Forecasting for the Renewable Power Grid with Myst

Today's guest is Titiaan Palazzi, COO and Co-Founder at Myst. 

Myst is a machine learning platform that leverages AI technology to improve demand and supply forecasting for an increasingly renewable power grid. Their platform helps companies accelerate the deployment of expert forecasting solutions that not only increase profitability but also reduce risk. 

We have a great discussion about the different energy use cases that more accurate predictive modeling can help, the customer types that use Myst today, and the transition of data science talent into the climate space.

Enjoy the show!

Cody Simms: Titiaan, welcome to the show!

Titiaan Palazzi: Great to be here.

Cody Simms: Well, I'm super excited to have you on here today. You were originally referred to us by Jules Kortenhorst at Rocky Mountain Institute, who was a guest on the show a couple months ago. And I know that the two of you worked together previously. You were at RMI. So maybe let's start with just a little bit of your own background. We'll obviously dive into myst.ai and what the company does, but at first we'd love to hear from you. How did you get into working on climate change in the first place?

Titiaan Palazzi: Definitely. Happy to share. Academically, I studied environment and engineering originally in the Netherlands, which is where I'm from. And then that also led me to come to the States. So I did part of my Masters at MIT in 2011-2012. And then had the opportunity through a few different paths to work with Amory Lovins, founder of RMI for about two years as his special aide. Amory had been a mentor of mine since I was in high school. And I had always greatly admired his perspective on the energy transition that really business can be a force for good in this transition and that we can bring everyone along by speaking to different constituents individual benefits and what they care about. So Amory brought me to RMI, and RMI was an amazing place. When I joined, we were I think about a hundred people. Jules and I joined roughly at the same time. So he took the helm of CEO about that time in 2014 or so. So it grow to about 200 until I left. And it was such a great learning environment with super smart, kind, knowledgeable people, fully committed to the energy transition. So it was a wonderful schooling environment.

Cody Simms: Yeah. And they've done great work with things like Third Derivative and everything too just in terms of really supporting, not just business efforts broadly in the climate space, but entrepreneurial efforts in particular, which I think is really powerful. And obviously I believe they're involved with Canary Media and just a whole host of different ways that they've helped bring more people into the energy transition, which is great. So also a big fan of the work they're doing. So you did some time there, built a good foundation of knowledge that complimented your educational background, and then you decided to jump into doing a startup. How did that path go for you?

Titiaan Palazzi: So I think maybe this is a good point to sort of like share two perspectives. So from my side, mine and then of my co-founder, from my side in the years I spent with Amory and then later I was on RMI's electricity team. One of the things I could witness from very close by was how different players in the energy transition were dealing with increasing amounts of renewables on the grid. So we would work in types of different types of technical consulting with a variety of utilities mostly in North America for my part. And we could see how they were struggling to understand, should we place batteries here or there on our grid? And how do we run this thing if we now suddenly have all this variable renewable energy generation? So I saw if we are going to go to a carbon-free power system, there are certain blockers that need to be removed.

Titiaan Palazzi: And this specific one around making sure the whole grid works in a reliable, carbon-free and low cost manner was one where I thought technology was a particularly good fit. I co-founded Myst with Pieter, my co-founder, who I knew from the Netherlands. And when I went to MIT, he actually joined Nest in the early days. I think there were fewer than a hundred people there then as a software engineer. And he helped Nest develop the original demand response algorithms rush hour rewards, and then built a variety of large scale software projects for Nest that leveraged machine learning in a variety of ways.

Titiaan Palazzi: He was there when they were acquired by Google for $3 billion and then stayed at Google for a long time. And so he had a very complimentary view and skillset where he had seen, look, when it comes to predicting the future in a sort of like short term manner, machine learning is an amazing fit, but it's also really tough to do this well because it's highly complex. You're getting data from millions or billions of data points. There is an opportunity to do this in a way that sort of like is really standardized and can be leveraged by many people. And so that was the nexus of Myst that led us to start the company together to explore initially. And then later start the company together. He left Google. I left RMI in 2018. And that was the origin story.

