Participants: Malte, Benedetto, Henrik
Malte:
Last night Mia asked me what I was doing all evening. I told her I was just jamming, responding to support tickets on Ento. At one point, someone was even trying to order their windows through our support. Not sure what happened there.
Okay, great, we’re recording.
Just a quick note on what we’re doing: we had this idea that we keep saying so many brilliant things in our weekly rants that we should start recording them. So here we are, recording one of our typical weekly rants.
It’s an organized mud fight. Or maybe a fist fight.
Apparently, Benedetto got four hours of sleep and I’m on a similar level, so this is going to be great – just running on coffee.
Henrik:
Alright, let’s bring a topic. I guess I’m bringing it this time.
We’re not prepared at all, but there’s been a lot of talk this week about submeters. We’ve had some rants on the internal chats, so why not elaborate on that here?
Malte:
I’ve had rants at customers and prospects who have invested millions of Danish kroner – hundreds of thousands of euros – in installing submeters. So some of my rants have definitely been directed at them, not just internally.
I think it’s a topic that just keeps coming up. As an example, we now have 35 Danish municipalities as customers out of 98. Every time we get a new customer, a key topic is:
“What about submeters? I need a submeter on hot water usage in the building,”
which is part of the district heating system.
And we’re like: no, you don’t. If you take the outside temperature and normalize your consumption against it, you actually get a very good understanding of when you’re using heating to heat rooms and when you’re using heating for something else – which can pretty much only be water.
This keeps coming up, and every time I explain it, people are very skeptical. They go through a process and then at some point they say, “I don’t actually need submeters.”
After managing this process for five years with so many clients, I’m wondering: how can we have this discussion once and for all?
Heating is only one example. On electricity, you often have cooling and thousands of submeters. You end up drowning in data rather than getting value from it.
Henrik:
Exactly. This topic is basically rants all over.
To be fair, maybe we should also talk about where submeters do make sense. I know, Benedetto, you’ve thought quite a bit about how to see it from the other side, so let’s make sure we get that in.
Benedetto:
Steel-manning, I love it.
If we think about certifications like ISO 50001 and similar frameworks, people often say they “mandate” submetering.
They don’t.
They mandate that you estimate what the individual sub-components of your building consume.
Until recently, there wasn’t a good way to do that using only main meter data. There wasn’t enough granularity, and there weren’t smart AI algorithms that could extract that information. So historically, this was mostly relevant for larger and more complex facilities.
That’s where submetering has often been implemented first: large or complex facilities, not the typical buildings we look at in Ento – fairly similar, large portfolios of buildings operating in a predictable way.
Also, a lot of traditional standards that organizations comply with – or choose voluntarily – like GRESB, focus heavily on submeters as necessary components for measuring things. So the demand for submeters can come from a compliance angle.
Up until a few years ago, using only main meter data wasn’t really considered an option. Often you didn’t have smart meters, or if you did, it was hard to access the data. And then you had no good way of extracting insights from it.
Malte:
A few notes here. I actually disagree with the idea that you ever needed submeters for ISO 50001.
You could implement ISO 50001-certified energy management even before smart meters were widely available – both main utility meters and submeters. The standard doesn’t say you need an energy management system. It says you have to be able to measure and collect your consumption data, but it says nothing about granularity.
What it does define is the concept of an SEU – a Significant Energy Use. That’s defined as a large consumption unit or a consumption unit with a large potential for reduction.
That’s the interesting part: where do you actually focus?
For a lot of applications, especially non-industrial, you really don’t need submeters in my opinion.
Henrik:
And this ties into one of the projects we’re working on with Jakob and 4BC in the EUDP project: seeing how far we can go using only main meter data while still staying within constraints like the ISO 50001 requirements.
Malte:
Think about it with the Plan–Do–Check–Act framework. As part of the planning phase, you decide where to collect data. If you have a building that accounts for, say, 50 percent of your total consumption, you might need to break that one down into further sub-components, beyond just the main meter.
Typically, that’s done with submeters. But you can also do it mathematically.
If one building uses 50 percent of the portfolio’s consumption, and half of that is driven by heating, a machine-learning model can pick that up and create a heat profile. You basically get everything you would get from a submeter.
In that sense, machine learning can act as a virtual meter.
The more interesting aspect for me is more traditional portfolios. Imagine a pension fund with 80 buildings that wants to do ISO 50001. Their buildings can pose a risk when it comes to water leaks. The question is: if you have main water meters, do you actually need submeters?
I’d argue no. With main meters, you can be notified when a risk occurs – when you have a leak. From there, it becomes a process question:
“Now that I know there’s a leak in one of my 80 buildings, how do I find out where it is and what to do about it?”
