Season 1 Episode 7: Samuel Young on Getting the Most Out of AI for Decarbonisation

This week on the Hypercube podcast, we are joined by Samuel Young, Practice Manager for Data Science and Artificial Intelligence at Energy Systems Catapult.

Working as part of the Artificial Intelligence for Decarbonisation’s Virtual Centre of Excellence (ADViCE) program, Sam shares his perspective on getting the most out of AI for Decarbonisation. From how to approach the technology to get the best results to the cultural shifts needed for AI to have an impact, Sam gives his take on everything AI and decarbonisation related.

S1E07 Samuel Young

In this episode, we covered:

  • The Artificial Intelligence for Decarbonisation's Virtual Centre of Excellence (ADViCE) program.
  • Avoiding the ‘efficiency trap’ when identifying AI opportunities.
  • Using AI and data science to improve heat pump adoption rates in the UK.
  • Accelerating decarbonisation by improving grid network connection times for renewables.
  • How future breakthroughs in time series forecasting methods could significantly help problems in the energy sector.

The weekly Hypercube podcast sits down with leaders in the energy and utilities sectors to explore how data analytics can help businesses make smarter decisions and accelerate business growth.


[00:48] Sam gives a quick overview at the Catapult’s and talks about the AI for Decarbonisation’s Virtual Centre of Excellence (ADViCE).

[4:41] Host Adam asks what to look for in an AI application to accelerate decarbonisation.

[6:52] Sam gives his view on using AI and data science to drive the decarbonisation of UK homes, particularly when it comes to encouraging heat pump adoption.

[11:49] Sam discusses the flexibility needed to ensure the energy network can cope with high heat pump adoption.

[13:40] Sam highlights the importance of getting more green, renewable assets live and connected to the grid.

[15:50] Sam highlights that organisational culture and mindset are essential for AI and data science adoption, not just individual skills.

[16:45] Sam illustrates the need for buy-in by sharing the ‘most successful and least successful data science project he’s ever worked on.’

[21:46] Sam gives an insight into how decarbonisation challenges impact the manufacturing sector.

[20:04] Adam and Sam discuss the need for a time series breakthrough in AI and the potential impact this will have on the energy sector.

[24:09] Sam runs through where listeners can go to find out more about the topics covered in today’s episode.


Hypercube Podcast Transcript

Title: Samuel Young on Getting the Most Out of AI for Decarbonisation

Host: Adam Sroka
Guest: Samuel Young

Intro: Welcome to the Hypercube podcast, where we explore how companies in the energy and utility sector leverage data analytics to make smarter decisions and accelerate business growth. I’m Adam Sroka, founder of Hypercube, a strategic consultancy that supports asset owner-operators, traders, route-to-market providers, and energy services companies to unlock the power of data.

If you’re interested in hearing real-world examples of how data and AI are advancing the energy sector, this is the show for you. 

Adam Sroka:
Hello, and welcome back to the Hypercube podcast. Really excited today to be joined by Sam Young from the Energy Systems Catapult. Sam, I guess, always like to start off with a bit of an introduction. For those that don’t know you, and also a little bit about Energy Systems Catapult, could you tell a little bit more about what you’re up to? 

Samuel Young:Yeah, great. So I’m Sam. I lead the Data Science and AI practice at Energy Systems Catapult. And Energy Systems Catapult is an independent organization focused on accelerating innovation in the energy sector to help us achieve net zero.

And so we do a whole load of different things from kind of working with government to help state policy, working with SMEs and innovators. And so my role within that is all things data science and AI. So very much thinking about when I joined the one line job description I was given was Think about everything that’s happening in the AI world.

Think about everything that’s happening in the energy sector and work out how you can make those two things come together and like accelerate net zero. So that was quite a big job description when I started. 

Adam Sroka: Just a quick morning task then, shortlist. 

Samuel Young: Yeah. 

Adam Sroka: Easy. Yeah. Well, right. Okay. Yeah. So that is very interesting because for those that are unfamiliar, because we have some listeners overseas, like that’s huge.

Actually, it’s a huge remit of areas. Do you end up spending more time in one area than another? Or are you like team of teams level orchestrating endeavors that way? 

