Harvard ALI Social Impact Review

View Original

A New Frontier: Generative AI, Business Risks, Opportunities, and Investments in Climate Change

Q&A with Harvard Business School Professor George Serafeim

George Serafeim is the Charles M. Williams Professor of Business Administration at Harvard Business School, where he co-leads the Climate and Sustainability Impact Lab in the Digital, Data, and Design Institute. He teaches the course "Risks, Opportunities and Investments in an Era of Climate Change" (ROICC) that he designed for the MBA program. It explores growing opportunities in energy transition, materials and product utilization and enabling solutions. Professor Serafeim also co-chairs the executive education program “Unlocking Growth with Climate Change Innovation” and regularly teaches in the HBS executive education programs. Previously, he taught “Reimagining Capitalism: Business and Big Problems,” which received the Ideas Worth Teaching Award from the Aspen Institute and the Grand Page Prize, and he created the Impact-Weighted Accounts Project that designed methodologies, produced large datasets, and incubated dozens of pilot projects globally on impact accounting and valuation. His book "Purpose and Profit: How Business Can Lift Up the World" explores the challenges and opportunities in building and sustaining profitable purpose-driven organizations that have a measurable positive impact on society. Professor Serafeim’s research currently focuses on the intersection between the sustainability transformation and the digital and AI transformation.

 

Emily Chien: Thank you very much, Professor Serafeim, for joining me to share the innovative work you are leading at Harvard Business School (HBS) at the intersection of climate solutions and Generative AI. To start, could you provide us some business context and the framework that you've defined on “risks, opportunities and investing in climate change”?

George Serafeim: First of all, thank you Emily for having me join you. If you actually take a step back and see what is going on in the world of climate change, my first thesis is that we are in a very dynamic world where policies and regulations are changing, technologies are changing, customer preferences are changing, business models are changing, and business strategies are changing.

Because of these changes, on the one side, you have the emergence of climate risks. Risk that your assets may become obsolete because now there are new technologies for a better product. Or risk that your assets might decline in value because of regulatory shifts. Or risk that you might be taxed because of carbon emissions and so forth.

Legal, reputation and brand risks are typically cast within the context of carbon exposure, meaning that the higher your carbon emissions, all else equal, you will be exposed to more risk. But this is not the only factor. Carbon sensitivity is another element which means that even if you have high carbon emissions, your customers might not have an alternative. You're not going to be very sensitive to that.

Chien: So, given all this attention on climate risk, how and why did you become interested in the climate solutions side? Why does it matter?

Serafeim: To continue this thread, since customers have the option to buy an electric vehicle instead of an internal combustion vehicle, now you're in trouble, right? So now, that risk is materializing. Of course, we have been doing a lot of work to try and understand, measure, and manage those risks. But I think that is only one aspect of the dynamic environment because the other side of the opportunity is that somebody has to provide the climate solutions.

Organizations are not going to be able to lower our carbon emissions if businesses don't create solutions that allow us to access quality and affordable products to decarbonize. So, we have been concentrating on the idea of climate solutions from an opportunity perspective. Everything that you see from battery, storage, and electric vehicles, all the way to heat pumps, energy efficiency solutions to digital solutions for reuse of materials and packages and products, and all the way to recycling solutions.

And when you think about that spectrum, then you actually start linking the two views of the market. You say, on the one side, this dynamic environment is creating a set of risks for organizations. But on the other side, this dynamic environment is creating a set of new opportunities for businesses. And what is going to happen is some businesses will provide the solutions to the organizations that face the risks, in exchange for a financial return. And that is how you connect the two sides of the market.

To summarize, we have done a lot of very important work on the risk side. But, what’s very, very exciting is when we're thinking about this whole climate challenge from an opportunity perspective for innovation and for bettering the lives of people through better products and services.

