AI-driven technologies present limitless possibilities, but work best when tempered with human judgement.
To effectively assess a corporate’s exposure to ESG risks and opportunities, asset managers are faced with the prospect of wading through a fast-flowing and unpredictable stream of underlying data. Given much of this data is unstructured, inconsistent and of questionable quality, it’s a time-consuming and costly task.
Many have effectively outsourced the number-crunching needed to ensure their portfolios match the sustainability expectations of asset owner clients by using ESG scores and ratings from third-party providers.
But over-reliance on third parties makes it hard to tailor solutions to the precise needs of asset owners in a rapidly-evolving environment, especially when accurate analysis requires a blend of qualitative and quantitative inputs.
More recently, asset managers have increased their use of artificial intelligence (AI) and machine-learning technologies for the speedy collation, handling and translation of unstructured company data.
By using basic AI and robotic process automation (RPA) to automate information-gathering, asset managers have more time to analyse streamlined information they know to be relevant to specific ESG-related issues that matter to their investors.
“AI and machine-learning [helps] asset owners and asset managers assess the long-term risks and rewards that [ESG-related issues] presents for their investments. What can be measured can be managed, and this is where AI and machine-learning can add particular value,” says Pooja Khosla, Vice President of Client Development for climate investing platform Entelligent.
Upskilling and innovation
But today’s AI-assisted tools are far from perfect, and shouldn’t operate in isolation.
Furthermore, the human element is still an integral part of corporate analysis and asset managers need to be upskilling their staff to better understand the data and technology they use, says Ashley Hamilton-Claxton, Head of Responsible Investment at Royal London Asset Management (RLAM). “Computers can’t read between the lines. They can still get it wrong,” she explains.
Nonetheless, it remains a market for innovation, with a plethora of AI-driven solutions being introduced by both third-party vendors and asset managers themselves to help deliver desired sustainable investing outcomes to asset owners.
Global data provider Arabesque S-Ray offers algorithmic tools to investors so that they can analyse the sustainability of their holdings at a portfolio and security level on a daily basis. The provider recently partnered with asset manager DWS and launched the DWS Concept ESG Arabesque AI Global Equity fund, which claims to capitalise on low interest rates, sustainability and digitalisation.
Meanwhile, BNP Paribas Securities Services’ Manaos platform partnered with Clarity AI to better enable institutional investors to store all fund data in one place, while further analysing whether, and to what extent, their funds contribute to the United Nation’s Sustainable Development Goals (SDGs).
A key use of basic AI technology is its ability to gather and sort corporates’ qualitative ESG-related information, including material contained in sustainability reports. To do this, asset managers commonly implement natural language processing (NLP) and sentiment analysis.
NLP is the technology used by Google Translate, offering immediate (and increasingly accurate) translations for a variety of different languages. Asset managers can use NLP to search through thousands of documents at a time, highlighting every instance in which a key word or phrase is used. These documents can come from the corporate itself or from third party content, such as news coverage.
For example, if an asset manager wanted to assess a corporate’s approach to decarbonisation, it could use NLP to identify each and every public mention by the corporate (or about the corporate) of ‘carbon’, ‘decarbonisation’, ‘net-zero’ and ‘emissions’.
“By layering all of those mentions together, asset managers can create a holistic view of how a company is actually addressing an ESG-related issue and then quantify it,” says Kevin Zacharuk, Product Manager of Textual Data at S&P Global Market Intelligence.
If the NLP tool reports that the majority of key words are attached to negative messaging – i.e. we don’t intend to commit to a net-zero target – the asset manager may decide to either engage with that corporate to encourage a change in approach or steer clear of the corporate altogether.
NLP is often used in tandem with sentiment analysis to further refine understanding. By assorting it into negative, positive or neutral, a sentiment engine allows analysts to generate a sentiment score.
If an asset manager wants to assess a corporate’s approach to mitigating climate risk, it can produce a sentiment score and measure that against the corporate’s competitors.
However, these technologies do come with their own risks, experts warn.
NLP could be collating information from sources promoting “fake news”, for example, or sentiment analysis could actually reinforce issues such as greenwashing, due to the predisposed bias of the individuals responsible for writing or implementing the programme.