Cody Simms: Super cool. And actually, Tony Fadell from Nest was also just recently on the My Climate Journey podcast. So fun to hear the convergence of those pasts as well. Maybe talk a bit about, okay, so you two knew each other, you each had these disparate experiences, but working in strangely similar problem sets. Maybe let's start with actually revealing what is myst.ai and then talk about how you got to realize that was the specific area you wanted to focus on.

Titiaan Palazzi: Yeah. So what we do at Myst is we help companies, all kinds of energy companies. So renewable power generators, load serving entities like utilities, but also distributed energy resource companies. So maybe EV charging companies or companies that install behind the meter batteries create highly accurate short term forecasts that they used to operate whatever they operate to manage their electricity demand or to make sure that their solar and wind assets are earning the revenue that they committed to their investors, or to make sure that a building minimize its energy costs. That's what we do. We got to this space really, I think back from those sort of like original perspectives that I shared. But maybe to sort of like add a bit more color there. So in 2018, we started. And in the early days we focused primarily on basically working very closely with a handful of load serving entities.

Titiaan Palazzi: So this is energy jargon. Basically, it means any kind of company that serves end customers with electricity. They could be utilities, but sometimes they don't operate a grid. And so technically, they're not a utility. In the United States, we have naming things. And so we'd spend time with them. We would go on site. We'd spend a week or sometimes weeks working with our technical teams to really figure out what is challenging when it comes to managing your electricity demand, what is the financial impact when things don't go well, how does this change as more renewables are added to the grid, and what's a place for us to help? Those early explorations, it was really user research, led us to conclude that both on the shifting winds, so in the sense that the weather is changing, we see increasingly frequently extreme weather events, energy demand patterns are changing. So for example, in California, we went in the last 10 years from very few homes with rooftop solar to lots of them. In Australia, this is even more strongly the case.

Titiaan Palazzi: Electric vehicles are being added. Periods like COVID dramatically shift energy patterns. So one of the problems we unearthed was that this led to these companies basically flying blind, and there was really an opportunity to add technology to the mix, both to reveal these kinds of patterns more quickly in a way that they were not facing undue risk, to enable them to integrate more renewables, and also to do this in a way where the technology they were using was highly reliable. Because the other thing is you need to make decisions every day or maybe even every minute, and you cannot face a system that fails.

Cody Simms: Yeah. So if I understand, I live in California, so I'll come at it with a use case that I understand, which is residential energy use in California. As you mentioned, over the last decade you've seen an increased number of rooftop solar projects going live from a distributed energy perspective. So you're seeing that, I assume, the demand curve less spiky in the middle of the day, because now you're able to power ... you're feeding more electrons back onto the grid through solar, but at the same time, you've also seen a pretty substantial growth in EV adoption. So I'm guessing between 6:00 to 8:00 PM, all of a sudden you're starting to see these huge spikes in energy usage when people are coming home from work and plugging in their EVs. So that is, I assume, a gradually ... both of those are gradually growing trends over time that presumably you're helping your customers understand, but then you also have sort of real time events that are happening like an upcoming weather event that you're seeing coming through. Or I don't know, you tell me what some of the other things that may be hitting. My understanding is you're helping these customers sort of measure the inputs of both of these in terms of being able to supply the right amount of power to the grid. Am I understanding correctly?

Titiaan Palazzi: Completely. So if we stick with the ... because we serve a variety of customers, but if we stick, for example, with the California utilities for a while, so we serve a variety of mostly community choice aggregators, so they might be providing power to your home. For example, in 2020, you may remember we had the August and September heat waves, fires, orange day, Mars day, in that time when the California ISO and the California Energy Commission did their retrospective on what happened, one of the three root causes they identified that led to these outages was actually the fact that individual market participants had not been accurately predicting what their energy demand would be and it's made complicated exactly because of the reasons that you mentioned. So we see part of our job is to make sure that all the data that has predictive value is there for us or for our clients to make those predictions so that the grid is more stable and individual market participants don't face up the risk.

Cody Simms: What's the status quo without you? How have they been managing this in the past?

Titiaan Palazzi: Well, I mean, that's one reason why the technology progresses but also in concert often with market demands. So I think one thing is that a lot of this is enabled by AI, cloud based infrastructure, et cetera. But a lot of it also comes from the demand side. So 10, 20 years ago, you could probably be okay by taking similar day values or even just using averages, like what we did yesterday is probably fine today. But that no longer cuts it. So often the status quo is an Excel spreadsheet where you take the average over the last few months and changes such as more EV adoption are either put in by hand or are not incorporated at all.

