speed of ESG data collection, as well as expectations from regulators (national and international) that technological advancements be harnessed for operational, cost and transparency benefits. Evidence from various studies [3] shows AI is a very useful tool in portfolio management, but one that requires close human supervision.

[2] https://www.accenture.com/_acnmedia/PDF-154/Accenture-Future-of-Asset-Management.pdf

[3] https://www.refinitiv.com/perspectives/market-insights/how-ai-and-big-data-are-reshaping-asset-management/

Getting that balance right and ensuring standardisation across the investment management industry will be instrumental in harnessing AI for good.