Furthermore, the activities of large corporates like Amazon will be reported by a wider range of qualitative sources than a much smaller corporate that might be doing some really innovative things to tackle ESG-related issues, Hamilton-Claxton points out.
The margin for error will narrow over time, as AI technologies are honed, argues Warren Breakstone, Chief Product Officer for Data Management Solutions at S&P Global Market Intelligence.
“The ability to train algorithms, AI models and machines to produce the richest corpus of content is a key component of machine-learning. It gets better all the time,” he says.
Transparency forces accountability
A current problem with ESG-related information on investee companies – whether raw data or scores and ratings – is that it is largely self-reported. Being able to access all publicly available information about a corporate’s ESG-related performance, rather than just the information the corporate provides, gives asset managers a less biased picture of the company, experts say.
“Transparency around ESG-related data is key, particularly when ESG reporting requirements are still being defined,” Breakstone says. “[Asset managers] are now getting a more comprehensive view of a corporate from multiple sources.”
By making a wider range of third-party information sources more easily accessible to asset managers, NLP is forcing companies to hold themselves accountable by ensuring information they provide isn’t overly biased.
“Corporates need to tell us the truth, because we’re going to find out anyway,” RLAM’s Hamilton-Claxton warns.
If there is too much variance between the sentiment of the corporate’s disclosure versus third-party data, then the asset manager is going to question them, she explains.
“Corporates have little control over what is being said about them, and so they need to be very aware of the information that’s out there and be able to explain it,” Hamilton-Claxton adds.
Human versus machine
Human judgement is still incredibly valuable in the era of AI-enabled ESG analysis. Particularly when it comes to weighing up the relative importance of ESG-related qualitative and quantitative information AI-driven programming has isolated and organised, and then incorporating it into the asset manager’s decision-making.
Zacharuk argues that this is because “there are components of a fundamental analyst’s role [in ESG-related analysis] that AI can replicate but not fully replace”.
Human analysts can be far more responsive to investors’ changing demands. After all, they are the ones who will need to figure out how to balance the impact corporates have on marine life against other overarching biodiversity performance factors, for example. Analysts are able to analyse information with these new priorities in mind instantaneously.
Comparatively, AI models would first need to be reprogrammed to take these changes into account, which takes more time. Aged data isn’t as useful.
“Human interpretation of ESG-related data is valuable because I don’t think a computer can always get that judgement right on its own yet,” Hamilton-Claxton says.
While human subjectivity allows for increased flexibility when it comes down to the number-crunching, a lack of visibility of methodologies used by analysts has been cause for concern.
Previously, ESG Investor highlighted the lack of transparency of third-party vendors’ methodologies when providing corporates’ with ESG scores and ratings. Different vendors have different scores for the same companies, meaning asset managers and other end-users have to perform cross-analysis on the data they purchase in order to make a best-informed investment decision.
“Asset managers end up with multiple sets of data on the same topic. They face a lot of heavy lifting in order to figure out what is actually accurate,” says Vicky Sanders, Global Head of Investment Analytics at global institutional investment network Liquidnet.
If an AI-driven programme is ever going to be the preferred solution to quantitative analysis, then the overall quality of ESG-related data needs to improve, too, says Khosla.
“Both quality and quantity of the input data matters a lot in AI and machine-learning models, as models are only as good as the data you feed them,” she explains.
Finding the right balance
For now, the best and most accurate investment analysis involves a blend of AI and machine-learning technologies with human analysis.
As Breakstone points out to ESG Investor: “A machine will always beat the person, but a person and a machine will always beat the machine.”
Asset managers largely recognise that AI technology enhances human capabilities, rather than overriding them.
It gives asset managers the ability to cast a wider net when hunting down ESG-related information, traversing different sources and languages. But human analysts have the final say as to what it all means. This is likely to remain the case, even as developments in disclosure standards and requirements yield more data, requiring comparatively less interpretation of company narratives on ESG themes.
“The human component is absolutely essential, despite ongoing advances in technology. AI is designed to support, validate and improve the applications of fundamental data that relies on human analysts,” Khosla agrees.
As asset managers grow more comfortable with applying AI and machine-learning programming to their decision-making process, these technologies will bleed into other facets of ESG-related data identification, collation and quantification.
“That technology already exists. It just hasn’t been applied in the right ways yet,” Sanders says.