Cody Simms: We talked about California use case. Let's take a different use case like Texas, right? We had the big extreme event a couple winters ago where they had abnormally cold winter. Texas, I think, has a very different grid mix than California. They're mostly powered by gas during the day and then at night, heavy wind usage on the Texas grid, as I understand it, and all of a sudden you have these abnormal cold snaps coming through a place that doesn't normally freeze. In what way could Myst have helped there and again, what was the status quo in that environment?

Titiaan Palazzi: Yeah. Maybe what I'll emphasize is that we've so far spoken about the load serving entities, but this is really a variety of players, all of whom we serve. So in Texas, actually, in addition to some big retail energy providers, we also help a variety of owners of battery storage and renewables. So solar and wind, sometimes those are coupled, but often they're also separate. So we might help those kinds of companies to anticipate what might come and therefore ensure that for example, their grid scale batteries are fully charged because there is a high probability that some kind of price spike will occur. Let's also be clear, I think that's important to do, what happened during Uri in February of last year was not something that a more accurate forecast would have eliminated. This was in part due to structural issues with insufficient investment in weatherization of gas lines and pumps and other things, but it could have been a contributing factor to help some of these companies reduce the amount of generation that was offline.

Cody Simms: Awesome. Yeah. So let's maybe dive into some of the different use cases for forecasting. In fact, one of the things I saw in a blog post you all recently published was the U.S. power grid boasts something like 99.9% up time and still has more power outages than any other developed country, which I thought was a really interesting data point. But in that same article, you sort of talk about the four use cases for forecasting, one being balancing the grid, two being optimizing renewables, three being dispatching battery storage, which you just talked about a little bit, and four being operating virtual power plants. Maybe go into each of those a bit and just unpack how a proper forecasting mechanism can support each of those use cases with better accuracy than how they're operating today.

Titiaan Palazzi: Totally, and then I'll also sort of tell you why we enumerate it in that way. Because I think we have a vision for what we are that actually comes more from the non climate space more from the technology space. So I think we touched on some of these already. So for grids, for the loads of entities within them, but also for the independent system operators or the other kinds of entities that are balancing the grids, one thing that is absolutely necessary to ensure that there are no outages, especially on the sort of shorter term day by day or hour by hour basis is to have a good insight into renewable production, thermal generation production. So fossil fuels, gas, coal nuclear, as well as energy demand.

Titiaan Palazzi: So for these kinds of entities, which is not to date our primary customer segment, they need aggregate level forecast sometimes built from a sort of disaggregated manner. In the sort of second and third categories that you mentioned, which is where we see most of the growth of our customer base, which is front of meter or utility scale solar and wind assets and battery storage assets, what we're seeing is the world has shifted in the last 10 years from most of those kinds of renewable types to have long term, fully guaranteed offtake contracts with no market exposure, to basically all of them having some kind of exposure to the market. So what matters for these kinds of companies is that they are dispatching their assets in a way that serve the grid, decarbonize power, but also ensure returns on their investment.

Titiaan Palazzi: So we help those kinds of companies typically with a variety of price forecasts as well as if it is solar or wind production forecasts that then really lead them to make their daily or hourly decisions, sometimes with trading teams or market operations teams, sometimes in a fully automated fashion, which we see more of, especially when it comes to batteries in [inaudible] for example, which are often dispatched on a five minute basis. Then the final one that you mentioned, which is actually I think a important growth segment for our company if we looked at the five or 10 year horizon, is everything DERs. So EV chargers, behind-the-meter, battery storage, energy demand, these kinds of things. We're starting to see more and more inbound attraction to our products from those market segments, companies that say, "Look, we are optimizing energy demand across 20 industrial sites or 200 industrial sites. Can we use your technology and build our own forecasts. And so, we see a lot of growth there too.

Cody Simms: DERs being distributed energy resources for folks who aren't fully up to speed on all of the climate and energy lingo. So things like solar panels and ways that people can generate energy outside of utilities.

