Automated answering of subjective questions on environmental and social governance


Challenge identifier: DPC2-2018


Proposed by

Bloomberg, a global financial services provider which provides business and financial information, news, and insights to decision makers in business, finance, and government sectors.



Environmental, Social and Governance (ESG) measures are a set of criteria via which analysts and investment managers can assess how a company is being run. These criteria typically include information about a company’s impact on climate change and the pollution of the environment, anti-corruption policies and performance, policies and metrics concerning the diverse and inclusive nature of its workforce and management team, and contribution to human rights advocacy and protection.

There are several reasons this information is valuable to financial services organisations. Investors may believe that managers that take these measures into account are more likely to build stable money making companies and will use ESG data in their stock selection strategies. Funds may build investment products that target socially conscientious investors – individuals and companies who wish to see their money invested in companies who behave in a way they approve of.

We have a number of data products that specifically target the ESG investment market. A key issue working in this market is that definitions of ‘good’ ESG practices are often subjective. This means that good technology solutions in this area need to be flexible to changing preferences about what constitutes good governance.


We have collected the majority of documents that companies publish about themselves. In most markets these include mandatory annual reports on company performance, and may include specific ESG reports. We are looking for ways to reduce the time it takes for analysts and investors to determine ESG ratings for companies. Typically this requires answering a set of questions and so this challenge could be cast as a question answering task for a natural language processing (NLP) system.

We are interested in NLP technologies that can be trained to answer the sort of subjective questions an ESG analyst may have. For example, an analyst that chooses to rate a company highly for pursuing strong diversity and inclusion policies would need to be able to quickly ascertain which companies in a set rank highest on these criteria.


Company filings and ESG reports with relevant metadata

More detailed information about the data can be found in our data catalogue.


Expected outcomes

  • Prototype question answering system
  • New approaches to assessing compliance to a set of subjective criteria

Expected impacts

  • Ability to answer a set of ESG questions as well as a human being could
  • Development of new sets of financial products
  • Greater incentives for socially positive investment behaviours