Samuel Young:A lot of what I’ve been looking at recently with the advice program, the AI for decarbonization center of excellence program is genuinely looking at the whole sector. So it is really broad. Historically, a lot of the work that me and my team have done has been a bit more focused on home decarbonization and some of the work we’ve done there at Catapult. But yeah, increasingly looking kind of really quite broad. Some of my history is with network companies. So thinking about kind of networks, um, but a big part of what Energy Systems Catapult tries to do is that whole systems perspective, because that is one of the interesting things about the energy system is.

It’s all interconnected and that kind of means if you just focus on one area, you end up missing some of the important features. 

Adam Sroka:  Really cool. So look, you touched on it. Big sort of exciting launch recently, the AI for decarbonisation. Can you expand on that a little bit more?  

Samuel Young: Yeah. So the department for energy security and net zero has been funding a program with some kind of funding for specific innovation projects, but also funding for this advice center of excellence where the goal is really to bring together organizations in harder to decarbonize sectors.

So things like energy, manufacturing, agriculture, and AI professionals in those sectors, but also outside of those sectors. to really accelerate innovation in that space to kind of bring in new funding, bring in new ideas and have catalyzed that. And so we, we kicked that off with a couple of reports working with Digital Catapult and Alan Turing Institute to look at, you Well, what is the kind of ecosystem around AI for decarbonization?

And then the edge system catapult, Steve and I kind of focused on, well, what are the challenges? What are the real decarbonization challenges? And did a report on that. What I’m hoping will be the most interesting technical appendix you’ve ever read, which is a series of kind of cards with one card per challenge, kind of illustrating that with the idea really being, you know, people had told us.

If you’re not in the energy sector, or if you’re not in manufacturing, you can get some like high level, vague understanding of, oh, you know, it’s difficult to connect renewables to the grid, but you don’t really know what the challenges are. And so you can’t really work out how you can apply AI to them.

So we were really focused on, well, let’s find the challenges, articulate them in a bit more detail than you normally get. But do that as broad as we can so that like, it’s not a, well, we’ve written a report on flexibility, but I report on as many different, um, AI for decarbonization as we can.  Hopefully, it’ll kind of like people pick it up and go, Oh, that looks interesting.

I want to go and talk to someone about that and kind of try and find someone and engage with that. Right. 

Adam Sroka: Do you want some raw feedback on the report from my first read? Very interesting. Yeah. So first of all, it’s brilliant. It’s quite, it’s a big, big report as well. It’s not a small document. It’s brilliant.

The thing that I really liked about it was covers a good breadth and a good depth of the stuff that you would naturally some stuff I hadn’t thought about but it was very accessible actually to both non energy experts and non data experts. So I think it’s a really good document to like, you know, get up to speed with, okay, this is what this space is thinking about and some of the technical challenges around it.

So yeah, kudos to you. Well done. I thought that was really good. To you though, and what were the kind of measurement criteria for what makes a good AI application that’s going to accelerate decarbonization? Like, what are you looking for? And how actually do you pick something and say, this has got the bones of something that’s going to be really impactful?

Samuel Young: Yeah. I think often people put it the wrong way around. So we focused on what is the decarbonization challenge and then wrapped our brains a bit to work out like, could AI help? 

Adam Sroka: Yeah. Okay. 

Samuel Young: Rather than often you get what I kind of call the efficiency trap where you have people go, Oh, AI is really good at this.

And they’ll look at something and go, Oh, well we could automate that. We could automate this. We could automate that with AI. And they just focus on like incrementally improving the efficiency of existing processes. But often with things like decarbonization, it’s not like incremental improvements to existing processes that are required.

It’s kind of wholesale changes to industry or like, Oh, we’re, we’re changing these two things and that’s causing a really knotty problem that we haven’t had to resolve. So that’s kind of the way that we kind of tried to look at it was we talked to lots of people across the sector, did lots of reading of like, what are the challenges?

Then do a survey of how is AI being used in them? But also, like, even if we’re like, well, that’s a challenge and we don’t really, there’s not loads of good work on AI in that sector, and we can’t really work out exactly how you’d use it. We still like, well, we should still talk about that challenge because someone who works in that area, they go, Oh, hang on.