Titiaan Palazzi: Yeah. Thank you for adding that. Totally agree. But then, maybe to sort of share why do we serve so many segments, I think really when we look at Myst, the lesson we see, the vision we see for the company is that we are kind of a layer in the tech stack of all of these kinds of companies. So any company operating in electricity markets in the next 10 years needs to have predictive power and we can offer that to them, very much like a company might use Stripe or Shopify to enable their online payments or in store. And that generic technology approach allows us to unblock parts of the things that need to be in place for the energy decision to happen.

Cody Simms: As a startup, has it been hard to go to market capabilities across so many different customer types and needs?

Titiaan Palazzi: Yeah, I think so. So I think we've gone through them sequentially, and I think that sometimes we're focused on one and others come to us. I think we've been very disciplined. And so, I think that has made it a lot easier. I would say the other thing that we've found is that it is imperative in any of these segments that you build domain expertise, even if you are just creating tools for your customers, to be able to empathize with them, actually understand where value originates, and also guide them in building these predictive models, which at the end, really depend in large part on the quality of the data that goes into them.

Cody Simms: So going through those use cases, just want to kind of touch on sort of what I heard from you and how that might apply to what I understand, being somewhat the theory of change, of how Myst not only is a viable business, but how Myst helps with climate change. Balancing the grid, which I assume are these mostly large enterprise scale utilities and whatnot, and they potentially could use you for helping to divert more supply to being produced by renewables, based on what they're seeing. But I heard you say that the bulk of your current business comes from optimizing renewables and dispatching battery storage. Which to me, what I'm hearing from your theory of change is right now, what you're trying to do is make the renewables business more profitable, which will thus result in more renewables projects winning out in terms of getting deployment. Is that broadly accurate?

Titiaan Palazzi: That is very accurate. So I'd say high level theory of changes. This is a foundational component to the energy transition across these four buckets. Let's make it easy, fast for everyone to do it. In renewable and storage asset owners, that is indeed the theory of change.

Titiaan Palazzi: Maybe to give some numbers, where we are now is we have about 1,000 gigawatts of solar and wind installed globally, roughly for each. And we need to get to 5,000 of each by 2030, according to BNEF and IPCC. So that's five X growth in seven years. And so, anything that can be done to make more investments flow there. But also honestly, to unblock scaling of the renewable asset developers and owners is critical. It's where again, IPCC says 75% of the emission reduction in the next eight years is going to come from new solar, new wind.

Cody Simms: And so, are these projects using you in the upfront financial projection side of things to anticipate IRRs? Or are they using you in day to day operations for the most part?

Titiaan Palazzi: Great question. Right now, exclusively in day to day operations. And the reason I use the word, sort of accelerating them, is that what we've seen for clients is that if you would do all of the work that we're doing in a truly open source manner, you're probably spending one to two years of software engineering time with a 5% to 10% team doing this and we simply do not have that luxury. As a society, but also as individual companies, they cannot wait so long. So we're making sure that this specific part of the stack of their activities is unblocked and they can move faster.

Cody Simms: And what are the primary data sources you're pulling in? How does a project operator today actually leverage you and what are they using from a decision making perspective in terms of the data you're providing? I guess, if I'm running a utility scale solar plant, solar farm, what am I actually getting from Myst? And how is that helping me?

Titiaan Palazzi: Yeah, totally. So I think that if you are a utility scale solar farm and you consider working with us, there are basically two things you need to believe. The first thing you need to believe is that you care or you will find it attractive to own some of the analytics about your assets in house, certainly in the long term. Because part of where we have gone from a product perspective is enabling our clients to contribute to, or to own IP of their predictive models. That's the first. The second thing is, you need to believe that by capturing the physical phenomena in machine learning models, you can predict the things that matter for you with high accuracy.

Titiaan Palazzi: And so, to then answer your question of, "What is the data that you provide?" It depends on the use case. If we're talking about renewable production, for example, we offer weather data inputs or historical and forecasts from our library of third party data providers. If we're talking about price formation, you're looking at a much larger variety of data. So in addition to the things I mentioned, also historical prices, gas prices, are there scheduled outages for generators or for transmission lines, all these kinds of things. Weather, of course, is a tremendous factor. And then either we or our users on the platform stitch those things together into a model and then assess what has the highest predictive power.

