Resources

• Sagewell Inc.
• Peak Load Management and Energy Efficiency
• USEPA: Smart Metering and AMI (Advanced Metering Infrastructure)
• International Energy Agency: Electrification and Decarbonization in Energy

Host and Guest

Climate Rising Host: Professor Mike Toffel, Faculty Chair, Business & Environment Initiative (LinkedIn)

Guest: Pasi Miettinen, CEO of Sagewell (LinkedIn)

Transcript

Editor's Note: The following was prepared by a machine algorithm, and may not perfectly reflect the audio file of the interview.

Mike Toffel:

Pasi, thank you so much for joining us here on Climate Rising.

Pasi Miettinen:

Oh, thank you so much for having me. This is an absolute pleasure.

Mike Toffel:

So it's always a pleasure for us to bring back some of our alumni from the MBA program who were doing work in the business and climate change space, and we met about six months ago when you were here once before, and we're delighted to have you back. Can you tell us a little bit about your background and how you ended up as the CEO of Sagewell?

Pasi Miettinen:

Yes. So after college, I ended up being a load forecaster for a utility, and when I came to HBS, one of the first classes that I took, I decided I will never go back to the energy business, and I ended up doing the Wall Street route and doing a lot of other financial services. And then about 15 years ago, I came back to the energy markets and have been enjoying it ever since.

Mike Toffel:

Wow. So what led you back into the energy space, a place that you said you would not go back to?

Pasi Miettinen:

Yes. It was maybe about 15 years ago, we were looking at the impact of energy efficiency programs, and we had sold a previous company and there was an opportunity to look at new things to do, and realized that the energy efficiency programs had a lot of opportunities to do things better. So we started the company 15 years ago looking for ways to deliver better outcomes on the energy efficiency side.

Mike Toffel:

Now, who is the driver of energy efficiency? These are on the utility generation or on the user side?

Pasi Miettinen:

See, these are the residential customers on the utility side. So we were trying to figure it out, because that's what defines the peak load for electric utilities and the systems as a whole. We also looked at the opportunities for just trying to deliver maximum efficiency for dollars spent. So a lot of the themes that we'll probably cover today are about how do we deliver the maximum results for climate results or economic results, or particularly for the utility customers and then the utilities that are our customers.

Mike Toffel:

So, we talked about peak load already. I want to make sure that we're just setting the groundwork to make sure folks understand it. So tell us a little bit about peak load, what that is, why that affects electricity costs and pricing, and then the idea of shaving the peak you hear sometimes, which I think is the domain that you're in. Just give us a little one-on-one.

Pasi Miettinen:

Yeah, so the simple context is that the utility business has really two parts to it. One is serving energy or delivering energy to the customers, but then the infrastructure is built for literally one hour of the year. There's a massive infrastructure that is dedicated to serving one hour of the year, and the rest of the year doesn't come anywhere near that consumption level. So you have the cost of energy, to procuring that and delivering that to customers, and then a massive infrastructure both on the generation side, as well as on the delivery of the distribution system to make sure that the power doesn't go out when you have the coldest day of the year or the hottest day of the year because those are the days when you have those peak consumption hours.

Mike Toffel:

Got it. So the whole system is designed to make sure that it can deliver heat and cooling when people need it on the most energy-intensive day of the year?

Pasi Miettinen:

Yes, and it even gets interesting from an economic point of view or from a business point of view, the cost structure of utilities is such that it's enormously expensive to sell electricity during those peak hours. So it can easily cost three or $4 dollars a kilowatt-hour to procure that electricity that they then sell at 15 cents or 20 cents a kilowatt-hour, or in Massachusetts, 40 cents a kilowatt-hour. So it's enormous economic losses during those peak hours effectively, but we get charged more during the rest of the year to make up for those losses. So there's a significant amounts of dollars that are involved in serving that few hours a year.

Mike Toffel:

Got it. So you say it's so expensive because we're using the most expensive generating assets to create that last bit of power?

Pasi Miettinen:

Yes, literally, and they might run for 20 hours every three years, and that's it.

Mike Toffel:

Wow, wow, fascinating. Okay, so tell us a little bit about Sagewell and how it tries to shave that peak.

Pasi Miettinen:

Yes. So a little bit of context for this is that as we were working on energy efficiency programs, about seven years ago, eight years ago, we started looking at meter data and we said, "Are we actually having an impact? Can we measure the impact in a way that is really significant? If we are making really significant headway on energy use reduction, we should be able to see it in the meter data." So we started doing analytics and, in the process, we discovered that we were having an impact.

Mike Toffel:

What's the programs that were having an impact?

Pasi Miettinen:

The energy efficiency programs, so basically insulating homes or weatherization in general, all these initiatives that are out there, like Mass Save, it's a very popular program in Massachusetts and there are similar programs all over the country. But in the process, about 10 years ago, we started running electric vehicle programs where we were marketing electric vehicles, and we were doing off-peak charging programs. We also started promoting heat pumps for utility customers. And then we started comparing the results. And what we discovered pretty quickly was that we were getting more significant carbon reduction impacts from doing electrification, getting people to switch to electric vehicles or to electric heat pumps than we were getting out of the energy efficiency programs.