I do something that’s very adjacent to that, but in a totally different sector. So that was kind of what we focused on. 

Adam Sroka: So you talked a little bit about, and I think a lot of people will know, ESC and the Living Lab and some of the work that’s been done around like heating and decarbonisation of homes and stuff like that.

And that is one of the big themes in the report. Do you have like a, a start for that? If we were to pick on one use case, like how do you think, what’s the biggest play in data and AI making an impact for decarbonization of the homes in the UK? 

Samuel Young: So I think this is a really good example because If I say, like, let’s pick on heat pumps, for example, they’re likely to be quite core to decarbonizing domestic heating.

If I say AI and heat pumps, how do you use AI for heat pumps? Most people think like control the heat pump,  right? That’s the thing that it’s like, well, obviously you’ve got, ah, well use reinforcement learning and that’ll be awesome. Yeah, sure. Go ahead with that. But the real blocker to heat pumps in the UK is not, we can’t get them to operate.

It’s an adoption challenge. And actually there’s a big kind of like feedback loop happening of people perceive them as expensive, unreliable. The installers aren’t kind of like really excited about them. They’re not kind of like pushing them and selling them. And so That kind of means that we don’t have rapid growth and competition in the sector and kind of improvements in quality and all of that kind of thing.

So actually coming at it from kind of like a systems, what are the systemic blockers? I mean, I go, well, the systemic block is heat pump adoption. Some of the factors influencing that are kind of like installer perceptions, installer skills gap, the length of time it takes to install a heat pump, all of that kind of thing.

So think about how can you apply AI. to accelerate that. 

Adam Sroka: There was some interesting stuff there on significant challenge, but like affecting consumer behavior, market demand, and is that not driven by the kind of financial overhead and some of the fear that people have about digging up gardens and all that around by actually getting the heat pump?

Is there anything we can do in that space, do you think? 

Samuel Young: So financial models are really interesting, right? So heat pumps are, do have a higher upfront cost than a boiler. And that’s a problem, but like, you know, lots of things in our lives have a high up road cost, like a house or a car. And we have financial models around those.

One of the interesting ones that Energy Systems Catapult has looked at and talked about quite a lot is heat as a service. Where rather than having a, you pay for the heat pump up front and you pay your energy bill too, you kind of, you package it all up and you pay for heats to be provided to your home and the kind of energy supply is responsible for how that happens.

And they choose to do it in a heat pump, low carbon way. But actually when you start to think about, well, what does that actually require? You find that there is quite a lot of data that starts to get in there and kind of AI. For example, I was thinking about it. This morning, if you charge people for heat in a different way from charging people for hot water, now you instantly have the problem of, well, you’ve got to separate out what’s heating and what’s hot water.

Maybe you’ve got that metered in already, maybe you don’t, but you also have to start to have things like, well, how do you forecast changes in behavior as kind of like price changes or like weather or all of that kind of stuff. Yeah. And so one of the other projects that we’re working on at the moment is really how do you use AI machine learning to measure in as much as you can measure energy savings after someone has installed insulation or a retrofit?

Because it sounds kind of easy and like, oh, we’re going to measure it. Like, sure. It’s been a 20 percent reduction compared to last year, right? But last year was a bit warmer and oh, after the retrofit, they’ve turned the thermostat up by a degree and it all gets really quite complicated. And the idea being like, if you can measure that, use AI to kind of like provide standardized quantified like measurements of things like that, then financial services providers can come in and go, okay, well we’ve got more confidence on it now, there’s no longer this kind of worry that people are going to turn their phone set up. We have some way to quantify. They have turned their phone set up and so we can design the final products around it. So there is that kind of thing of like, if you can use data collection AI to quantify things that gives the financial people kind of more confidence to develop new business models and build kind of new ways of serving consumer needs.

Yeah. Which I think is quite important. Yeah. Absolutely. 

Adam Sroka: The big challenge I think though is, I think it’s hinted at in the report as well, is with a lot of this stuff, like dynamic control of heat pumps, for example, it’s to say we get to like 80, 90 percent adoption across the UK, right? Can the grid handle that? Like, is the infrastructure there to support not only supplying that level of additional power, but also, To the homes, but having them hopefully intelligently move the demand around. But also, yeah, just provide the flexibility that it would need. 

