AI and ESG: Better Together

Synergies from foundations to outcomes

Last week, I was on a panel for practicalESG.com and was challenged to explain how AI can be applied to ESG problems. I wrote up a twelve-minute talk track, which I edited and decided to share here!
As a reminder, I work at Microsoft, so some of this pivots that way.

In my new book, ESG Mindset, I discuss technology as another pillar of ESG with similar risks and opportunities. I also discuss how technology can be used to solve ESG challenges, and today, we’ll focus on the latter, more specifically on AI opportunities.

Before we jump in, a little context on ESG and the broader technology implications.

As you well know, I like to define ESG around its original meaning:

The material Environmental, Social, and Governance risks that affect a company and opportunities that drive toward long-term value and sustainable growth.

ESG Mindset argues that while companies can help save the world and its people, they must approach these goals from within their core business. A non-material approach will only lead to transient efforts.

Both digital technology and other types of technology can move a company forward, whether related to ESG, sustainability, social justice, or impact. Still, every technology, digital or otherwise, will likely have its risks and opportunities, similar to the other pillars of ESG.

Lastly, AI can mean many things. We’ll mostly cover three types here: Machine Learning, Large Language Models (LLMs), and Generative AI. For the purposes of this article, Machine Learning is best focused on specific tasks, often quantitative and analytical in nature. LLMs are better at understanding qualitative content and human language while the related Generative AI is good at text summarization or generating content.

With that framing, let's talk specifically about the opportunities that AI offers for ESG-related issues.

Applying AI to ESG starts with data

Today, many companies chase disclosures due to rising ESG regulations globally. This means that they gather data from purpose-built business systems, like ERP, HR, facilities, and more, and calculate those activities into carbon, water, waste, diversity, and others. In other words, new quantitative data sets are created from existing purpose-built systems for ESG reporting.

One of AI’s first applications is helping to collect this data. For example, through optical character recognition (OCR), AI can transpose handwritten invoices into digital data to be calculated. AI can be further applied to automate the task of uncovering variances in the data collected. This can be a simple automation script, as in ‘tell me when the numbers vary by xx% month over month,’ and then be supplemented with automation to execute specific tasks that follow, such as alerting a sustainability team to investigate.

All of this collected data sits in new purpose-built systems for recording and reporting ESG metrics to stakeholders, investors, and regulators. Due to its quantitative nature, AI tools like Machine Learning can be applied these metrics to uncover trends, correlations, and forecasting. In fact, Microsoft has a tool called Sustainability Manager that does just that. For the Sustainability Office, this can be invaluable for uncovering areas of improvement to pursue.

However…it is critical to note that if we only follow the sustainability data collecting workstream, the value of non-financial ESG data is locked out from the company and is only of use to external parties, as it is reported.

The business value of ESG data comes from connecting it with other business systems and then applying AI and other digital tools to conduct advanced analysis. The challenge is that many companies are built on organically grown data siloes. I spent about 15 years in IT and I've seen just how these siloes can form as a company grows. Breaking down these data siloes securely and democratizing access with quality governance is the first step in valuable ESG analysis, but also has applications across the adoption of AI and other new technologies more broadly. It also can have additional value to other business functions, like Compliance and Risk up through to the management team and board.

In other words, the sustainability office at your company may be a force for Digital Transformation.

If a company can break down these siloes to pull the context of calculated or extracted ESG data alongside those original purpose-built business systems, it can gain additional insights. Finding patterns across these complex datasets is, again, where tools like Machine Learning alongside data science tooling and visualizations can inform higher-quality decision-making.

AI plus Quantum Computing can deliver sustainable R&D

The nature of business is shifting around ESG, and focusing on collected metrics for regulatory disclosures, even when combined with business data, will only take you so far. In breaking down data siloes and conducting analysis with AI, the company may uncover opportunities to address material ESG challenges. These opportunities may further use AI to support a transition.

Each company has to work to find its unique intersection with these topics. Doing that means looking at a lot of internal and external information in new ways.

For example, extracting lithium to support the renewable energy transition through solid-state batteries is often debated as a serious environmental trade-off. In the past year, Microsoft Research and the Pacific Northwest National Laboratory (PNNL) announced that we had worked with Quantum Computing and a new AI model on this challenge.