So we actually switched the business over. So about seven years ago, we were really focused on data analytics for utilities because we started realizing that there's a real value in actually understanding how people use energy. And then when we presented the utilities saying, "Look, you have all these opportunities either for climate purposes or economic purposes," they often came to us saying, "Well, we can't run those programs. We don't have the staff to run the programs." So we ended up then doing the marketing of the heat pumps or the marketing of the electric vehicles and running the peak reduction programs for the electric vehicles in particular. So it was all about the focus on, "How do we deliver the maximum results for these programs for the dollar spent?"

Mike Toffel:

Now, these energy efficiency programs, these are, if I remember correctly, funded by the utilities-

... that then customers, whether they are commercial or residential, maybe industrial too, can avail themselves of these programs usually for free or at nominal costs to get an assessment. "What are my energy efficiency opportunities?" Sometimes their opportunities might be insulation or, "You need new weatherization," and a lot of those are themselves subsidized or even free.

Pasi Miettinen:

Yes.

Mike Toffel:

And the purpose behind these being funded by the utilities is because it allows them to continue serving customers without having to invest in yet more and more generation capacity. Is that right?

Pasi Miettinen:

That is indeed the theory.

Mike Toffel:

Yeah.

Pasi Miettinen:

And then one of the things that we discovered is that it's not quite that simple because a lot of the program decisions and funding decisions, literally billions of dollars, are based on engineering estimates. They're not necessarily measured on actual measured outcomes from the meter data. So one of the fascinating things to us was we discovered that things that we thought were not impactful turn out to be very impactful, and some of the things that got significant funding were actually not producing the outcomes that we're hoping for because if these programs were effective, you would see significant reduction in the energy consumption on the energy efficiency side. On the electrification side, you should theoretically see significant increases in electricity use because they no longer use natural gas or oil for heating or gasoline in their car. So we started going, "Hmm, there's something else going on here." And so the question comes then, where do you deploy the resources? Because if the outcomes aren't what we're hoping for, how do we actually deliver the maximum impact?

Mike Toffel:

Yeah, that's really interesting. So what's an example of an efficiency program that, based on engineering standards, should have been leading to less electricity demand, but in fact, based on your data analysis really weren't having that impact?

Pasi Miettinen:

Much to no surprise, a lot of the weatherization work didn't produce the impacts that we were hoping for. But later on, it became relevant in the context of installing heat pumps and being able to make sure that those heat pumps were able to deliver the necessary heat during the very, very coldest days. So we found that there were complementary relationships, like weatherization in the context of a heat pump conversion could make sense. Weatherization in a house that uses very little electricity or energy in general probably didn't make sense. So one of the things that the meter data analysis quickly shows is that about 20% percent of the customers are responsible for 50% percent of the peak load as well as 50% percent of the energy used. And as you go through the analysis, and you quickly realize that it really matters who you go after. So if you want to deliver results, if you want to deliver the climate impacts or the economic impacts, you really need to identify the customers who are the best fit for those programs.

Mike Toffel:

No, I can imagine, my first blush of who would benefit most from these or where you'd move the needle most would be on the most energy-intensive customers, but with large houses, for example, or particularly, which require a lot of space heating. Or places with big SUVs where you could electrify it and have a more dramatic impact on reducing greenhouse gases by transitioning them from fossil fuel to internal combustion engines to an electric vehicle. But first of all, is that true? And secondly, if it is true, then isn't there a political concern about most of these subsidies gravitating toward wealthier customers?

Pasi Miettinen:

Yeah, it's an interesting question and we'll get to the equity piece of it in a minute, I'm sure. But when we look at the data, it's not necessarily the biggest houses that are the biggest consumers, they can be. But there's a second piece of this, and one is, and we'll come back to the economic impact, the utilities, and why they want to do this. But if you actually look at what time of the day people use electricity, it's really, really relevant because electricity costs vary radically. So we are now in a dynamic where the late afternoon and early evening hours and morning hours are very valuable from an energy procurement point, it's expensive electricity. And when we look at the daytime electricity, now it's often close to zero value or even negative values. And nighttime, in many places where there's a lot of wind resources, nighttime prices have approached zero or sometimes be negative.

So when we look at the economic impact combining with the environmental impact, you end up identifying that certain hours are really, really valuable, and you want to go for both maximum climate impact and maximum impact, focus on those. And there, you find that the houses that are valuable could be sometimes very small because they use really significant amounts of power regardless of their size, but often they're also big houses. If you want to have equity impact, you want to specifically target households that meet certain criteria. But again, the solution there is focus more on marketing and trying to deliver the outcomes they're looking for because every segment that you're pursuing has more valuable targets and less valuable targets, and going after the most valuable targets is what we try to do.

Mike Toffel:

I mean this whole thing is complex because there's variability on demand, that is people over the course of a day use different amounts of energy to heat their homes and to power their homes. And now, there's even more variability in supply with renewables where the wind blows sometimes and it doesn't other, suns up sometimes, sometimes it's not, where it's different from the olden days when you had nuclear power that was 24/7 or fossil fuel power, which was relatively consistent in cost, whether you run it in the day or the night.

Pasi Miettinen:

Right.

Mike Toffel:

So it seems like all this additional complexity has entered the space with the renewable energy capacity growing in proportion to the whole grid.

