In this interview, we'll talk to a startup co-founder whose daily work is based on using AI to optimize the work of renewable energy companies. Proving once again that AI and machine learning could do anything, even help keep the planet carbon neutral, we encourage you to read this interview with Thomas Sherman.
Thomas's Bio
Thomas Sherman is a co-founder of CRCL Solutions, a Ph.D. in Environmental Fluid Dynamics, and a Forbes 30 under 30 alumni. His company provides power traders with farm, regional, and system-wide wind and solar power forecasts by facilitating AI algorithms. This way, CRCL Solutions ensures the accuracy of wind and solar forecast data to help power traders effectively participate in the US market.
Questions
Nat: Environmental Fluid Dynamics/Mechanics sounds like a complicated and important topic. Can you explain what your job is and what CRCL Solutions do as if I was a 10-year-old?
Thomas: Environmental fluid mechanics is generally how you study the Earth in terms of water and air. We try to answer questions like "How do the oceans move?", "How does the weather form?", "How do rivers flow?" etc. My Ph.D. was focused on such topics and I had a wide variety of natural systems to study. A lot of my work focused on hydrology and how contaminants spread in water systems, like if there's a pollutant spill in a river, how long will it take to go from point A to point B down the river? And how does it spread? The movement of the river is going to push the pollutant downstream, but then it's also going to disperse and spread. So, I looked at problems similar to this one and developed methods measure and understand air and water systems.
And, regarding what CRCL does, we predict how much energy is produced at wind and solar farms in the US. So just like you would go to your local weather station to get your weather report and learn that today at noon, it's going to be 15 degrees Celsius. Similarly, we tell you that tomorrow at 12 o'clock, your wind farm will produce 100 megawatts. That's what we do, very broadly.
To be more specific, we take weather data that is publicly available (similar to the weather forecasts you see on TV). We pull that data to learn what the wind speed is or what solar irradiance is (how sunny it is) on a certain day. Then we take the data and run it through a series of physics-based algorithms to convert weather forecasts to energy production forecasts. Groups in the energy industry access our forecasts to understand what the energy supply looks like for the day in a specific region.
Nat: At what age have you expressed interest in nature's use of air and water?
Thomas: I've been interested in the weather for a long time. I'm from the Washington, DC, area, and when it snows there, the whole city shuts down. So, as a kid, I always got a day off of school when it snowed. Because of that, I was constantly watching the weather in the winter. I didn't want to go to school and wanted a day off. And that's how I got interested in the weather.
Apart from that, I always liked watching extreme weather conditions. I'd watch National Geographic shows about tornadoes or Animal Planet shows like Planet Earth, where you see the mountains with harsh weather or the desert and its extreme heat. I always liked the beauty of that. So, I just liked the weather in general, animals, and other natural sciences.
To be honest, from an energy company's point of view, I knew very little about the energy industry until two years ago. And I still am learning a lot. My co-founder and I had very similar backgrounds, and we knew how to improve weather forecasts. We also knew we had the technology and the expertise to correct basically any public forecast with some new methods, using those buzzwords like artificial intelligence or machine learning.
We understood that wind and solar farms rely on the weather, so we tried to sell the weather forecasts to them initially. What surprised us is that they didn't want the weather forecasts but energy forecasts instead, e.g., how many megawatts are being produced per hour. Then through a series of hundreds of customer interviews in the industry, we concluded that most people mostly wanted energy forecasts, not weather forecasts.
We figured out how to convert our weather forecasts to megawatts (energy forecasts) and so that's how we started predicting energy.
Nat: What was so fascinating about that for you?
Thomas: For me, it's fascinating to learn how the power system works. E.g., turning your light switch on and getting power requires people to produce energy from wind, solar, natural gas, hydro, and other sources. And that's amazing how the power ever gets to you. People have to buy and sell energy; there are all kinds of financial interactions in the background that I was never even aware of (before I started working in the industry).
So, just learning how the whole system works, how complicated it is, it's fascinating. I'm still learning, and it's so complex it's shocking that it actually works as a system because there are so many steps, from turning your switch on to power getting there.
Also, I didn't know commodity trading existed. Commodity trading is the place where you can basically bet or take financial options on all kinds of different commodities. And energy is one of them! But! Some people bet on all sorts of things like cattle, minerals, and coffee and there are all these really wealthy people who made their money in these really obscure markets.. E.g.. Personally, I didn't even know such a thing existed.
Nat: Was it difficult to combine the roles of scientist and entrepreneur?
Thomas: It was definitely difficult. There are all kinds of rules that you learn from books or podcasts. For example, according to those, you need to have a business plan and go to market strategy. But in the beginning we didn’t have these, we were still learning what it means to start a business.
