Who among us would not like to have the gift of speed reading? While most of us must plod through documents laboriously, a fortunate few are able to scan texts much faster and even have photographic memories. But can any one individual read 19,000 sentences in a minute? Probably not—however a machine can. This figure is not plucked from the sky but is rather the verified read speed of a new artificial intelligence-based tool that IFC is developing to help its staff assess the environmental, social and governance (so-called ESG) performance of clients and potential investees.
The tool, called Malena, has been trained to analyze three million internal documents—impact assessments, news articles, and sustainability reports—stretching back decades and covering over 60,000 projects in 186 countries. While IFC’s ESG experts are more than capable of doing this initial filtering and analysis, AI can do it far faster and generate insights that enable more accurate risk assessments.
Malena’s creation comes as the market is witnessing many similar AI-driven innovations— all aimed at meeting the growing demand from investors for ESG performance data. For instance, credit rating agencies who previously ranked companies purely on financial performance are adding ESG metrics to their offerings and new rating agencies are emerging specialized in producing ESG rankings.
Corporates are eyeing the space intensely. Over 80 percent of the world’s largest firms by revenue now report on sustainability aspects of their businesses and nine in ten S&P 500 companies disclose ESG data. Governments are getting involved too. The United Nations has created a Sustainable Stock Exchanges Initiative to promote networking and research among capital market stakeholders. Regulators are developing standards to make ESG ranking methodologies more transparent. This should help make ESG data sets easier to understand and compare for non-specialists.
Malena works using a technology called natural language processing. First developed in the 1950s, it has become very sophisticated in recent years thanks to advances made by large tech firms like Google and Meta. IFC began building Malena in 2019 as a navigational aid to mapping out the ESG landscape. IFC’s ESG experts trained Malena to analyze language in a way that would generate the most useful ESG insights. And this is a key point to bear in mind: Malena and tools like it do not magically create themselves; they are made by people.
Human language is highly complex. Context is often key to understanding meaning and prevent errors of interpretation. A great deal of effort has gone into programming Malena to recognize language’s myriad nuances. For example, one could make an AI tool that extracts all references to ‘waste’ from a batch of documents. Unless trained otherwise, the tool might consider these references as negative indicators of a client’s environmental performance. But what if the word ‘waste’ was immediately followed by ‘management’. In that context, it may be describing a company’s efforts to manage waste in an environmentally sustainable manner—in other words, it’s a plus, not a minus.
Close attention must be paid to these kinds of nuances for Malena to be a truly useful, accurate tool. So far Malena has been trained to identify more than 1,200 ESG-related terms in unstructured text. Currently, it can read only English but other languages may be added to its repertoire. Malena generates dashboards on ESG performance which staff study when making assessments, which in turn inform investment decisions.
Although conceived as a purely inhouse tool, consideration is being given to widening its use. For instance, IFC recently teamed up with Amundi Asset Management, Europe’s largest asset manager, in a project aimed at validating Malena. Amundi tested it on a sample of their documents associated with more than 800 financial institution issuers of hard-currency debt in 60 emerging markets. Malena was able to validate Amundi’s in-house ESG scores and perhaps more significantly, added information for the 30 percent of the test samples for which Amundi had no scores previously. IFC continues to refine Malena for optimal performance and usability, with its official rollout planned for 2023.
As demand for higher quality ESG data grows exponentially, including in emerging markets where such data is often lacking, AI tools like Malena have an important role to play in improving data quality. They can help investors meet their ESG goals in innovative and nontraditional ways.