About this Episode
Patrick Wood Uribe, CEO of the fintech company Util, joins the Essential Podcast to talk about how his team is using AI technology to evaluate the ESG performance of companies against the U.N.'s Sustainable Development Goals.
The Essential Podcast from S&P Global is dedicated to sharing essential intelligence with those working in and affected by financial markets. Host Nathan Hunt focuses on those issues of immediate importance to global financial markets—macroeconomic trends, the credit cycle, climate risk, ESG, global trade, and more—in interviews with subject matter experts from around the world.
Listen and subscribe to this podcast on Apple Podcasts, Spotify, Google Podcasts, and Deezer.
- Util uses machine learning to measure the real-world impact of how 45,000 global listed companies affect the 17 United Nations Sustainable Development Goals (SDGs) and thousands of other sustainability concepts, empowering investors to make more informed investment decisions.
The Essential Podcast is edited and produced by Molly Mintz.
Nathan Hunt: This is The Essential Podcast from S&P Global. My name is Nathan Hunt. When the UN first announced its 17 Sustainable Development Goals in 2015, the initial reaction from the financial industry was muted. The goals felt so big, so high-level that it was difficult to see in that moment how they could be used as a tool to evaluate companies and investments. However, in the years since, the SDGs have begun to be used as an evaluative framework within financial markets. One of the companies that is leading in this space is the artificial-intelligence-meets-ESG company Util, which happens to be led by a former colleague and an old friend.
Patrick Wood Uribe: My name is Patrick Wood Uribe and I'm the CEO at Util.
Nathan Hunt: Patrick, to get started, journalistic integrity demands that I acknowledge our past professional relationship. Not so long ago, we worked for the same mothership. You were at Kensho and I was at S&P Global. I'm still at S&P Global. Now, you have left us to serve as the CEO of Util, to which I must ask, how could you do such a thing? What's so great about Util that you would leave us bereft like this?
Patrick Wood Uribe: It is funny and appropriate that you should put it that way. I very much enjoyed my time at Kensho, which was just absolutely wonderful, and I think the relationship between Kensho and S&P was a very healthy one while I was there, and so it was a matter of being presented with an opportunity that was, that was simply too good to pass up. I was just very excited about the work that Util was doing at the time and used to do and it's a real privilege to be able to jump into this particular role. But as you say, the way in which I could do it is to be presented with something even better.
Nathan Hunt: Util is a financial technology company using big data and machine learning to derive the social and environmental impact of every company and portfolio. I'm obviously quoting from your website right now. I'll try to paraphrase that. It's impact investing using unstructured data and machine learning. Is that more or less accurate?
Patrick Wood Uribe: Absolutely. Yeah. That's really, in a nutshell, what we've been trying to do is that traditionally something like impact investing or understanding the impact of companies is something that takes a lot of human analysis and that comes with some downsides. The first is that even with a perfect human analyst, that limits the amount that you can actually cover, so you don't get a huge amount of coverage. Then there's obviously the limits of being human, which means that biases creep in, we're not consistent with one another, so different humans come up with different ideas, and then of course, that means that if you really want to do this at scale, it's an enormous challenge because you don't have incomparable across all of the different possible companies you might invest in. That's where what we're doing using machine learning helps us to bypass some of those issues.
Nathan Hunt: Util evaluates companies against the United Nations' sustainable development goals. Why did you choose the SDGs? They're objectively pretty ambitious, but they aren't really bullion. For example, goal 10: Reduce inequality in and among countries, that's not really a matter of checking a box.
Patrick Wood Uribe: Exactly. I think there are some distinct advantages with looking at the SDGs that come precisely from some of those kind of nuances that exist within them. There are a few things that we especially like about them, one of them is that they are internationally recognized, and so that helps from the standpoint of simply having an external framework that is recognizable, and that is something that has been developed separately and outside the company so that we're not just saying, "You can trust us for offering work." We're actually using something that is holistic and that has been developed elsewhere.