Samuel Young:Yeah. And there’s a really interesting thought experiment here that I think really demonstrates the importance of looking at the whole system, which is we’ve got two different groups, right?

Of people with like slightly different incentives. So you have the heat pub designers and heating system installers. And they are trying to design your heating system so that it is just large enough to work out on the coldest day, whatever the coldest day means, which means that essentially it will be running 24/7 for that coldest day or 24 hours for that coldest day.

And then you have the kind of network flexibility people who are looking at heat pumps going like, Oh, it’s good because we can use them to flex things. And we don’t have to reinforce the network fully.  Actually, if you think about it. On the coldest day when demand is going to be highest and most heat pumps are going to be running at their capacity, there won’t be loads of flexibility.

If you’ve designed it right on one side, there won’t be enough flexibility. And so that kind of, that thinking around, actually, there’s lots of flexibility lots of the time, but you don’t have it when it matters. And so that’s where you need to think about the whole system. And then like, are you incentivizing  the right forms of heat pump design?

Like, how do you wrap all of those into one cost? because you tend to think about the network reinforcement costs when you’re a network.  So as things currently stand, you know, the, there’s increasing amounts of work trying to bring those two things together and work out how do you consider it. But the amount of investment that we need to reinforce the network for those coldest days, really extreme peak demand is still really large.

And we need to think about like, To some extent, one option is you just reinforce the whole network. That’s incredibly disruptive. Or you think about, Oh, well, what other forms of flexibility? How do we, how could we use technology as well to kind of help with that? 

Adam Sroka: So we’ve talked a lot about demand management, I suppose, like, can we flip a little bit to what are your thoughts on how data and AI can help with getting more green, more renewable assets live and on the grid and sort of build a greener energy system as a whole.

Samuel Young: So I think this is a really good, again, I’m just going to keep repeating myself, like, think about it like a system, find the systemic blockers. But when you think about getting renewables onto the grid, most people don’t realize that one of the biggest blockers is building the network connections. And the time that it takes to secure a network connection is now kind of, you know, 10, 14 years.

And so what is it that’s holding up renewable deployment? It’s not, it takes a long time to build a wind farm. It’s, it takes a long time to get the connection to the wind farm built.  And so that’s where there’s kind of like, okay, so if you’re going to use AI. to help renewables. You can use AI to improve solar panels, come up with new materials, so they’re more efficient.

Yes, good. You can use it to improve the design of the wind farms, so they’re more optimal. Yeah, all of those are good. But if you focus narrowly on what’s going to actually accelerate decarbonization,  One of the big things you should be focusing on is how do you accelerate the actual building of the connections. And so looking at that queue, there’s lots of work on this problem as a whole rather than like AI focused going on. But there’s also organizations that have focused on it from an AI perspective. So Continuum Industries have been using AI to accelerate the process of planning a route and kind of like taking into account the different terrain factors and different things that are along the route. So that you no longer have to have multi month studies by engineers kind of working through these parameters by hand. But if your first go can be automated, and then if you can have people collaborate with adding comments and it iterates all on this platform, that can really accelerate the cycle of planning process, which can then kind of shave months off the process of getting approval.

So I think that’s a good example of like, focus on the, uh, what can we do to. Accelerate connection applications. 

Adam Sroka: Do you think we’re ready from like a skills perspective in the industry to really? Do this at the place it needs to happen?

Samuel Young: I think that’s a really interesting question. I think for me there’s a skills intersection with culture and mindset question, right?

So ’cause you have like individual skills, but you also have organizational culture and capabilities. And I think actually. It’s not, it’s not that hard. We always moan about it, but it’s not that hard to find someone who has the skills or like train them up in the other half of the skills they’re missing, whether that’s AI or power systems.

But actually the organizations as a whole and the processes that they work in the culture, that can be harder to kind of change relatively quickly. So I think that’s a bit of the challenge that we face. It’s not skills gap, but there is a skills gap, but it’s the kind of like. Organizational culture change.

That’s actually really important to me. That’s like my favourite story is the most successful and least successful data science project I ever worked on where I was working in organization. And an engineer came to me with an Excel spreadsheet with information about some assets. And said, we think there are problems with this asset class.