Cody Simms: And utility scale solar farm may not have even been the best customer type for you. It sounds like it's going to be more large scale demand response aggregators and the like, who are probably more likely using you in the near term, is that accurate? Or I guess, it's all of the above.

Titiaan Palazzi: Exactly, it's all the above. So big solar farms now, especially the newer ones, often their revenues depend on the market price. So they can do smart things around when they sell, if there's dispatchability because of batteries, when they dispatch. And so, shorter term predictions matter a lot.

Cody Simms: Excellent. And I'd love to talk about how you've sort of financed the business to date. I know you've raised a couple of rounds of financing. It seems like Gradient Ventures, which I think is Google's AI fund, led your initial seed round. And I'm sure it didn't hurt that Peter had a nice Google background coming into that. And then, you raised a series A couple years ago now from Valo Ventures. And maybe share a bit about what that pathway has looked like and sort of how you view needing to capital... You're a software based business, so I assume you're a capital light, but would love to hear how you view needing to finance the company.

Titiaan Palazzi: Totally. Your historical perspective was accurate. So Valo and Gradient are primary investors through the two rounds we've done so far, about 8 million to date. And then, we're preparing for our next round.

Titiaan Palazzi: And I think you're totally right. So 85% of our team is engineers and data scientists. Right now, our technical team is fully in the Bay Area. And as we grow, we'll grow that. So we'll grow everything that's product related, but we're also going to invest significantly more in our go to market. Because a large part of making this successful is actually having a really close collaboration with our clients. And those today sit exclusively in the U.S. and Europe, but we see a future in which we serve other areas of the globe too.

Cody Simms: And from a competition perspective, do you view competition coming from other players who are doing almost the same approach as you, but just claiming to have better accuracy on forecasts? Do you see competition coming from companies that have other forms of revenue coming in, like a demand response aggregator, who has to get good at forecasting for their own use case and decides to platformize their technology? How do you see the market evolving for you over the next few years? I guess, you also mentioned that you saw that helping with operationalizing a virtual power plants being a large growth engine for you in the coming five plus years. So that may be some degree of signal of how you see your business evolving, but curious how you see the market landscape changing.

Titiaan Palazzi: Totally. I think that's honestly one of the most exciting parts, because even though we started in the first two years, in the first two years all of our revenue came from us essentially building predictive models for clients, productionizing them, and giving them access through APIs and web dashboards and these kinds of things. But now increasingly, we are actually enabling clients to build their own predictive models. And that's super exciting. I think we really look at Myst, it's also what we call our product, much more as a platform that enables clients. And so what that means is that really when we have early conversations with technology leaders and our clients, often the way we describe it is in this way, you have a choice of a spectrum of choices. On the one hand, you can go with end to end solutions. So those might be forecasting services, companies that are really good at predicting market prices or predicting wind production at a site or predicting energy demand. In each of those segments, there is at least a handful of players.

Titiaan Palazzi: Or you might go for end to end more integrated solutions. So you might go for companies that offer a full asset management solution for your battery, for example. And I think that's fine. I think the benefits there are that it is probably relatively fast and you do not need to worry about building in-house expertise. On the other hand, if you say, "Look for the value of our company and for our future revenues, it is imperative that we build expertise, that we have something to say about how we operate this group of assets or our log portfolio." You could build everything in house and you could hire five or 10 software engineers and build this from scratch, leveraging open source tools and the big cloud providers' toolboxes, but that's probably going to take you a while. It will take you... To configure everything so it works for you, it will take you two years, three years, if you're lucky and then there's maintenance on top of that.

Titiaan Palazzi: And so what we say that there's really a third way where we sit in the middle, where we from day one, enable data scientists and analysts to build predictive models. You can do it totally yourself, all IP resides with you, or if you need some of our help in the early days, we can help you build your first models based on our in-house expertise. But we're taking you on a journey where we're helping you build this own internal IP in a way that it is not dependent on any one engineer on your staff, but it sits in a platform where it can be used across engineers. And you're not being held back by two years of build it yourself investment.

