How AI is Transforming ESG and Data Governance

Authors: Claire Walls, ESG Product Manager

Diana Rose, Head of ESG Solutions

With the recent developments in Generative AI (Gen AI) and Large Language Models (LLMs), there has been a flurry of discussions around the possible business use cases for these tools, as well as the risks. A promising area is in the ESG space, around data collection, analysis, and reporting. From our experience, while people are generally optimistic around AI's potential benefits, there is also a fundamental concern around what good data governance looks like.

There are various aspects of ESG management and reporting that AI will become widely used for, and each has its own considerations for data governance. Here’s what we’re seeing.

1. Use AI to do the boring bits of gathering and categorising ESG data

By far the biggest difficulty of ESG reporting is collecting and organising data, which has obvious potential for Gen AI and LLMs. AI models can be tasked to search databases at speed, extract relevant ESG metrics, and classify them into specific categories. It can also help provide estimations, when given data and instructed as to what calculations are to be made. For example, results from supplier and subsidiary surveys can be compiled faster through AI.

These tasks in ESG reporting create huge workloads for sustainability teams, leading to burnt out analysts, prone to mistakes and uneven results depending on the team member in charge. AI capabilities are already proving their worth to help with these most tedious aspects of ESG data collection and validation, alleviating the load for analysts.

But analysts can’t put their feet up just yet!

Good data governance here involves ensuring the integrity and provenance of any data being fed to AI models, expertly designing instructions and prompts for the AI to follow, and overseeing the outputs to ensure validation is done to the right standard. Picking the right technology solution should be done while bearing this in mind so you don’t risk rubbish in AND rubbish out.

2. Can’t AI just generate my report for me?

Although there are various solutions that would have you delegate your report creation to AI, we believe using LLMs to generate entire corporate reports is a bad idea. Having said that, we see there are genuine, targeted uses for AI in the creation process for ESG reports. Gen AI can help prepare data and generate analysis, even creating drafts and images, which team members can review and validate. AI can also be useful in proofreading and quickly identifying small mistakes and typos, cross-checking for consistency on disclosures, and against the house style. 

Beware the AI ghost-writer

When it comes to retaining the trust of your reader, and using words to paint a truthful picture of a business’ activities, it’s important to remember that LLMs don’t understand what they are writing about and therefore can’t be reliably delegated to replace a professional in certain creative or analytical tasks, such as interpreting a set of ESG information to convey a strategic message on behalf of a company’s leadership.  While AI used well may support accuracy, AI used badly may compromise it, and protecting truthfulness as a pillar of corporate reporting is the domain of human intelligence and judgement.

For newly released research on the use of Gen AI in corporate reporting and guidance on how best to utilise its capabilities, see our joint research with Falcon Windsor.

3. Making sense of ESG reports

While various ESG ratings providers have been using Machine Learning and Natural Language Processing to analyse company reports for years, the potential for generative AI is meaningful to help the consumers of ESG reports make sense of them.

AI is due to make report analysis faster and smoother, and this is already being foreseen by legislation, which increasingly requires companies to publish in AI-friendly formats. The European Union, as always a frontrunner in sustainability regulation, has included a requirement for reports to be published in machine readable format such as XBRL in the Corporate Sustainability Reporting Directive, the most comprehensive and wide-ranging ESG reporting regulation to date. 

Get ready for increased scrutiny, accountability, transparency

This type of regulation is ultimately intended to level the playing field on sustainability reporting and help investors to scrutinise, compare and essentially make sense of reported metrics so they can make clearer investment decisions based on that data.  Greenwashing should in theory become harder and harder with AI tools available to regulators.

With that in mind, published corporate documents are starting to be evaluated for their AI-friendliness, and data governance sits at the heart of that for auditability and accountability. This is becoming essential as AI will increasingly be used to parse through documents, extract the relevant data and provide overall interpretations, including gap analyses, benchmarking, finding contradictions and providing insightful summaries. 

Data Governance: Boring but more important than ever

The various applications of AI for ESG data processes are due to relieve some of the burden around ESG reporting, and will reduce the time taken both to create reports and gain insights from them.

However, to mitigate the risks around the Gen AI models of today and tomorrow, we need water-tight data governance. Data has become the most valuable commodity globally, and accuracy, integrity and security are of paramount importance. The output of AI models is only as good as the input data, and to retain their stakeholders’ trust, companies will have to ensure that they not only have full confidence on the source of the information but also on the systems there to ensure transparency – both in terms of traceability of information and what mechanisms were used to process data and make decisions.