Patrick Wood Uribe: The other aspect is its holistic nature, which is very helpful to us because ultimately what we're doing is trying to measure multiple impacts of companies on the world. The framework of the UN SDGs allows us, for instance, to understand both positive and negative impacts simultaneously and holistically, so there are ways in which things that might have a negative environmental impact might have also a negative consequence on health outcomes via something like water or air pollution. Things like that are captured using the SDG framework that are very difficult to do otherwise. The last thing is precisely that they are goal-oriented, which I think is a really key thing in order to affect change. I think if we just do incrementally better than we did last year, or five years ago, we're not really going to get to the place that we need to in terms of the overall goals, and so the very ambitiousness of the SDGs is something that ultimately helps us.
Nathan Hunt: Who are the clients for this type of analysis? Is it more private investors or large institutions? Who are your customers?
Patrick Wood Uribe: Our customers are primarily asset managers. These are kind of institutions managing money and they use it in a variety of ways. The principle way is to monitor and understand the impact of investments. That happens in a slightly different way from what I would call "traditional ESG" in quite a complimentary way to traditional ESG in that we are looking at precisely what you described so accurately, which is the impact of companies, and that gives us a slightly different set of metrics to look at. We use the UN SDGs as our framework. Ultimately, that's where asset managers are using it is to understand that impact. They may report that to their own clients so they can incorporate it into their client's understanding of the funds that they're invested in and it will also allow them to make slightly different decisions as to which companies they invest in, and in what quantities.
Nathan Hunt: How much can an asset manager customize this analysis you're providing? Let's say I care about SDG goal 14: Conserve and sustainably use the oceans, seas, and marine resources, but I'm less concerned about goal nine: Build resilient infrastructure and promote inclusive and sustainable industrialization. Can I see the data just that way?
Patrick Wood Uribe: You can, and that's exactly part of the design of the dataset. What we do when we look at the UN SDGs, and we have a whole methodology that ties companies through our evidence base of 120 million academic and scientific journal articles, ultimately through to the UN SDGs, but the data output itself is precisely designed to be as granular as possible and to give as much objective information about companies as possible so that our clients can make these types of decisions.
Patrick Wood Uribe: To give you an example exactly along the lines that you mentioned, for every UN sustainable development goal, we have a metric of positive and negative alignment. I say "And negative" because something that may be positive in one way may be negative in some other way. A classic repeatable example is the way in which energy, for instance, increases economic development, but ultimately has negative consequences for the environment. If I am a client of Util's and I get our dataset, I have granular data for every company, both positive and negative on all of the UN SDGs, and then I can weight that myself. I can say, "I'm only interested in companies that have a positive impact on these three SDGs," or this one SDG, "Oh, and at the same time, I also want companies that have no more than X amount of negative impact on these other SDGs," and so we've designed the dataset to be easy to use in that sense, so that you could conduct quite basic filtering on the overall dataset and still get a very new nuanced ultimate output when you do that.
Nathan Hunt: What is the relationship between this type of analysis and the existing ESG scores that are in the market? Would this be used instead of or in addition to an ESG score?
Patrick Wood Uribe: I would say it's complementary, so I would say definitely in addition to, and that's partly for, I would say, two related but slightly different reasons. One of them is that the ESG metrics that are currently used in a very widespread way tend to look at the extent to which companies manage the ESG risks, and so it's very much focused on not only internal choices by companies that may relate to their practices, et cetera, but it's also related to risks to the company's business, and that of gives a very specific picture of those risks and of those environmental, social, and governance factors altogether.
Patrick Wood Uribe: What we are doing, we actually start by looking at the revenues companies generate from what they sell, and so we're actually looking at a different dimension to begin with, and that gives us what I sort of see as the other 180 degrees of a 360-degree view of the company. If you have a traditional dataset that gives you a sense of the external risks from ESG onto the company, then our dataset gives you a sense of the company's outward impact on the world via the products and services that it sells, so it's highly complementary in that sense.
Patrick Wood Uribe: I think the other component that is important to bear in mind is that it's tied to a separate starting metric, so if we think about ESG, those tend to be risk-based. They tend to be incorporating more risk factors, whereas we're looking at revenues and then impact. That is, as I mentioned earlier, it's related, but it's slight different in that it provides, I call it "a different focal point" for the analysis as well, so ultimately, I think it sits quite well alongside traditional ESG datasets.