Chien: That's incredibly inspiring, George. Now, let’s turn our attention to the opportunity side of the climate equation alongside another very exciting innovation that is taking the world by storm – generative AI, and large language models (LLMs) in particular. Today we’re exploring opportunities at the apex of both sides, and how we are bringing them together. AI is not a new tool in terms of climate change, we know many of the use cases that are being developed and deployed such as better disaster forecasting, smart grids, optimizing transportation and supply chains, methane leak detection and many other use cases.

LLMs can be general, like ChatGPT, or can be domain-specific models, such as BloombergGPT, that focus on natural language processing applications in finance. What drew you to LLMs, since they are relatively new? How do you view LLMs, and why are they a useful new tool to better understand climate solutions companies and performance?

Serafeim: AI is a fascinating field, and just in the last 12 months, we have experienced a leapfrog change in our ability to use AI tools for many, many use cases because of advancements in generative AI and LLMs.

So let me take a step back. When I started my research on the opportunity side of climate change, we wrote a paper on Climate Solutions Investment. I said, okay, we have written several papers around risk, but what if we could write something about the emerging set of climate solutions opportunities? And the first problem that we found was that there was no measurement and identification of which companies are climate solutions companies.

So, a few years ago we did something that didn't use any AI at all. We developed a systematic literature review of climate opportunities and solutions, what constitutes a climate solution, relying on lists such as those coming from the World Bank, Project Drawdown, and several others.

From that review, we developed a set of keywords for different things, such as transportation-related solutions, or energy-related solutions, or alternative fuel-related solutions, and so forth.

We then looked at the firm’s business descriptions, and we systematically went through business descriptions that data providers provided for different firms. We imposed a very strict threshold where we said, for example, the first few lines of your business description need to include these specific keywords, and also, they cannot include certain other keywords. This is because we're trying to identify pure climate solutions firms, not transition firms where the core business is still not about climate solutions, but firms like Tesla and Beyond Meat, who are trying to disrupt an industry.

As a result, we identified about 900 firms around the world, that we had a lot of confidence are pure climate solutions firms. And we wrote the paper about those firms trying to understand their financial performance over time, how much they're investing in R&D and capex, etc. This was our first iteration.

Chien: So, using this foundational taxonomy, you then augmented this work to train a new model, using LLMs such as GPT4 but also other models such as BERT (Bi-directional Encoder Representations from Transformers) which is a Google deep neural net model.

Serafeim: Yes, these are powerful models because it attaches context into the words. And we can systematically classify firms using the LLMs to try to understand not only if a company is a pure climate solutions firms, but also any company trying to transition their products and services.

We trained the model, and then we deployed it to the Item One “Business” description of the company’s Form 10-K filings with the U.S. Securities and Exchange Commission (SEC).

Let’s take the example of General Motors. If you go back to the Item One “Business” description of their Form 10-K, 10 to 12 years ago there was very little mention of any climate solutions. But now, it is all about electric vehicles, batteries, and energy storage, and many of those things are very important elements of the climate transition in the transportation sector.

Chien: In this longitudinal analysis, you studied Form 10-K filings from the 2005-2021 period. Curious, why did you focus on the Form 10-K Item One “Business” description section, when you could look at the firm’s entire annual Sustainability Report or other documents?

Serafeim: Well, the reasons are twofold: 1) the text is standardized and generally, AI works very well when you have relatively standard training datasets. So, the regulation enables us to leverage training datasets that are comparable over time and useful. And 2) the Form 10-K is a regulated filing for the U.S. SEC for publicly listed companies, with supervision of not only regulatory authorities but also scrutiny of lawyers, the legal system, analyst, and investor communities. As a result, it's a much more credible set of information for the firm.

For example, in a firm’s annual Sustainability Report, they might be more likely to exaggerate what they're doing, versus in the Form 10-K. And the Item One “Business” description is more likely to address exactly what the firm’s products and services are so we can understand what the firm’s business is.