Pasi Miettinen:

And not only has it entered, but it's going to continue evolving over time. So the things that are valuable today may not be valuable to or three years from now. So there's a need for continuous monitoring of what delivers the outcomes. So for example, on the electric vehicle load management side, we really want to make sure that those cars are not charging between, say, 5:00 PM and 9:00 PM. Those are the summertime hours that are really taxing the grid.

Fortunately, that's the easiest load to shift over. There's no penalty. I've been driving an EV for 11 years, there's no penalty in charging that car at nighttime. It's ready in the morning when I'm ready to go. So moving that charging to off-peak hours is very, very straightforward. Heating and cooling, not so straightforward. People like to be warm when it's cold outside and vice versa. So turning their heat off or air conditioning off when it's an extremely peak day is not very popular among those customers.

So it really matters what kind of load you go after shifting it to off-peak hours. Not all of it is the same, not all of it is equivalently shiftable.

Mike Toffel:

So probably the least time shifting is temperature, and maybe the most time shifting would be something like if you need four hours of plug-in charge for your car and you have 10 hours to do it, you're pretty indifferent as to which particular four hours you run it, or if you run it for eight hours at half charge, for example. And somewhere in between would be dishwashers and washing machines. If you're running it overnight, you don't really care when exactly overnight, but you'd like it to be done by 7:00 AM, so you can dry it or something.

Pasi Miettinen:

Indeed. But it also varies depending on what part of the country you're in. So for example, when we look at Florida utilities, and quite a few utilities, even though Florida has the second most EVs in the country after California. What's interesting is stuff in many of the utilities, pool pumps and pool heaters can be a more significant load than electric vehicles today. That will change, in the not so distant future that is going to change. But when we look at where is the value, what kinds of things can you do to have maximum value, understanding what types of end uses are causing it is really critical. And we discovered that from the utility meter data analytics. Our software algorithmically using our AI algorithms goes through and figures out, "Oh, this house has this much heating load, this much EV load, what part of it can be shifted?" So we classify that load usually into things that can be rescheduled and things that are really hard to reschedule.

So we first go after those loads that can be rescheduled. I mean, electric vehicles are one, pool pumps in Florida. And there are a whole bunch of other things that are relatively straightforward to load manage. But when you actually look at it, say, on those very coldest days or if you look at the future, what will electrify future look like when we have a really significant amount of heat, electric heating? Well, a lot of the country's already there. We have, I mean, 20 million all electric households in the southern half of the US already, and some of those places get very cold. Kentucky, it can get colder than in the Massachusetts, or at least in the Boston area. So when we look at those opportunities, there's a lot of insights and knowledge that can be brought back to whatever the home territory because somebody else has already experienced in the country, and AMI meter data is super, super valuable for telling us that.

Mike Toffel:

So now we have a great foundation on the variability of demand and the supply of electricity. Let's talk about what Sagewell does because you talked about we could go after these rescheduling opportunities, like for charging your electric vehicle. What does that entail in practice? And how is Sagewell bringing those insights and delivering value to its customers? And who are its customers?

Pasi Miettinen:

Yes. So our primary customers are electric utilities. We also do some work for gas utilities. We also work with some of the investment arms of utilities as they're trying to think, "Where do we deploy resources for the next 10 years or 20 years?" Because we have a lot of insights from the meter data on that. But as a company, what we do is identify these opportunities that are either maximum climate benefits, for example, through electrification, or we also do electrification programs and load management as I mentioned.

But the primary driver is all of our AI modeling, all of our analytics as a core part of the business that then drives everything else. And there's a other theme that's here, which is if we are working on something and it works, let's do more of it. If it doesn't work, pivot. And a big part of it is that our business pivoted seven or eight years ago significantly from energy efficiency towards electrification because we saw that to be more impactful. So our customers at electric utilities, we run the electrification programs. In some way, we deliver the problem because we get people to buy electric vehicles and then heat pumps, but then we also manage that load, making sure that the grid can handle it.

Mike Toffel:

Got it. So an electric utility with the insight, for example, that, "Oh, these households have..." And they can tell from your analysis of their data.

Through your algorithms, your machine learning algorithms, that you can help them identify who has electric vehicles, who has heat pumps based on the energy signature of their demand.

Pasi Miettinen:

Precisely.

Mike Toffel:

And then what do they do with that? Do they send out mailers or email notices to their customers saying things like, "Hey, did you know you'd benefit by plugging in your car later in the day," or how does that work?

Pasi Miettinen:

Yeah, so it's either direct mail or email. Email is very common. All these programs are open to everyone. Utilities want to make sure that everyone has access to them. But when it comes to target marketing these programs, we're specifically looking at the ones that have the biggest impact. And one interesting nuance is that, so there's an investor-owned utility perspective versus a non-regulated utility perspective. I mean, we have some 20-something percent of the customers in the US, they get their electric service through a non-regulated utility or differently regulated utility than investor-owned utilities. And so the nuance of this is that there aren't enough subsidy dollars to deliver these outcomes because you can't spend enough money on them to deliver, get everyone electrified, you can't afford to do that. So we are often looking at what is the margin impact of these technologies, the utilities, especially on the non-regulated side, to basically make sure that the economic interest of the utility is aligned with the climate interest?