Nathan Hunt: The big concern right now in ESG circles is greenwashing. You have Tariq Fancy, ex-BlackRock chief investment officer for sustainable investing basically accusing the entire industry and his former employer of faking the funk. Can your approach help prevent that kind of greenwashing that Fancy is referring to?
Patrick Wood Uribe: Absolutely. I say that with a good deal of confidence, but at the same time, I do think that greenwashing is very, very hard to combat, so I don't want that to sound flippant. But I do think that it's important in thinking about greenwashing to think about the spectrum of how it could possibly happen. At one end of the spectrum, there's obviously the cynical view, which is that essentially, there's a bad-faith effort to sell basically products identical to what they were before and rebrand them and claim that they're somehow more ESG-friendly than they were before. At the other end of the spectrum is the non-cynical version, which is that it is very, very difficult to capture all of the necessary information to understand whether a company or a fund is greenwashing or not.
Patrick Wood Uribe: That's where I think our data and where I'm very confident our data is really helpful is because what we are doing is we are introducing brand new information about impact of companies that helps separate companies from one another that might otherwise look quite similar. If you take, for instance, two energy companies, one energy company promises to invest in renewable energy sources. A second energy company makes exactly the same promise. Five years later, one has invested money and actually now generates revenues from wind power and the other one has not really made any investments and they just keep claiming every year that they want to do more renewable stuff and nothing has changed and their revenue sources remain identical.
Patrick Wood Uribe: Our data shows the change over time between those companies. Their ratings may in other circumstances have stayed the same. What we're showing is that the fundamental business of the company now generating more of its revenues from wind power. The fundamental business has changed because it's generating that revenue from a different source, and so if you are looking at companies and trying to separate them from one another, being able to do it with this kind of product focus or this kind of outward-looking focus from the company is really, really helpful, so I think that's where data like ours is really helpful for, I think, just picking through the issues when it comes to greenwashing and can certainly provide a clarifying new piece of information for companies.
Nathan Hunt: Patrick, as you had mentioned earlier, the challenge of human analysis of sustainable factors is that it is very easy for bias to be introduced, but the challenge of machine learning is always that as good as it is, it's only as good as the data on which it is trained and to which it is applied. Given that companies have gone to great lengths to obfuscate their environmental and social impact, are you concerned that your model may still be missing key details?
Patrick Wood Uribe: I'm so glad you asked. That's a really wonderful question, sort of fascinating topic in terms of how we try to... This is going to sound very philosophical, but it's actually how we try to improve on ourselves as humans using machines. It may sound slightly odd, but I do think that there's an opportunity to use machines in a way that can help us with this.
Patrick Wood Uribe: I can give you a couple of examples. The first example is actually a choice that we Util in terms of how we process the data and what we're basing it on. If you were, for instance, highly cynical and assumed that every company is obfuscating its activities in some way, the only piece of information that we get that is tied to a company is the revenue data. Claims about its products and what they do are actually not something that we get from the company at all. They may promise to make something in a very specific way, but we don't track that as a claim, so that's not the type of analysis that we do.
Patrick Wood Uribe: What we're doing instead is we are looking at academic and scientific journal articles across every discipline published over decades, and that's what gives us our volume of 120 million academic and scientific articles. What we're doing is we're decoupling products from the companies, as it were, and looking for the impact of those products in that impartial academic database. As a data source, it's incredibly high quality because of the quality of the work that goes into the publications in the first place, they are all peer-reviewed, so there is a layer of quality control before we even see it, and then, of course, there's the modeling that goes into pulling out that accurate information.
Patrick Wood Uribe: Now, that takes time to build as well. I think one of the things that is striking about using machine learning in the way that we do is that it always seems like because our machines read 120 million articles every night, it sounds like that's something we could just build in a day, but actually, it takes a very, very long time to train those models, to do it accurately and to do it in a way that we can trust, so it's sort of a combination of those elements, but ultimately, I think it is really, really important to consider the source data really carefully and to make sure that it's something that either is already quite high quality and that you can then obviously work with or that you know how to turn it into a high-quality data source by cleaning, structuring all of the other things that can go into that process, and then ultimately, obviously, you have to make sure that you are monitoring and refining your model so that they ultimately do the work they're supposed to do.