So we’re going to replace them all as going to cost us tens of millions of pounds, but you’re a data scientist. Can you, I don’t know, like do some data science on it and tell us whether that’s what we need to do. And classic, you know, customer request of like, Wave your data science magic wand. I was like, sure, I’ll have a look.

Started spitting out, like, did you know this? It’s like, yes, we knew that. Did you know this? Yes, we knew that. Did you know this? And eventually I got to a, did you know that this set pays very differently from this set? He’s like, Ooh, no, I didn’t know that. And I’m like, ah, well, please some more charts.  Then he goes out and goes away and like, looks at the design drawings, goes actually.

Now that you point that out, these are like this, and they do some, like, electrical stimulations, they kind of go, yes, there’s a design flaw on these ones, but not on these ones. And I’m like, I am a data scientist superhero, I have, like, done what I’m aiming for, like, data has found something about the real world that the engineers didn’t know.

So we, you know, wrote a paper to say, don’t replace them all, only replace these ones. Took it to the senior engineer, who said, uh, but we thought there was a problem with all of them. And if one of them that we say is now okay explodes and kills someone, like, it’ll be my name on the line for that having happened.

So I need more evidence. And so I do the back of the envelope and I go, even if we looked at every of this asset that we have, it still wouldn’t give us like much more evidence. But the engineers, no, we need to connect a year’s worth more of data. So collect another year, go back. I’m still not happy putting my name on that from a safety perspective.

So we need to collect another year’s worth of data. And so as far as I know, tens of millions of pounds were spent replacing assets that might not necessarily need to be replaced because of a cultural thing and kind of like a, it wasn’t that he lacked the skills. To maybe a tiny bit, but it was, it was this cultural element as well that I think is kind of really key. 

Samuel Young: And that’s really hard because it’s like infosec is another domain that just could trample over any other requirement in that ultimately if it comes to safety, I mean, our money might have been spent that needed didn’t need to be spent, but that engineer was probably right. And yeah, feel for that.

Samuel Young: I’d argue because it, what really struck me in the weeds of that one was the, how.

You like to think as a data person that you’re rational, and most people like to think that they’re rational. And then you present a rational argument, and you find that it doesn’t change people’s minds. So I looked at it like, oh, well, if the safety risk if you’re saying this is a level of unacceptable safety risk, we need to replace all the assets on our network. 

It was a like people, classic decision making biases, but someone had anchored to something and the data didn’t overturn it. And it taught me lots as a practicing data scientist about like, don’t assume that the data will convince people. Like you have to understand their drivers and their motivations and their emotions to kind of think about how do I frame this in a way.

Once I’m convinced that the data kind of tells a certain story, how do I tell that story effectively? Because otherwise, you kind of like don’t necessarily get the change that the data is suggesting you have. 

Adam Sroka: That probably sums up about 80 percent of all of my blog posts I’ve ever written. Like is to the point where I often say, like, start at the other end. Don’t do any analysis. Don’t look at any data at all. Start with the people. And really push them to give you on paper or something in front of a group of peers, chiseled in stone, the conditions by which they will change their mind, then go away and do all the hard, like, all the techy bit, because you’ll find that, yeah, like, so many of us have been in that situation where it really doesn’t matter what’s, what comes out of the other end of this analysis. If it doesn’t agree with the decision that was made, we’re just going to sort of gloss over it. So I feel your pain. Yeah. 

Samuel Young: The way that I kind of. The mistake I think lots of data scientists make, particularly if you come from a scientific background. If you’re, one is you just throw yourself at the data and you have, you don’t like ask what are people trying to achieve?

But if you’re slightly more sophisticated, you go, I’ve learned I need to ask the customer, what are your hypotheses? Like, what do you want to know from this data? And they, you go away, you ask what you want to know. But the thing is that people really curious, they want to know lots of stuff. And as you say, it’s the like, what would change your decision?

That’s the question that you need to ask at the beginning. Because so often, otherwise you run down rabbit holes of interesting, but not actually the core of what would change my decision. 

Adam Sroka: Okay. So it’s been a really good conversation. Is there any other parts of the report you want to touch on before we wrap up?