So making sure the specific types of heat pumps get installed, not the least efficient heat pumps, as well as making sure that the EVs that matter, the grid-impacting EVs that they're charging during off-peak hours. And again, what matters there is that only about a third of the EVs out there are creating problems. Turns out that two-thirds of them are actually pretty good for the grid, and they generate very healthy margins.

The problem is that one-third of them, again, can cost... I mean, a couple years back in Massachusetts, we had very expensive peak power. Capacity costs were so high that a Tesla charging on peak could cost $2,000 dollars for the utility for that one hour.

Mike Toffel:

Wow.

Pasi Miettinen:

The only problem with it was that the entire household might deliver $500 dollars of contribution margin in a year. So basically your entire margin for the year is gone, four times over, in one hour. So when we are looking for those opportunities, we're specifically looking for those cars that are contributing to the peak the worst or that have the worst peak contribution, and then being able to focus on the ones that we can bring them so that we can also get downward pressure on electric rates, and deliver the economic benefits of electrification and climate impacts.

Mike Toffel:

So when you say a third of the electric vehicles are the problem, you're not talking about the models or the people, you're talking about the charging behavior?

Pasi Miettinen:

Correct? Yes, the cars are fine. So the way I would describe it is that, if you think of it as a two-by-two matrix, like any consultant would say, what is the probability of a car to be on peak and on the x-axis? And then if you look at, what is the charging rate on the Y-axis? We tend to focus on the top right quadrant that basically says if you are charging frequently at a high rate and you happen to be on peak, yes, we are very interested in you. That car is a primary target for us.

But if you charge at work during the daytime where there's no shortage of power or if you charge at a very low level, just plugging into a regular 120 volt outlet, those cars are not primary targets for peak reduction. They are not contributing to the problem; they're actually generating very healthy margins to the utility. And it's actually very, very good business for the utility. It's those third of the cars that are actually potentially money losers, in some cases, significant money losers.

Mike Toffel:

And your ability to detect at the customer level, the variation in their demand is because of the advent of smart meters?

Pasi Miettinen:

Well, that's the interesting bit. We have about 120 million of these meters that provide at least hourly interval data, so they can at least tell us what the consumption was in any given hour. Some of them produce 50 minute interval data, some even down to five minutes, and now increasing even shorter. But fortunately for us, all the loads that matter from a utilities point of view, either for their business or their distribution system, all the large loads can be detected from the hourly meter data. So that means that the market size from an analytical point of view is about 120 million households in the US and businesses.

Mike Toffel:

And right now, the tool that you're describing to shape people's behavior is information.

Pasi Miettinen:

Correct.

Mike Toffel:

And how far away are we, do you think, from having the ability to actually differentially charge or through pricing or incentive programs, to encourage people to actually change their behavior, but not just because they now know and want to be a good citizen?

Pasi Miettinen:

We are there already, and what's interesting about that is that we started... So take one step back. A lot of people in the utility business or even electronics or even startup environments associate complexity with sophistication. And what we found is that a lot of utility programs are actually remarkably complex, and they are sometimes difficult for customers to navigate. We were guilty of some of that ourselves. So when we started doing EV load management programs six plus years ago, we used smart chargers that you can turn off remotely during high peak periods. Some customers were okay with that, but they were not necessarily thrilled with it.

Mike Toffel:

So they had to opt into this program? Which basically said, "I'm agreeing for you to cut my EV charging in times of peak loads." Is that what they're opting into?

Pasi Miettinen:

That is correct. And basically what happens is that they had to make that active choice to join those programs, but when you looked at the market share, those smart chargers, they were less than 10% percent of the market. So even if we had gotten a 100% percent of them, we wouldn't have had an impact. So we were like, "Okay, what else can we do?" And that's how we ended up identifying what we, at that point, called a Tesla problem. So Tesla drivers really liked buying their remote controllable chargers because all the smarts are in the car. The charger itself is just delivering electricity, there's no electronics beyond just delivering the electricity. But what we discovered in the process was that "Why don't we just ask the drivers to schedule their cars to charge during off-peak hours?" Nighttime mostly, sometimes during the daytime as well.

And then we went and said, "Let's use our algorithms to detect when they're charging." And in the process, what we discovered was that it was really simple for the customer. It took them less than 10 minutes to basically do the settings in the dashboard to get their car to charge during nighttime and the enrollment process. So what happened was all of a sudden, we had a program that became significantly more popular than our smart charger program and we're like, "Hang on. If this is working so well, why don't we migrate all of the customers over?"

And we did that, and this was now, six years ago. In the process now, we can turn off that charging in your car as well, remotely. But again, you have to participate.

But one of the challenges that we have found is that things that involve getting people to accept somebody else to control their lives can be really challenging. This is also true of thermostat remote control programs. If you turn someone's heat off during heat wave or when it's really cold, you're not very popular.

So we basically figured out that, for example, in TeleMAT and Smart Chargers, there was less than the 10% market share that could be captured. That's the remote control ability to turn off charging in your car.

What we've discovered in the process was that customers’ preferences really, really matter. And if we want to have impact, we better take those into account.

So, our AMI-based program is routinely enrolling 20 to 50% percent of the customers that have grid impacts, and we are able to deliver the outcomes using the AMI technology, automated metering infrastructure.