Cody Simms: And so really for that key customer you are a data API provider who you just have to be really good at anticipating what datasets they're going to need access to and ensuring that you have the most accurate versions of those datasets for them to utilize. Is that correct?

Titiaan Palazzi: That's a core component. But we are really what we call, to use some tech lingo, a machine learning operations platform for everything time series. And so-

Cody Simms: So, they'll build the models on your technology stack?

Titiaan Palazzi: Yeah. Totally.

Cody Simms: Okay.

Titiaan Palazzi: And productionize them [inaudible] our stack.

Cody Simms: Okay.

Titiaan Palazzi: And we're responsible to make sure that every minute the models run as necessary for them.

Cody Simms: So drawing parallels here, obviously there's so many parallels between what's happening in energy and what has happened over the last decade in cloud compute. Drawing parallels here, there's obviously huge businesses that... SAP and IBM and others, and Google operate and Amazon operate in terms of enabling you to build predictive capabilities for cloud compute power. I'm hearing similar feedback from you. It's just where the datasets are around electron usage and exogenous effects like weather.

Titiaan Palazzi: You're totally right. And so to put another thing out there, a belief, is that what we are seeing and what you must trust, the thing that Myst is going to be a big company, is that there is value of the domain expertise, even in a generic platform. And because we've spent four years and we're investing a lot in building that expertise also on our own team, that allows us to not only bring in the right external data sources, but also to unlock certain functionality on top of the generic functionality that really helps these companies. But you are right. I think that the shift that you described is very similar and we see it, we've seen it happen in the last 10 years.

Cody Simms: Yep. Microsoft too. Don't want to not give Microsoft a shout out. They obviously are a big player in this space also. So what's live today and what's next?

Titiaan Palazzi: Yeah. So the company is about four years old. We're about 20 people, 20 full time staff. We're getting close to 20 enterprise clients in the US and Europe, including some of the largest. So the world's three largest power companies are all Myst clients and then many more. And as I shared earlier, we're getting ready for our next phase of growth. So growing the company to really expand, both within some of the companies we already serve, as well as accessing new clients. And one of the things that's been most exciting for us, also for our technical team, is to see that after years of product development, we're now really seeing an acceleration of this self-service approach. So the growth we've had in the last six months is as much as we've seen in four years previously. So real scaling.

Cody Simms: That's not bad. Titiaan, what didn't I ask? What haven't we covered that is important to make sure to communicate?

Titiaan Palazzi: I'll close with two thoughts. I think one is the change we see in electricity is a tremendous opportunity. Already today, there is about $1 trillion spent globally on electricity yearly, about a percent of global GDP. And it's only going to be more, it might... $2 trillion, it's gigantic. And going back to the IPCC thing, this is one of the most urgently needed shifts. So I would say anybody who is working on accelerating solar and wind deployment go hard. We need this.

Titiaan Palazzi: The second thing I'll say is, from a technology perspective, this is a really exciting challenge. And so what we're seeing is the shift we see in talent of all kinds, but including software engineering, to the climate space is so needed. And we're really grateful for all the great people who have already come our way through communities like MCJ. And as I shared, we are scaling. And so anybody listening to this who feels particularly drawn to this kind of challenge that we're going after, please reach out. You can find us on LinkedIn and on Twitter, send us a DM and we're happy to chat.

Cody Simms: Yeah. And I will just underscore, the amount of talent in things like data science and machine learning at MCJ Collective that we're seeing every day join our member community and put their hand up and say, "Hey I've spent the last decade optimizing advertising yields. And I'm realizing that my skills can maybe be put to work in other areas that I care about personally, like helping to grow renewable power generation or helping to work on things like synthetic biology." That's all just really inspiring. And I think for anybody out there who's in data science or machine learning, obviously it sounds like Myst would be interested in hearing from you if you're interested in moving into that space.

Cody Simms: Great. Anything else you need help with right now? You've got you've got our audience and our listeners and our member community tuned in to you at the moment.

Titiaan Palazzi: Great talent, please reach out, share the word. If you are running a solar wind battery storage operator, or you're running a climate tech company and you realize this is important, I see an opportunity for partnering. Please also send us a note. And Cody, thank you for having us

Cody Simms: Titiaan, thanks so much for your time.