Samuel Young: I guess the other interesting one heading into manufacturing is the, one of the big things when you think about manufacturing, how do you decode those manufacturing? You often get pushed in that, well, let’s make everything more efficient. That’s good and useful, but often like there’s a good cost incentive to make things efficient already.That’s not going to drive the systemic change where a big part of the challenge is the how do you decarbonize the inputs? And that may be like changing to a different fuel, but it may also be kind of changing some of the raw materials. And that’s a much bigger challenge because it often requires kind of changing of the equipment, changing of the processes.

Um, and so again, it’s not one where I’m like, and the AI answer is use product X to do this. But if you’re looking for kind of where should you be applying your kind of AI talents to help decarbonize manufacturing, don’t just focus on that. I’m going to do quality monitoring and I’m going to like find defects, but think about how, how do I get into that space, which is the.

How do we change the kind of the inputs and decarbonize inputs? Because that that’s actually the hard part of the process. 

Adam Sroka:  It’s not the step change outcomes. Well, if it’s achieved, it’s just, yeah. Okay. Two last questions for me. What’s your great big hope, any emerging technologies or things you’re really excited about that you think is going to make a big impact?

Samuel Young: I’ve got my, I feel like it’s a slightly optimistic hope, but I do feel like time series has been the forgotten child of AI. You know, we’ve had that big vision breakthrough. And we now have had the big language breakthrough, so I’m like, please can we have the time series breakthrough next? I mean, there’s early indications of some work in that direction, but it’s like, that’s, I think would be really powerful in the energy sector because a lot of what we do is time series.

So yeah, that’s my, it’s not that I’ve got my eye on something like, Oh, it’s going to happen. It’s going to happen. But that’s my, like begging all the researchers out there. Switch your focus to time series. Do that, please. 

Adam Sroka:  Yeah, I wholeheartedly agree. I think it’s one of those areas that it hasn’t seen like drastic, like revolutionary change in a long time.

Like the old methods are still the best methods and like good old-fashioned stats and maths is what’s going to pull you through most use cases. Well, before we wrap up, one thing I always like to ask is Is there anything you’d like to plug or promote whilst you’re here, or is there anywhere we can send the audience to come and find out more if they want to learn more about what you’re up to?

Samuel Young: Yeah, obviously, definitely have a look at the advice website and our report and the challenges. One thing that we’re going to be doing is running data science sharing circles. And my framing of these is kind of like, you know, peer mentoring for data scientists deliberately  For different organization types, an SME and large network and an academic kind of like all coming together.

My rule of thumb is if your job is to go to meetings, you’re not allowed. Myself may be accepted, but it’s the hands-on people doing the kind of data science. With the idea being, but yeah, the senior people get more opportunity to network and see what’s going on and see things. But when you’re like at the coalface, you don’t necessarily have that same opportunity to bump into people, chat about something that’s totally outside your domain.

And actually when you do, you find the adjacent, Oh, you know what? There’s kind of like a, Oh, in financial services, they use this time series forecasting package for. Oh, you’re working on EVs. I’ve got a friend who knows this data set that that’s really useful. So yeah, we’re kind of trying to spin up some of those again, like the advice website is really great.

But if you know people who you’re like, ah, they’d be really interested in some of those for like the energy sector or manufacturing or agriculture was trying to spin up some of those. So that’s the good way to get involved and to find other people who. might give you interesting ideas, whatever you’re working on.

Adam Sroka: Hopefully find some of our team there. That sounds right up our street. So very exciting. Well, look, thank you very much for today, Sam. It’s been a pleasure to talk to you. My pleasure. We’ll make sure we share the report and yeah, hopefully, we’ll get a chance to catch up again soon. 

Samuel Young: Great. Thanks very much, Adam. 

Outro: And that’s it for this episode of the Hypercube podcast. Thanks for tuning in today.

If you have any questions about the topics we covered, you can reach out to us on LinkedIn or check out our website at You can also join Beyond Energy, our Slack community of data leaders from the sector. There’s a link to sign up in the episode description. We’re just getting this show off the ground, so if you like today’s episode, please leave us a rating, review, or subscribe wherever you get your podcasts.

It all really helps. See you next time.