Mike Toffel:

So let's talk a little more about the AI element of the work of Sagewell. So you're getting, as I understand it, this data feed from the utilities that are your customers, which allows you to see hourly or more frequent data from, adjustable market now, 120 million customers out there. And you're trying to then turn back and say to the utility, "Hey, we think these are the high priority places to target our efforts to try and get them to load shift."

Pasi Miettinen:

Yes.

Mike Toffel:

What's the AI ML story that allows you to go from these data feeds to those insights?

Pasi Miettinen:

Yes, that is an interesting part, and in some ways it's no news at all. And what I mean by that is that when I originally started as a load forecaster, literally 30 years ago this summer, we were doing neural networks back then, all the models and the basic concepts happened around forever. So in terms of the analytics, what has particularly changed in the last few years is it has become a lot more accessible to more people, you didn't have to have heavy computing infrastructure. We used to have heavy computing structure; we used to be able to do a lot of things because we had access to those resources.

Now, you have broad access to those resources. What hasn't changed, though, is that while it has become really easily accessible, and literally anyone can have access to most AI models, what's interesting is that it has made the value of the data itself more important. It's no longer a competitive differentiator what models you necessarily have, the differentiator is, what can you extract out of the data, and how good is the data quality? So before we even get to the actual analytics part of it in using AI or any other machine learning or any other methodology, we actually have to spend an enormous amount of time making sure that the data is good quality.

And it's a historical artifact because people used to not look at hourly meter data. It was good enough that you build things on a monthly basis. As long as the monthly totals were correct, you didn't necessarily worry about, was that electricity consumed at 2:00 or 5:00? Well, for us, it matters a lot. And what you'll find is a lot of the systems were configured in such a way that they didn't worry about the hourly accuracy of it. So we spend a lot of time worrying about data quality. Are we analyzing the things that we're doing? Because you end up introducing a lot of artifacts. So the key part here, before we even get to the, how does one actually model things or what models do you use to do it? Is you have to make sure that your data is of good high quality, and we spent a lot of time on that analysis.

Mike Toffel:

So in my world of data analysis, we call this pre-processing the data, and things we're looking for are outliers like, "Oh, 1000 times spike, that's probably a data error somewhere." Or maybe data sparseness. Like, "Huh, that house all of a sudden doesn't have any data for two weeks. That seems unlikely." Are those the types of animals that you're looking for? Are there other types that you're looking for?

Pasi Miettinen:

Those are very common. There are probably 30 or so items on our checklist. I mean, a really common thing was that people didn't worry about time zones too much, or they didn't worry about daylight savings time. We just adjusted them just recently. So you will end up being shifted by an hour or two or three hours, and that's the most basic, a very, very common error out there. And as a result, if you just apply the models directly, you will draw really, really incorrect conclusions because the peak was three hours ago, not when you actually thought your model was predicting the peak to be.

So there are a lot of examples like that, that there are a lot of technical details in terms of how these can be set up incorrectly. Fortunately for us, most of them we can correct later. Like time zones, we know how to correct that. It's rare that we actually find fundamentally incorrect data that can't be, but it has occurred as well, and then we have to work hard with the local utility to do it. But the good news is that your billing is not incorrect, your billing is fine for most utilities. It's just some of these challenges that occur.

Mike Toffel:

I want to understand the difference between analyzing the user's experience because you can see over time how much they're using. And from that, you're gaining insights, like you're able to say, "Based on that signature, we think this customer all of a sudden now has heat pumps or now has electric vehicles." So tell us a little bit about that. And then the second part of that is I think you're getting into prediction.

Pasi Miettinen:

Correct.

Mike Toffel:

And so help us understand, where does prediction come into play here?

Pasi Miettinen:

Yes, there are actually a whole host of derivative consequences of all that. One is being able to identify these end users and being able to say, "Okay, there's a savings opportunity for this customer. They could really have a significant economic benefit to themself if they switch to some other technology." So there are customer benefits that we identify, there are utility benefits that we identify. But as we go through it and look for those opportunities, the next level is actually a lot harder. The part about identifying things reliably is, fortunately for us, the bigger the impact a customer has on a utility, the easier it is to detect.

If you use a lot of electricity, you are easier to identify as that particular end use. If you use very low electricity, we are not so concerned that our predictions might be wrong because you're probably not going to be a target for any of these, at least energy savings programs. For other programs, yes. But what we've learned over the years is that you identify two identical houses that have exactly the same challenges or is opportunities for electrification, or what are the measure that utility is promoting to that customer? What becomes interesting is only one of them actually acts on it.

And the other one doesn't. So then, there are two aspects of prediction. One is, can we predict which customer is more likely to do it? Because then, when we look at the customer lifetime value, we are looking at the cost of conversion and we're looking at how easy it is for us to deliver the outcome. Where do we deploy the resources? Can we predict who is most likely to act? Then the second part of that is that not only does that define who is most likely going to take the action that we're hoping that they will, is also then impact on load forecasts, and that basically defines which areas are more likely to adopt electric vehicles first or heat pumps, or are these buildings really well suited for heat pumps whereas those other buildings are not? So it's not even a customer preference thing, it's just literally those buildings are not good targets today.

