The Green Finance Podcast Ep. 15: Why climate AI is essential to reach net zero
- Today, we're talking about artificial intelligence – a tool uniquely positioned to help manage the complex issues presented by climate change.
- To help us understand more about how AI can help us solve the climate crisis, I've invited BCG's leading sustainability expert Mike Lyons to expand upon the study findings and explore real-world, practical applications for climate AI.

Today, we're talking about artificial intelligence – a tool uniquely positioned to help manage the complex issues presented by climate change.
Due to its capacity to gather, complete, and interpret large, complex datasets on emissions, climate impact, and more, it can be used to support all stakeholders in taking a more informed and data-driven approach to combating carbon emissions and building a greener society.
Climate AI could really accelerate our path to net zero, but there are still challenges, like insufficient access to expertise, limited availability of AI solutions and even a lack of confidence in AI data and analysis.
Companies want to use AI to help with their own climate efforts, but only around half have a clear vision of how to do this, according to a Boston Consulting Group study.
To help us understand more about how AI can help us solve the climate crisis, I've invited BCG's leading sustainability expert Mike Lyons to expand upon the study findings and explore real-world, practical applications for climate AI.
Tell us a little bit about your background - How did you start working in this area of sustainability intersecting with artificial intelligence?
If you wind the clock back far enough, I'm a fairly classically trained chemical engineer. I grew up in Houston, went to Rice University. I had a career in the chemicals industry prior to joining BCG. That was 17 years ago, when I made the jump. I come from a heavy industries background, I know what it looks like on the front lines. And I've always been in and around, you know, heavy energy demanding companies.
Here at BCG, I've worked on a number of things, but over the last, let's say, handful of years, I've come to lead and develop a lot of our offerings in the climate and sustainability space. Specifically, I lead several of BCGs climate topics specifically around carbon footprint, baselining and mitigation. We also have a pretty advanced analytical suite that we've had to build from the bottom up. And even some CCUS and hydrogen, you know, when it comes to decarbonizing heavy industry, we're also pretty active there.
I basically spend every waking moment of my day, looking at either emissions or climate risk, but with the intent really to serve all of our clients globally, also in the banking sector. So we've had to build our offerings with that scale in mind, to really support all geographies and all companies.
That's awesome, and pleasure to have you here today to help us understand how this all connects. Because climate change is complex and we need proper tools to handle that. AI can interpret complex data sets, so it really sits at the core of the data driven approach that we need to take in order to combat carbon emissions and to build a greener society. So in your view, and to speak in general terms, what's the potential that AI can unlock?
The ability of AI to unlock value, I think, is tremendous. And we've written on this extensively, you know, in our perspectives, and we see it with our clients. I would say the topic areas, at least for me, fall broadly into carbon intelligence and abatement, everything around your footprint. Perhaps the footprint of your banking customers, for example.
The other side is more around climate risk adaptation and resilience. Within both you have the aspect of huge amounts of capital spent, you have the aspect of fairness within your customers and your companies and even within your societies. As you can imagine, you brush up against company data, country level data, socio economic data, jobs data, capital spend data and individual project level data.
If I take a step back, you're trying to go end to end, you want to both measure carbon or measure risk. You want to enable decision making around that, whether it's what actions to take to decarbonize. Eventually, you want to reduce carbon footprint, you want to reduce risk, but you want to do it in a way where you can afford the journey, so to speak, being able to put on top of this, I would say, an optimization layer.
And then also on top of that, expanding the boundaries of your company by collaborating, bringing in new data sets that you don't traditionally have access to, in order to improve your analysis and ultimately your recommendations and actions. For me, it is the end to end workflow that both companies, communities and governments are trying to enable.
Now, as you rightfully pointed out, there are some obstacles, I think data availability, and uniformity creates a huge pressing issue, because you do need the harmonization of many different datasets to even turn loose a machine learning model or to be able to use a more advanced analytical tool. And so without that, you're stuck. So there's a big piece around knowing what data to get, what's the format, what's the structure? How do I bring it together? So we definitely have invested a lot of time to do that. And on each project that we do, we find that that's the unfortunate front end of the work to really then get you to the more interesting stuff where you're actually optimizing and making decisions.
How do you go about designing an AI software with ethical behavior in mind?
I'll go back to the two main pillars, both carbon management and then climate risk management. In the carbon management side of things, where we have used AI a lot is around again, data ingestion and harmonization, but then also the matching of client data to this big database of emissions factors.
Then on top of that, using it to prescriptively offer recommendations to decarbonize or pull from our let's say, our initiative libraries, so all of our work where we've done decarbonisation, but matching an initiative to a baseline element. In all of that, you are relying on natural language processing technologies to both bring together emissions factor databases, which in our CO2 AI tool, we have about 160 databases that we bring together. Massaging that data into a harmonized hierarchy and taxonomy was a huge lift, we probably spent two or three man years to get that organized. And then I would say, luckily, we use that same NLP matching technology to match client data to this emission factor database.