Benedetto:
That does require your water utility to collect and send data at, say, hourly resolution, so you can detect things in time.
Henrik:
Which many do – and most others will at some point. In many places it’s becoming mandatory.
You also need an analysis that’s intelligent enough not to spam you with alerts every time there’s a spike – for instance, when everyone takes a shower at the same time. It has to process the signal and turn it into the correct insight. Otherwise you end up with alert overload, which is exactly what we see in many organizations.
Malte:
That’s one of my main points. In at least three cases now, Ento has replaced a legacy energy management system with lots of submeters. In one case, there were around 5,000 submeters.
Now they just have main meters – fewer than a thousand – and it’s hard for them to grasp that they can get more relevant information out of fewer meters combined with intelligence than from 5,000 meters without intelligence.
With those 5,000 meters, they assumed they were getting value. But they weren’t. In one case, a water leak went on for a year and a half. Nobody had the capacity to manually analyze 5,000 meters in a spreadsheet, so it just didn’t happen.
For me, the flip point is when they see they can get more insights from less data – which is quite funny coming from a data company.
I was working on a headline:
“Can we confidently say that more data from submeters and sensors often leads to less savings?”
It flips the traditional thinking on its head.
Henrik:
Still, there must be cases where it makes sense. We have one very large facility with thousands of submeters. What are we actually doing there?
Malte:
That’s a fun story. In this facility, we would say it makes sense to have submeters to identify problems.
We did find a water leak there – but we found it on the main meter.
Let that sink in: there are thousands of submeters, but we found the leak on the main meter. The submeters were then used afterwards to narrow down where the leak actually was.
That’s where submeters make sense: as a second layer, once you already know there’s a problem.
The same can be true for electricity. The main meter might cover too many square meters, so you use submeters as “local main meters” for parts of the building.
Henrik:
Exactly. That’s the opposite of how ISO 50001 tends to think about it, where you focus on the energy-intensive systems: cooling, heating systems, process units, and so on.
Benedetto:
I’m developing a theory that the usefulness of submeters is inversely related to the size of your portfolio. The bigger your portfolio, the less submeters make sense. The smaller the portfolio, the more they might.
If you have hundreds of buildings and a couple of energy managers, what you really need is to know which building has the problem. With good analysis on main meter data, that’s enough.
“I have 500 buildings. In 10 of them, something is running at night; they have a water leak or poor performance. I need to know which 10.” Once you know, you send a technical team to those buildings and fix the problem.
If you have two energy managers for 10 hospitals, it’s less important to know which hospital has a problem and more important to know where within each hospital the issue is. If the hospital is thousands of square meters, and all you know from the main meter is that there’s a leak, it can take months to locate it. There, submeters can make sense.
Malte:
I think you want to link this to ISO 50001 again. A rule of thumb – not in the standard, but often used – is: if one unit represents a certain percentage of your total consumption, you break it down further.
So say in your large portfolio one SEU is 40 percent of total consumption. You break that into sub-units, so that no single unit represents more than, say, 10–20 percent of total consumption. The exact threshold depends on the absolute size and the economics, but the principle is sound.
It’s a less mathematical way of expressing your “inverse” theory.
Henrik:
I’m not sure the inverse relationship really holds. If you had five small buildings, you wouldn’t necessarily need lots of submeters either. It’s not just portfolio size; it’s the size of individual buildings relative to the coverage of the main meters.
Jakob often mentions a rule of thumb for SEUs: below 10,000 square meters, you might think of it one way; above that, in another. It’s not precise, but it helps structure the thinking.
Still, that “break up the very large units” principle can apply to anyone. Suppose your portfolio uses 100 GWh per year, and one building out of 100 uses 20 GWh. Your focus should be heavily on that building, and breaking it down further makes sense.
Benedetto:
I’d even challenge the SEU concept here: you might not want to add submeters even if the system is large, if you can estimate its behavior precisely.
We have all the data and context to do that now. We can extract clear heating and cooling profiles, see setpoints for process or comfort cooling, and so on, just from main meter data plus weather and other features.
We’ve seen that in big hypermarkets and similar sites: they save millions of euros just from main meter analysis. In those cases, it’s not about complex metering. Submeters might help localize a problem in the final step, but if you can already see hundreds of thousands of euros in savings potential from the main meter, you’ll be motivated to find the exact cause. That last step is not the big issue.
So even there, submeters are not strictly necessary.
Malte:
I feel like I interrupted you there.
Benedetto:
I was trying to defend my beloved inverse theory I just came up with.
Henrik:
You might be in love with it and need to kill your darling.
You had the role of steel-manning submetering. Did you get through all your arguments? Are there more reasons why it can make sense?