But there's a second compounding factor, which is, that people sometimes, their consumption tends to cluster. So people who are not only in their house but amongst their neighbors. So the people who are more likely have a solar, but also more likely to have an electric vehicle and a heat pump. So not only is there load concentrating, so we can probabilistically predict, if you have one in use, you'll more likely to have another. But then we also know that your neighbors are more likely to do it because... All kinds of cascading events and net promoter score, for example, is a very good way of predicting which of these technologies actually are going to get adopted.

Mike Toffel:

Interesting. So there's some contagion modeling going on.

Pasi Miettinen:

Very much so, yeah.

Mike Toffel:

Yeah. Now, do you have access to demographics of who's living in these households? Because I imagine that might influence the question of the responsiveness to these programs. Is it a family, is it a empty nester? Is it single? Is it even a rental property versus owner occupied?

Pasi Miettinen:

So that's a really good question, and the answer is it depends. In some cases, these customers are completely anonymous to us. We may not know, the utility has not given us any information on these customers, we only know their consumption pattern, and we are using that in this case. If we have demographic data that is attached, and some of it's still anonymous to us, we may not still know, but we now know that the building characteristics are a certain type and the household characteristics are these, then we can further segment them and use our propensity scoring models to, again, figure out who's more likely to adopt these things based on previous adoption rates.

Mike Toffel:

Got it, got it. Super interesting, yeah.

Pasi Miettinen:

But these things change. I mean, what's fascinating to me is that we had fantastically well-working heat pump prediction models years back, like four or five years ago. And it was essentially predicting that if you had oil heat, during high oil prices, you were very likely to convert. Well, when prices dropped, they were a lot less interesting, customers were a lot less interested in converting.

So there's a time element to it, there is a relative value element, there is a contagion element, there is proximity to other neighbors who are doing things. All kinds of things influence the outcomes, and this is where AI becomes fascinating to me because we used to do all this modeling by hand, essentially. And when I said hand, we had still sophisticated models, but we were looking for specific relationships that we could explain. Now, particularly in the context of marketing, we are now very much used to being able to take data quickly and adjust the models as dynamics change, and we don't necessarily know why. The models are looking for relationships that we as humans can't. And they will say, "These are now the target houses, that these are different than last year."

Mike Toffel:

Right. So this is the concept of explainability in these models, which in some industries, in particular, when you're dealing with government regulators, they want to know, "Why are you suggesting we should target these entities versus others?" And so then the movement toward explainability sometimes comes at the cost of accuracy.

Pasi Miettinen:

Correct.

Mike Toffel:

But in your industry, it sounds like in your segment, explainability is not a big deal.

Pasi Miettinen:

It varies depending on the context. For marketing, increasingly we are doing things that we can't explain the relationship. The model is a black box to us. It just knows that, and if we have a way of iterating quickly and adjusting marketing messages, we're able to act on it. But in other contexts where there are particular focus, for example, there is a desire to make sure that people are equitably participating in programs. It used to be that people would say, "You have to make sure that you don't target anyone."

But when we actually looked at those outcomes, it turned out that those programs were incredibly good at targeting people who were not good candidates for anything. So by saying we don't want to target, they actually were targeting by basically getting the... A lot of resources were spent on trying to get very few people to participate.

Mike Toffel:

Interesting.

Pasi Miettinen:

So we are now spending more time in equity. When we look at equity, for example, or from utilities, they're very focused on equitable delivery of program resources. We are actually doing more targeting in those segments to make sure that they are represented, that we get the participation from the best customers in those segments. So times have changed, but the themes stay the same and the models are completely different.

Mike Toffel:

Interesting. Now, you've mentioned neural net a few times. I want to make sure that the listeners who are unfamiliar with machine learning technology and jargon have some sense of that. Let's assume people understand regression, maybe, they have one Y and a bunch of X's. Take us from predicting one dependent variable with a bunch of independent variables. What's the leap to get to neural net?

Pasi Miettinen:

The leap is that the model makes those conclusions, the model decisions for you.

Mike Toffel:

It figures out which axes.

Pasi Miettinen:

Yeah, exactly. So for us, it really is the black box. Actually, to me, it doesn't matter if it's a neural net or any other model. We are increasingly dealing with black box models. They are making all the decisions for us, in terms of saying, "This is important, variable. This is not an important variable." But we don't necessarily know which they were.

Mike Toffel:

Right. And the interactions between those variables as well.

Pasi Miettinen:

Interaction, exactly. And the fact that it changes, and it's continuously learning, and this is the key part. As we get more data from program participation, we feed it back in, we get new insights and new conclusions, and that's why none of these things stay the same, they evolve over time as long as we can deliver the outcomes. So the key part here is that we're not doing this because we enjoy models. In fact, it's a tool to deliver an outcome.

And that's why all of our models have changed over the years, and they continue to change. And one of the interesting things for us right now is investment. We had a theory about applying AI to a particular customer relationship opportunity or how to serve customers better, but we couldn't make the models work last summer. Well, now the models, Open AI, large language models, those tools have evolved radically in the last year. Things that we couldn't do last year are potentially now within our reach, and we're experimenting with those concepts again.

Mike Toffel:

Now, who are your competitors in this space? Are there competitors in this area?