From that perspective, we have a very good setting to train the model and then classify these different types of firms – the pure climate solutions firms, and the firms that we observe year-over-year to transition their products and services.

Chien: So, as a result of this work, we now have a new “Serafeim Climate Solutions Firm Taxonomy” for climate solutions companies! Thank you for explaining your thought process. I’m curious, what are some of the most interesting “aha's” that you have discovered?

Serafeim: Emily, I hope you have patented the taxonomy! Let me highlight first that this is important work done with my colleague Shirley Lu, an Assistant Professor at HBS, and involves a truly multidisciplinary collaborative effort at our Data Digital Design Institute (D^3) lab on Climate and Sustainability Impact, with Anna Bialas for example on data science, and several more. On “aha’s”, the first thing to note is that we are right now in the process of analyzing the final results, so this is “hot out of the oven.” But can I share a couple of things that I think are very, very interesting?

We have classified firms into three groups: 1) firms that are not transitioning; 2) firms that were not pure climate solutions but are transitioning over time; and 3) firms that are pure climate solutions firms offering products and services in the market to speed our decarbonization.

This is important as it allows us then to do several things. The first is to understand the geography of those climate solutions. We have found many climate solutions firms that are actually headquartered in Republican states, with an enormous amount of innovation, jobs, and wealth creation. From the policy perspective, whether those climate solutions have been created because of people’s preferences, or because of business opportunity, we now find that the climate solutions are driven by business opportunity.

The second very interesting “aha” is that industry boundaries are blurring over time. The reason is because in the transition of some firms which were previously unrelated are now becoming more similar to each other. However, there are other firms which are becoming more distant from each other than they have been traditionally.

Great examples are General Motors and Panasonic. Traditionally they had very little to do with each other. However, now, they are both actually in the energy storage and in the battery transition business. So, over time, their supply chains, operations and products and services have converged and are overlapping in very substantive ways. And we find this shift happening systematically. Thus, our traditional ways of thinking about industries might not apply anymore because of these industry lines blurring.

And then what we're finding is that those climate solutions exist in many sectors across the economy, from real estate to consumer goods, to transportation, to energy, to hospitality, to materials. We are observing that those companies who are transitioning or trying to disrupt industries are elevating their innovation, raising R&D investments, and are accelerating revenue growth.

But at the same time, because of the higher investments, their profit margins are suppressed. And over time, we're finding that the firms that have successful business model transitions are actually accelerating revenue growth. They are also closing the margin discount that is being generated because of that.

And that leads us to now, and we are in the process of working on several very interesting questions.

One of the questions is what is the relationship between transitioning firms and their human capital strategies? For example, do we observe those firms who are trying to transition their products and services starting to hire more human capital externally to bring new skills and capabilities into their organizations? And if that is happening, then how does this affect their organizational culture and incentives?

Chien: This work gives us many new insights we haven’t seen before. What other questions arise, and which stakeholders could benefit from these learnings?

Serafeim: A second question we are studying is venture capital funding of new entrant/start-up innovation, which can create industry threats and disruption. Is this actually mobilizing the incumbents’ development of climate solutions? Yes, we're finding that basically when you have specific solutions from venture-backed start-ups, for example, in heat pumps, they’re actually generating a response from traditional large-scale manufacturers of heating and cooling equipment.

This prompts us to ask: Do Wall Street analysts understand the future profitability of these businesses? Is it harder for analysts to understand and forecast earnings of companies they are investing in as these companies are transitioning and innovating their climate solutions? But then, these companies’ trajectories become more uncertain, right? That's the nature of innovation. So, we're looking at whether earnings forecast errors become larger for such companies, and whether there is more disagreement about the future profitability of these companies across sell side analysts.

We have several papers underway that seek to understand the measurement, management, governance, and financing of the business transition for the climate solutions opportunity.

