Harnessing the power of AI and big data
Environmental Finance: What are FactSet’s AI solutions in the space?
Hendrik Bartel: Truvalue Labs, a FactSet company, focuses on harnessing AI and big data to provide financial professionals with actionable information on how companies are behaving pertaining to issues like carbon emissions, labour relations, data security, and product quality, to name a few. We leverage complex algorithms that sift through millions of data points (e.g., articles, blogs, social media posts, and legal filings) to help our users uncover ESG information hidden in over 100,000 sources of unstructured, third-party data in 13 languages.
The powerful AI behind our products results in over 300,000 signals generated monthly, as well as four key scores across 26 ESG categories and 16 United Nations Sustainable Development goals (SDGs). Our AI-driven analysis is distinguished by the fact that to avoid greenwashing it mines sources but does not consider companies’ own ESG statements.
Furthermore, analysis is produced in real time, thus enabling subscribers to gauge direction of travel for positive or negative ESG investments.
EF: Is AI the answer to ESG data challenges?
HB: No, AI is not the answer. AI and machine learning are computer science methodologies that allow us to understand context between disparate datasets or allow us to extract structured data from unstructured documents. AI is a set of technologies that can be applied to problems, allowing us to find new solutions.
Yes, AI is a tool in the toolbox but in this case, not the “be all, end all.”
EF: How exactly can AI allow investors to better understand ESG data and analyse risks and opportunities?
HB: Technology gives us the ability to build a more objective data process. It is also highly scalable and allows us to uncover insights in real time. The use of AI and machine learning benefits the capital markets industry by providing more efficient ways to mine massive amounts of unstructured data and enabling us to deliver the most important and financially material insights to our customers.
EF: Which tasks can be automated?
HB: There is a long list of very granular tasks that can be automated but, to summarise, FactSet is building and delivering the most efficient, scalable and highly integrated front-to-back ESG data and solutions suite. AI and machine learning technologies allow us to be more efficient and scalable when it comes to extracting data from unstructured sources. It also allows us to find connections between different data points, and even in datasets where we wouldn’t have found connections before. This gives us the ability to make disparate data more valuable and uncover the insights hidden within it.
EF: How are the parameters of such programs set and how can bias or issues be both identified and avoided?
HB: This is something we have thought about for a long time. I would be more inclined to call a manual collection and analysis process more biased and subjective, as this has been established in numerous academic papers when comparing the assessments of multiple traditional ESG providers with each other.
Using modern computing technology allows us to build an objective pipeline from data collection to data analysis, with the ability for humans to step into the process at any point in time for random sampling. These random samples and a careful topic curation process allow us to iteratively improve on data quality and on the breadth of topics we cover.
EF: What are sentiment analysis, natural language processing (NLP) and deep learning and how can they help in this context?
HB: Sentiment analysis allows the machine to infer the overall sentiment or mood of an entire document, phrase, paragraph, and so forth. This is done by a complex calculation of words being used in a sentence or document, distance between words, and more. This technology is helpful for downstream processing and general understanding of a document. It is also a branch of NLP.
NLP is a field of computational linguistics – it is technology that programs computers to understand large amounts of text similar to how a human would understand it.
Deep learning is a branch of machine learning, which in itself is a branch of AI. Deep learning uses multiple layers of neural networks, whereas machine learning uses a single layer neural network. This technique is used to understand the context and connectivity between different datasets.