Pasi Miettinen:

There are, and there are, companies that do utility data analytics. And there are companies that run programs, they're called program administrators. They're often different companies. And then there are specific technologies like electric vehicle load management programs, they can be done with the smart chargers or it can be done with telematics. When it comes to the AMI base approaches, we're certainly the largest provider. It's the smart meter database approach. We are the largest provider of that in the US, but when you have big companies, like Oracle, has a significant data analytics operation.

There is a company that offers data analytics. And we are just trying to take it to the next level, which is delivering the outcome. Because of the challenge that we saw out there, there's a big gap in terms of identifying an opportunity and then actually getting someone to purchase or install some in their home. And there's a large gap that still exists, even if there are very, very good analytics companies delivering, the outcome is difficult. And there are very, very good delivery companies, like program administrators for utilities, and we partner with some of those companies where we might deliver the analytics piece and they deliver the boots on the ground, the customer service, and those kinds of initiatives. So the whole host of companies in this space, but we're one of the few companies that takes it from customer analytics all the way to delivery of the outcome.

Mike Toffel:

Got it. So it's more of an integrated producer.

So let's take a moment to look backwards and then we'll switch the gears and talk about looking forward. So as you look back in the decade or so that Sagewell has been around, what are some of the key challenges you've had to overcome?

Pasi Miettinen:

Yeah, it's a really great question because they have changed over the years, but there's one really common theme. When you're doing data analytics, when there are others, that most others are not doing the analytics. So we lack a common language around outcomes. And what that means, in practical terms, is that a lot of dollars are spent on climate initiatives, on energy efficiency initiatives, peak reduction initiatives, that are geared towards a narrowly defined technology, saying, "Can you weatherize this many homes?" And the goal is, how many homes can you weatherize? It is not how many kilowatt-hours or BTUs did you save.

Mike Toffel:

Right, more of a process metric.

Pasi Miettinen:

Process metrics rather than outcome metrics.

Mike Toffel:

Right.

Pasi Miettinen:

And when we started looking at the outcomes, it became challenging sometimes to say, "Hey, we are meeting all those process metrics." Yes, you are ranked number one in the country in terms of having the most successful process, but if we want to measure outcomes, there might be different things that we do, and a lot of these things have inertia. So finding the places where the data and insights can be deployed and where the incentives are aligned to maybe change directions sometimes is challenging.

And for us, what that has meant is that time has had to go by before the market caught up to the inevitable things that we saw in the data set. So there is a challenge that comes from, we know where the puck will be, but boy is it takes a long time sometimes for the puck to get there.

Mike Toffel:

Yeah, yeah.

Pasi Miettinen:

That's a big, big challenge in that business. And from a climate point of view, in many cases, we are spending years on things that we can already tell instantly from the meter data that that is not probably the best place to deploy resources. And we see the positive side of it, which is, there are these places that are fantastic places to deploy resources, and let's do more of that, but the mechanisms that are in place just aren't very fast to make those changes.

Mike Toffel:

Got it. And now let's look forward.

Pasi Miettinen:

Yes.

Mike Toffel:

What are some of the opportunities that you see going forward for Sagewell? Are you looking to hone the product line that you have now and just find more utilities to work with, to try and reach that currently 120 million total addressable market, and that I'm sure is growing? Or are you diversifying to offer a wider suite of services? Where are you seeing yourself go in the next five, 10 years?

Pasi Miettinen:

Yeah, so we have a unique offering that relies on the data from the utilities. So as long as the data is within the control of the utilities, our primary target market continues to be a utility market. However, there are 120 million meters in the US. Canada has some additional numbers; Europe has 200 million plus meters already installed. The rest of the world is about 400 million meters and growing. So the universe of the analysis opportunities is actually pretty significant, it's not just limited to US. And for example, I just spent some time in Norway recently and looking at the opportunities there, and one interesting nuance, as a sidetrack here, is that all the residential customers have to, their electric prices are market based. So literally the day before, you get the next day's electric prices. There are some caps and limits in terms of how badly you can be impacted if the price goes sky-high. But you have an entire ecosystem built around people, in theory, responding to prices.

Mike Toffel:

Yep.

Pasi Miettinen:

Which creates entirely new problems, in terms of load concentration and uncertainty about your energy bills. But when we look at these kinds of opportunities globally, each region has its own challenges, unique things. Norway has, over 90% percent of new cars are electric, and roughly about half the households have an electric heat pump. So the point of this is that we already know what happens to these distribution systems and the electric utilities when certain parts of the world are electrified. And so the second part of it is not just the analysis results, but also bringing some of these insights back. So Massachusetts, for example, in Massachusetts, we like to say we are the first to do something and we are always doing the leading and stuff. And a lot of the discussion right now in electrification and decarbonization of the grid assumes that we are the first to do it, but turns out that we're not, we have a lot of insight that can be brought from elsewhere in the country and the world.

So we have found doing more advisory work as well. It was not intended to be that way, but we thought we were going to be in software business, but we're doing a lot more advisory work saying, "Here's what we know is going to happen in the next 10 years because the trend is already clear in the data. If we get this many EVs, we know the outcomes already." So those are the kinds of things, so doing more both domestically and globally on the analytics side of it.