We used advanced AI to screen over 32 million candidates to discover and synthesize a new material to create better batteries. This is one of the first real-life ESG examples in this new era of R&D driven by AI. Those 32 million were quickly whittled down to 500,000, and through filters, 150 were selected. From here, 18 were looked at with humans in the mix, and 1 moved into real-world testing. Scientists are hopeful that a new sodium-lithium electrolyte may reduce lithium in the batteries by 70%.

With AI, this type of complex problem-solving no longer needs to take years but can take days, accelerating a sustainable transition.

Not all data is metrics-based

The previous examples mostly sit with Machine Learning, but with LLMs, we can also analyze qualitative information and context. The possibilities are massive as much ESG information exists outside of metrics with the stories that companies and their suppliers tell.

However, if a company doesn’t back up its stories with quality data and actions, no amount of AI can save it.

Generative AI is really good at summarizing large amounts of text, so building on the solid-state battery research use case, a company could turn an LLM over to investigate industry research at the material intersection with the company. For example, if I'm an EV or solar company, I could create a crawler to look for news articles about solid-state batteries and pull that information, summarizing the latest innovations for the R&D team.

But to go back to data collection, the ESG regulations out there are meant to streamline and standardize ESG data for investment decisions, but inevitably, companies publish those metrics and content in their CSR reports. Procurement teams and other stakeholders use the metrics to compare players in their value chain. There's just one issue. Since ESG information exists in qualitative storytelling, analysis at scale has been difficult since Machine Learning is better at quantitative analysis.

With LLMs, we now have tools to understand content. In addition to scraping CSR reports for tables of quantitative information or purchasing it, now Procurement teams, investors, and other stakeholders can understand the context behind the numbers as the company explains them. A Generative AI can output the results for review.

Think about the disruptive power of Generative AI on something like ESG scores. If I can ask a natural language query on a company's ESG position, specifically focusing on either what I think is material or perhaps focused on thematic impact, do we need ESG scores anymore? Generative AI could surface relevant information to the person asking the question, ultimately leading to the specific data that stakeholders and investors need to make a more informed decision.

LLMs can also be used to understand stakeholders better through marketing campaign response, call center logs, and social media monitoring through sentiment analysis. Previously, sentiment analysis existed only in the realm of Machine Learning, but LLMs can uncover and summarize this information more effectively because they understand the content. The difference here is between understanding that 10% of callers were having issues with a particular function to surfacing what the customers were doing at the time that caused the issue or how this impacted them.

Managing ESG risks with AI

Besides uncovering new opportunities and information, ESG risks that can be uncovered with AI. For example, there are AI models in use today to understand different global warming potentials and the related localized climate risk. This information is invaluable for property protection. Here in the US, we've seen homeowner’s insurers pull out of states entirely for their climate risk. Germany's reinsurance industry almost fell over a few years ago due to severe flooding. The CEO of Aon last week stated that insurance needs new models to deal with climate change risk.

Again, the nature of business has changed.

There are many stakeholders, companies, investors, and governments who need to plan around localized climate risk for their physical assets and along their logistical routes. It also isn't only climate risk either. Two of the world's biggest canals are suffering right now. While the Panama Canal is experiencing a drought, the Suez Canal is under siege from pirates. Over the past few years, we’ve seen escalation of global conflict disrupting supply chains and people. With the right mix of supplier and logistical information, companies can find new ways to move inventory from place to place and keep their employees safe.

We're also already seeing AI paired with satellite imagery to examine systemic biodiversity collapse, although I haven't observed this yet at scale. For an agricultural, processing, or even retail company, this type of work is critical for ongoing resilience and sustainability.

Summary

Let me close by saying that AI can do many things, but there are still two things it needs.

First, you have to break down data siloes securely in order to gain the biggest benefit from its use. This is a big challenge, especially in highly regulated industries, but it can be done (with surprise!) data governance.

Second, you still need people in the mix to help the models and deliver the value that the company needs, whether tackling a material issue or trying to save the world.

Don't underestimate the value of AI in helping to understand your ESG issues more clearly. While the examples here are practical ones that can be implemented today, AI has the potential to help connect the dots across interconnected risks in new ways. Today’s business issues flow in and out of the E, S, and G and when you break down data siloes and layer on a responsible AI model, you may uncover the information and insights needed to build a resilient company.