Benedetto:
I think the two main areas are:
- Very large buildings.
If you have thousands of cubic meters of water consumption, a leak that’s still relevant might be invisible at main meter level unless it’s massive. There a submeter can help. - Industrial applications.
Places where you really need to know what is consuming what, because you take decisions based on production units, process temperatures, and other process constraints.
That also ties into why Ento doesn’t work much in heavy industry. These processes are very custom and hard to predict using standard features, unlike more predictable buildings like supermarkets, offices, schools. Without a dedicated meter, it’s hard to build a robust proxy.
Malte:
The way I describe it is: in manufacturing, consumption is not driven by the variables we typically include (weather, opening hours, etc.), but by production.
We actually have some manufacturing on the platform – one example is a furniture producer. I expected we wouldn’t be able to give them recommendations from main meter data. But they operate five days a week, with a very structured schedule.
Even though their consumption is driven by manufacturing tables and chairs, the pattern is very regular. Because we integrate weekday and hour-of-day, we get a good understanding of their baseline.
Later they had an issue with a ventilation system that started running at night, 24/7. We detected that from the main meter. The relative change was not huge, but they could still save around 200,000 DKK, about 30,000 euros per year. They went out and fixed it.
So even in manufacturing, you can sometimes find problems with main meter data. There could be process optimizations you’d see more easily with submeters, but you can still detect big issues with mains if the operation is relatively stable.
If it had been a plant with constantly changing processes and very variable consumption, we probably wouldn’t have been able to do this meaningfully – there would have been too many false positives. But here, because it had been stable for so long and then suddenly spiked, the main meter was enough. I forwarded the alert, and they tracked it down to the ventilation system – which is funny, because it’s almost always ventilation or HVAC.
Henrik:
So I think we can at least conclude – and this is how we usually advise – that you should always start with the main meters. Even in industrial settings, but especially everywhere else.
Get the main meters in, get the overview, get the perspective. You’ll probably find issues; it’s very rare that we don’t. From there, you can discuss whether there are good reasons to dig deeper with submeters.
Malte:
I also wish more people thought about the trade-offs. One part is buying the meters. Another is maintaining the data streams. If it’s not utility-grade, it’s your problem: you have to maintain loggers, meters, communication.
Then there’s the analytics: is 5,000 meters better than 1,000? Why should it be? Those are the questions we try to instill in people.
We’ve already had a few cases where organizations were about to install hundreds or thousands of submeters for a lot of money, plus ongoing maintenance and analysis effort.
Instead, we got their main meter data from the utility. They realized they didn’t need the submeters, and they saved not only on energy, but also on installation and system complexity.
Henrik:
Nice. Let’s end that part here.
We also wanted to catch up on anything else interesting. Benedetto, you’re going to a conference in the UK to speak soon?
Benedetto:
Yes. It’s called “Verified”, taking place November 5 in Birmingham. It’s about M&V – measurement and verification of energy efficiency savings.
I didn’t even know there could be a whole conference just about M&V.
Exactly! We’ll be there to explore the role of AI in the M&V world, which until now has been very manual. Believe it or not, most M&V plans today are still based on monthly data, using simple regression. Then smart meters come in, AI comes in, and suddenly we can do much more accurate analysis of what’s going on in these facilities.
I could speak for hours about M&V, but the main theme is: how do we understand what we need to invest in, and what has no impact?
We have limited budgets for energy efficiency and decarbonization, so it’s crucial that what we do actually moves the needle.
I’ve been asked to run a workshop session on how to verify the outputs of AI. That will be fun. If you’re in the UK, you can join us.
Henrik:
That’s super nice. I feel like that should almost be the main topic for another discussion, because there are so many touchpoints with how people used to work and how we should be working today.
Benedetto:
Think about it this way: with M&V, you see the return on investment on your time and money. If you don’t do it, you’re acting blind. If you do it, you know that whatever you did last week had this impact.
That leads to two things:
- You hopefully stop doing the things that have no impact.
- You do more of the things that have a bigger impact than you thought, and you can prioritize them properly.
And this goes back to submeters as well. Many people think they need submeters to do M&V, but that often just stops them. They get stuck in Excel, pulling in weather data, normalizing, and so on.
All of that is what we care about automating as much as possible.
Henrik:
Nice. Is that a teaser that we have to do another episode?
Benedetto:
Let’s make a note for one of the next weeks: dig deeper into M&V and AI.
Malte:
If anyone listening has survived until now, they know they can’t miss the next episode – or your conference in Birmingham.
Henrik:
Exactly. Go ahead and book the flights.
Alright, thanks everyone.
Malte:
Thanks.
Benedetto:
Thanks. Cheers.
.jpg)
.jpg)