But we also have had a lot of inquiries from trades, for example, saying, "If these trends continue, what does it mean to have we need...? We will need X number X of transformers. We need thousands of miles; we need very large numbers of cable. What does it imply for batteries? What does it imply for virtual power plants? " Both the business and the distribution system implications. So we are now doing a little bit more predictive analytics.

Mike Toffel:

Got it. Yeah, you mentioned day ahead pricing. I was going to ask that. It seems like right now, in the US context that you're working in, you're basically trying to convince people to load shift the timing of their demand dealing with Q on the price and quantity that in an Econ 101 class would talk about.

Pasi Miettinen:

Right.

Mike Toffel:

And it seems like price has been mostly missing from the story in the US context about at the consumer level trying to adjust prices either by day or even within day.

Pasi Miettinen:

Yes.

Mike Toffel:

And I imagine that once that arrives, your models are well positioned to have to figure out, again, a new level of responsiveness because some customers who might not have been responsive to just information nudges might be very responsive to price changes, and vice versa.

Pasi Miettinen:

Yes, and that's actually another one of those areas that has surprised us a little bit. One is yes, well, the punchline in the story is that there's a segment for everything. Every technology or an approach will have a customer who responds to it. The same applies to time of use rates or rates that are changed throughout the day. Some customers are very price sensitive and others are not at all. And one of the fascinating things to us is that we have a large utility in the Midwest that has a very healthy nighttime discount on their electricity rate, but relatively low take up on it, even though customers in general would save money.

And we have done a lot of the analysis in terms of how to help with that. We actually are doing certain things there to help the adoption of that. But there are two things that we are seeing. Not only are voluntarily people not switching over to those as much as we anticipated potentially, but even when they do, they aren't necessarily changing their behaviors. Some of them are still continuing to charge their cars during peak hours. But one of the simple explanations for that is that, for example, if their nighttime electricity rate is, say, 10 cents, and their daytime or peak hour electric rate is 20 cents, that seems like a big discount, half off, literally 50% off.

However, the equivalent price of that 20 cents a kilowatt-hour is $2 a gallon.

Mike Toffel:

Yeah, interesting. All right, well the last question that I would want to ask you here is what advice you have for folks who are interested in working at the intersection of AI and climate or more nearly in the space that you've found yourself, which is data analytics and energy demand analysis?

Pasi Miettinen:

Yeah. So there are a couple things, and even going one step further back. I mean, I have two professors at HBS who were very influential. Two Bobs, actually. Bob Burden in terms of how option pricing works, and there's a lot of optionality right now being introduced to this ecosystem. Price volatility in particular, and Bob Kaplan who, activity-based costing was a very, very popular topic back then. We tend to focus on activity-based margins today at the utilities and try to twist on that concept, trying to figure out which activities are generating the highest margins. When we apply those kinds of lessons, the lessons that I took from classes and then looked at where the opportunities are today, the reason why those things are really relevant is that we're seeing increasing volatility in prices and supply of electricity as well as the consumption, as well as trying to figure out what are the economic implications of all these, which are radical.

So it's literally hundreds of billions of dollars’ worth of shareholder value will change hands based on these dynamics. Not only are investors looking at it, but if you're a new person coming into this business or looking at the ecosystem, it would be very healthy to look at it as an investor because the most valuable commodity or resource you have is your time, and trying to figure out how do you deploy that time is an investment decision. So just as you decide where you spend your time and where your career is going to be, it would be very helpful to look at whatever the intrinsic value is, it's really helpful to understand what is the return on investment that you will get on your personal time.

So that's one. I would certainly encourage people to listen to previous episodes of Climate Rising because there's a whole bunch of things in there that are useful advice in there in the past. So I was just trying to think about what has not been mentioned in the past yet.

So one thing that I would suggest is going to industry conferences because there are a lot of things that we hope to be true by the companies. Go to the industry conferences, talk to the competitors, interview people, and use LinkedIn very assertively. Most of us are happy to talk to people and give us advice. And I keep saying this that my advice is free and you get exactly what you pay for. But most people, particularly if they're HBS alums, they will respond to younger students seeking help and others as well.

Mike Toffel:

Yeah, I would broaden that to say folks are generally helpful to people who have some connection to them. Right?

Pasi Miettinen:

Absolutely.

Mike Toffel:

Undergrad or even a town they're from sometimes, or a sport they play, find that connection.

Pasi Miettinen:

Absolutely. And not everyone will respond to it, but I would put those feelers out and just talk to people. Sooner or later, you will find somebody who will tell you about their experience, and that experience hearing is valuable. And the last item that I would say, we have an internal saying when we look at candidates and the question is, "Can they read?" And what we mean by that is that do we have evidence that they are really interested in something, they're intellectually curious, that they will go and read and learn something? Because all the things that we knew 15 years ago really aren't applicable today. Things that we knew 10 years ago aren't necessarily applicable. Knowledge expires and experiences that we have are helpful, but our ability to learn new things is critical. And we look for that. Can they read? There's a universal catch-all phrase that means, basically, are they able to adapt and are they able to learn something new? And those are the kinds of things that we look for in people we hire.

Mike Toffel:

Interesting. Really, really helpful advice. Well, Pasi, thank you so much for joining us here on Climate Rising.

Pasi Miettinen:

Oh, my pleasure. Thanks for having me.