To your point around fairness and problem solving, and how can you be sure that that is working? Well, a lot of the software design to the actual front end that we built, allows humans to quickly let's say, interact with and diagnose the quality of these matches, and then also to quickly make edits and changes. I think we find that I mean, not only do you need that core functionality, because you need to audit things. But just from an emotional standpoint, humans are not always super comfortable with that black box solution. Just on the adoption of these software's or of these analytical platforms, you have to create that easy way for humans to get inside the loop. Because otherwise, you're just not going to find adoption comes, people just won't understand it, so they won't use it.
And on the other side, on the climate risk side, I would say, we use AI in a lot of places, whether it's statistical and machine learning models to interpret past weather events, the signal processing that can turn 3000 signals into 30,000 or 300,000 signals, so that you can develop a great predictive tool. And it could be things like crop yields, you know, how do weather and crop yields correlate? How does that correspond to certain input signals that we see? We use machine learning really well on both sides of it, but it has to be supervised, and you have to have those outputs so you can control it and continue to learn it, because you would hate to do all of this work, and then end up with socio economically unbalanced outcomes.
There's a lot of data gaps, a big barrier for companies not going in this direction, because they feel they have to do loads more work, and it's not going to yield an effective result. So how do we get over this hump?
Fair enough, and getting over the analysis paralysis is definitely a very human reaction. But again, it's humans making the decision, so we have to address it. And I do work with a lot of, let's say, broadly engineering companies who are very analytically driven. Telling them not to worry about accuracy is just not in their vernacular.
Part of the way that we do handle this, at least on the carbon management side, there's a few steps that you can take. Going from corporate level spend baselines, and yes, some more generic, you know, average emissions factors, but moving into much more reliable and granular activity based data. So how much of this did I buy, what volumes, what quantities, what weights. We've been able to characterize power drills and tractors and movies, you know, like motion pictures, really anything that we can break apart, we can characterize from a carbon perspective, and that can be scope one and two, so your own operational energy, purchased energy, and then even into the scope three space.
But the final frontier, I would say, is even moving away from emissions factors, and being able to directly collaborate with your suppliers. I think that's where the industry wants to go. And so those who step into that space, I think, well, it will serve a greater good of creating a very broad global ecosystem of data. And that's some of the work that we're actually doing right now with CDP, which is one of the largest self disclosure organizations in the world. That data will start to come into fruition.
Then you're really unlocking a lot of advanced use cases, where you can not only fill in data gaps, you can create, you know, connections and relationships between different products and suppliers. So that, again, to your point around how can we progressively advance our analytical capabilities as you get more data and as the data granularity increases, of course, you can do more with it, but it's only going to pull harder on some of the machine learning models or any of the statistical work that you do. But the value is there because if you can pick apart your climate risk, you're always going to be able to have a more tactical and operational change, you know something that's more likely to be very tangible and reduce your footprint or reduce your actual climate risk and exposure.
I'd like to focus here a little bit on the financial services sector. Obviously data heavy, data driven, lots to uncover. In your view, is there an opportunity here that we're just not tapping into currently?
I do agree that reform and change is coming, I think you can read the tea leaves with some of the SEC commentary that's been put out for consideration and ultimately, to vote.
You're just seeing a huge uptake in the types of questions that are being asked of companies, the type of data that they have to report, the detailing of both the carbon and the climate risk out of a company. I think people are looking around the corner and they know it will become some sort of a requirement.
But that said, there are many entrepreneurial and enterprising business owners out there, and they say, well shoot, if I need to do this anyway, let's think about how we can turn this into a business. And so I've seen everything from you know, I as an institutional investor or I as a market maker, so like actually a market operator or even just a bank, how can I provide a service to gather a significant amount of data from those who I have contact with? How do I monetize this both to provide benefit to my own company, the shareholders at my bank or institution or what have you, but also to reciprocate and provide some sort of service that takes advantage of that scale.
And then any add-on analytics or optimization layers that you build on top, provide that benefit back to your customers and start that virtuous cycle. I provide a service therefore, people toss in their data, my service improves and so on and so forth.
I've personally been involved in discussions where banks are looking to accelerate climate based financing, so decarbonizing farms and manufacturing facilities, you know, across the world. These banks are looking to us to help them upskill their people, how do I get my bankers more climate intelligent? How do I help them characterize their baseline and therefore can offer back to them, decarbonisation or abatement opportunities, such that they can get more comfortable that this is a real opportunity, this helps my business and therefore, I should ask for a loan from this bank, for example. So the bank is happy, the customers are happy, and in that instance, it's truly virtuous.
In other instances, I've seen market makers and large financial institutions looking for providing a service perhaps around characterizing someone's carbon footprint, keeping in one place all their abatement initiatives. So again, providing some sort of a service that's valuable to companies who perhaps don't have means to do it otherwise, but then they are in the middle of this huge data pipeline. And so then these companies can turn that around and either sell the data as a service. Where companies are finding the mutually beneficial pathway, that happy path is where you're going to see growth. Because the advanced analytics, the user interfaces, and the strategies that are well suited for both sides of the table, I think will catch on and they will scale.