Nathan Hunt: This sounds quite laborious. How many companies have you done this analysis on?
Patrick Wood Uribe: We've done the analysis on 45,000 companies, which is every listed equity in the world. The laborious component is sort of interesting because it is laborious, but it does happen at scale, so we've iterated a great deal and we've built from a small number of companies to a large number of companies and we've been making sure that the models get more and more accurate and all of those things, and so there is a lot of labor involved in getting it, I would say, improving the quality. Ironically, the scale is not as much of an issue, so that's how once we have the quality of output that we want, it's a matter of scaling then to the size of coverage that we have. We cover 45,000 companies. For each company, we have several hundred data points, if you can imagine. We cover the UN SDGs positively, negatively. We have some additional metrics on top of that and we have a five-year history as well. There's a huge amount of data that comes out of a relatively small team and has been developed laboriously over time.
Nathan Hunt: Patrick, one final question. You have had a career path that might seem unusual to some people. You have a PhD in music theory, you were a tenure-track academic and a concert violinist. That might seem like an unusual career path for a fintech CEO, but I suspect that given the complexities of machine learning and ESG, your past may be more applicable than it might first appear. To illustrate the point, I intend to ambush you with an unfair and needlessly theoretical question to end this podcast, so here goes: How is ESG measurement like the concepts of consonance and dissonance in music theory?
Patrick Wood Uribe: Oh, goodness. Well, first of all, what a delightful question, and second of all, it's very funny you should bring this up. I will answer your question directly a second, but my career path has come up quite often, and very often the link between music and mathematics comes up as a frequent pairing that happens quite commonly. I do think there are some very interesting parallels between how we understand, for instance, an enormously large body of documents or the patents in financial markets. We understand those using abstract models and that's actually very similar to the way in which we understand pieces of music via music theory. There are some kind of linkages between my history and my presence, so to speak.
Patrick Wood Uribe: In terms of consonance and, and dissonance, I would say the thing that strikes me immediately is that the important features of a company, and by that, I mean both financial metrics of a traditional kind, like their profits, et cetera, and their non-financial metrics, such as partly covered by ESG and partly covered as Util measures, I think broadly speaking, those should be consonant. I think when they do, that's ultimately the type of companies that we want to reward and that we want to anticipate rewarding if we want to make sure that we capture that value for investors.
Patrick Wood Uribe: I think dissonance is, in many ways, the opposite, but we do also find it when we see that there are companies that have a good track record in the environment, but not so much on the social side or whether they have good governance practices, but they actually have negative environmental impacts. We see that quite often as concepts that I would call "dissonant" in a broad way. I would sort of bring that back to music theory in those two ways. Since you asked the question, I'd love to know if you had a theory of how these concepts might link to ESG yourself.
Nathan Hunt: I had a theory. It is not as good as your answer, but I thought that consonance and dissonance can vary a bit over culture and over different forms of music and that they are meaningful concepts that, at the same time, vary over time and need to be responsive to the circumstances of the music itself, and I thought ESG measurement could, in some ways, be similar.
Nathan Hunt: Patrick, as hard as it is for me to say this, I forgive you for leaving S&P Global because it sounds like the work that you are doing at Util is fascinating and useful and potentially quite valuable for the world at large, so thank you so much for joining me today.
Patrick Wood Uribe: Thank you. I'm delighted to hear that and I'm obviously very proud of the work that we're doing at Util with the team. I'm very glad that you think so. Thank you so much for having me. It's been a real pleasure talking to you.
Nathan Hunt: The Essential Podcast is produced by Molly Mintz with assistance from Kurt Burger and Camille McManus. At S&P Global, we accelerate progress in the world by providing intelligence that is essential for companies, governments, and individuals to make decisions with conviction. From the majestic heights of 55 Water Street in Manhattan, I am Nathan Hunt. Thank you for listening.