So, we learned you don't need to do everything at once! As you and your business mature, those rules and requirements make more sense, but you don't need to start with all of them ready at hand. What you really have to understand is what your customers want and what they are willing to pay for. Because if you don't have customers, you don't really have a business. You have a project instead.
Among all the things for your business, getting customers is the hardest part. Everything you theoretically need for your business isn't required until you have actual customers. Usually, people think when they start, they need to raise venture capital, organize marketing and sales, and have all these pieces in place.
But the initial thing you need is to know what your product market fit and who actually wants your product. That's where doing all those interviews really helped us in the beginning. We got an excellent understanding of what the market actually wants. But, finding out what people will pay for turned out to be very challenging. So, combining those two is indeed a learning process.
"We've learned that what people need and what people are willing to pay for are very different things."
We made two products that didn't work. Initially, people said they wanted one product but weren't actually willing to pay for it. That's where we learned that what people want and what they're willing to pay for are very different.
After we've found the golden middle, we've slowly matured and started adding pieces in place to make it more like a business. We went about a year without money in terms of funding. We were fortunate to be in a position where a few factors conveniently came together: we didn't mind living very simply at that moment, and my business partner's father had a house that he couldn't rent out because of COVID. So, for three to four months, we stayed there and reduced costs as much as possible to get the business started.
Financially making that transition was a hard part. It was very stressful in the beginning, but the fact that we could find that product market fit and actually build something that people would pay for was a great headstart for us.
From a technology point of view, I think a lot of scientists and engineering folks really focus on creating great technology. But people ultimately want your technology to provide a certain service. So, you have to figure out what that service is and articulate the technology in a way that is usable for whoever your customer is, so they can actually use it to solve their problem.
My advice is to fail and fail quickly. Don't be a perfectionist. It's perfectly fine to give people half-ideas and let them critique and design your product for you. The market very much designed our own product. If you want to quickly get to what people want, what's good, and what's bad, then show it to the people in exchange for honest feedback. Even if you show people something that sucks, they'll tell you! So then you won't spend time and money on something that sucks and no one wants.
Get our premium e-book with SaaS and Tech interviews + industry predictions for 2023 for free.
Nat: How much are AI and Machine learning involved in the process of your forecasts?
Thomas: So, we take the weather data that is publicly available and make it better. Right now, we're using NOAA, the US weather agency, which has all their historical weather forecasts available via an archive. We take this archive of NOAA forecasts and optimize them over a number of variables.
If you take the weather forecasts with temperature, rain, cloudiness, and wind speed (which is a particular point of interest for us), what's best for the whole US forecast is different from what's best for a single site like a wind farm.
Using machine learning or AI, we train the models to understand where the NOAA model biases are. For example we try to learn trends, like if the temperature is really hot, and the wind's blowing from the west, then we typically know that the wind speed is over-predicted. Our machine-learning models will correct such forecasts in real-time. So, our technology has an AI component. We show our models what the NOAA forecast is, where error happened, and train the model to correct that forecast. And then, when NOAA releases forecasts in an operational setting, our AI makes the correction in real time and produce pretty accurate weather data.
Nat: How precise are your forecasts?
Thomas: We did a case study in Texas (one of the big US energy markets with more wind energy, which is very important for us). We improved forecast accuracy using a metric mean squared error by 30% compared to what a NOAA model currently produces. And then it's ultimately turned into power, which is harder to validate. Wind farms and solar farms usually don't register how many megawatts they actually produce. So, we have to trust our physics-based algorithms that they can properly convert weather into energy.
Nat: If you met 10-year-old Thomas, what would you tell him?
Thomas: I would say: "Just do what you think is interesting." One mistake I've made is choosing jobs for financial reasons versus genuine curiosity and interest. And, usually, when you follow your interest, curiosity, and passion, things work out better, even though, in the short term, it might be a sacrifice of taking less money.
Listen to your gut.
Rapid-fire round
Your biggest regret?
Honestly, I don't really have regrets. You make a choice in life, and you learn from it.
What does it take to be on the 30 under 30 list?
You pick a particular, less popular topic (like energy), and then they're looking for 30 people to fill that topic. If you like music or sports, it is much hard to get because lots of people are doing it. But if you pick science or energy, for example, how many people do you know who are doing things in these fields on their own? I think there are not that many of them, so the 30 under 30 in these categories is probably not as impressive as it seems.
What does it give to be on the 30 under 30 list?
Of course, it's a cool thing to have, in terms of the network. It opens up opportunities. Personally, I was approached by an accelerator in Italy at the University of Genoa, and they invited me to help out with the accelerator.
But if you expect to get it and then your business takes off, it's not going to be like that.
What's your forecast for 2023 in one sentence?
The world will become more stable, and we will move in the right direction.
Get to know Thomas