Chien: Let’s peak into the future. As LLMs become more powerful, will we be able to expand this type of analysis beyond the Form 10-K Item One “Business” descriptions? And I'm thinking, in particular, about the European Sustainability Reporting Standards (ESRS) and Corporate Sustainability Reporting Directive (CSRD) standards, and the recent State of California climate and emissions disclosures requirements (SB 253 and SB 261) for public and private organizations doing business in California. If we fast forward, could this approach become a much broader tool you are thinking about using in new and different ways?

Serafeim: For sure. Actually, we are thinking about in the Climate and Sustainability Impact Lab of the Harvard Data Digital Design Institute how we might apply the same LLM algorithm to different forms of disclosures. So, imagine how we could apply it to corporate Sustainability Reports, earnings call transcripts, or even to Wikipedia pages to understand what companies are doing and how they are changing over time. And for many other forms of disclosure, we could actually try to understand the level of consistency across the different information mechanisms that are becoming available. The thing to be mindful of is as you increase the amount of information, you trade-off benefiting from more useful signals/information but also including a lot more noise which can degrade model performance. As we move forward, we plan to deploy and expand to many different settings and see how the model might perform and be enhanced.

Chien: Final question. Given the speed at which LLMs are improving and how quickly your work is progressing, have you encountered any challenges? At a minimum, there's probably a lot of data sourcing and cleansing at the early stages, the unglamorous part of AI.

Serafeim: Yes, there are huge challenges. Where do I start? One of the biggest challenges we have faced is that context is very, very important. Here’s a very simple example: if the organization is a real estate development company, a REIT, and you lease out a building, obviously, your product is the building, right? So, you’re making the building very energy efficient and with lower carbon emissions. Because you have superior buildings and insulation, you have energy efficient equipment, you might have low carbon energy for the building. So, is this a climate solution, because the building is your product. But what if the exact same language is being used to describe the headquarters of Proctor and Gamble? That is not a climate solution, right? Instead, it is part of Proctor & Gamble’s company operations, so it is not the firm’s product. But in today’s LLM model, that is quite challenging to separate.

So, one of the big challenges that we face is introducing nuance into the model to be able to separate when somebody is discussing products and services that lead to climate solutions, versus from their own operations, or from a market development, or from a regulatory perspective. These models can be tricked to think that this business is a climate solutions company, because of all the mentions of all these things while the company is not actually providing them as climate solutions.

The work ahead is to be able to further refine and develop the contextual and the nuanced understanding so the model can accurately identify the types of companies that are providing those products and services that constitute true climate solutions, and not misclassify companies based on other types of language that is being used.

Chien: Thank you for sharing your “hot off the press” work. It’s so encouraging and exciting to consider climate change from the lens of innovation and new solutions so we can better understand businesses that are capitalizing on opportunities while also addressing their climate risks. We can’t wait to learn more as your research continues to advance!

Serafeim: Thank you so much, Emily. It’s my pleasure.


About the Author:

Emily A. Chien is a Senior Fellow at Harvard University, specializing in Climate Finance, the Net Zero transition and commercialization of new Climate Solutions. She led IBM’s Global Climate Offerings and Partnerships and is a founding co-chair of the 100 Women in Finance ESG C-Suite Peer Advisory Group. As a long time champion of and innovator in data, digital and AI, Emily led AI transformations for IBM clients in banking, investments and insurance and was a 2021 AI/ML Fellow with the World Economic Forum.

In her prior leadership positions with JP Morgan, Fidelity Investments, Prudential Financial and American Express, Emily built, commercialized, and ran new solutions, platforms and markets working at the intersection of business, technology, and partnerships. Emily holds a patent for digital capabilities now used by Amazon.com, is a former CPA and member of the Economic Club of New York, 100 Women in Finance (Global Angel), and CHIEF women’s leadership network. She was a Chair of the 100 Women in Finance 2023 Impact Investing Symposium.

This Q&A has been edited for length and clarity.

Cover Photo Credit: Image by Freepik