Speakers:
- Maria – Moderator
- Benedetto – Speaker
Maria:
Okay, here we are. We’ve got some people joining. We’ll start in a few minutes, just giving everybody a chance to log in.
While you’re waiting, if you can access the chat, feel free to post a message and say hi to everyone.
I see more people joining, so while we’re waiting I’ll start sharing our presentation already.
[Screen share starts]
Let’s give it one more minute so people can join.
For those of you who are already here, you can grab a glass of water or a cup of coffee or tea and get comfortable for this one hour session.
I still see a few people jumping on, so we’ll wait a little bit longer and then we’ll get started. Also feel free to introduce yourself in the chat so others can see who’s here and what your background is.
Okay, I think we should slowly get started.
As Benedetto just said earlier, if you didn’t hear him, feel free to introduce yourself in the chat. We also just learned before starting that some of you might have issues accessing the chat. If that’s the case and any questions come up, please feel free to send me an email at maria@[company].com and I’ll make sure to pass your question along during the presentation or afterwards.
Don’t hesitate with any questions that might come up.
So, welcome to this webinar. My name is Maria and I am hosting together with Benedetto. Or, I should say, Benedetto is mostly hosting – he’ll do most of the talking today about IPMVP and how you can verify energy savings.
For those of you who might not be familiar with Ento, we are an online platform that analyzes energy consumption and helps you find ways to reduce it. You can connect all your energy data and we’ll analyze it and tell you if there are issues or problems. Based on that, we can also verify whether you have actually achieved savings.
Benedetto is one of the masterminds behind this measurement and verification feature, so I’ll hand it over to him and let him do most of the talking.
Benedetto:
Thank you very much, Maria, for the introduction, and welcome everyone to this webinar on IPMVP and how you can use it to verify your energy savings using our platform.
Let’s start with a quick overview of today’s session. I’ve organized the presentation into five sections:
- A general look at energy efficiency and one of the big problems in that field.
- A case study of how we measured savings in over 10,000 buildings in Denmark.
- How we at Ento developed a measurement and verification product to verify energy savings within the platform.
- First a bit of theory
- Then a live demo
- Finally, a Q&A session.
And again, if you can’t access the chat, you can send your questions to Maria by email and she’ll pass them on.
Energy Efficiency and the Core Problem
Benedetto:
Most people here are likely aware of how important energy efficiency is. In one of the latest IEA reports from 2023 they state that around 40–50% of the carbon reductions needed to stay in line with the Paris Agreement can be achieved simply by implementing the right amount of energy efficiency. So it’s a major part of the solution.
But we also have a big problem: avoided energy consumption cannot be directly measured.
What does that mean? Let’s compare it with renewable energy technologies.
If you install solar panels or a local generation plant, you typically have meters showing:
- How much energy you produce
- How much you consume in the building
- How much you inject into the grid
That makes it straightforward to estimate the expected ROI before implementation and then monitor the actual performance afterwards.
With energy efficiency, we have no meter that says: “this is how much energy you did not consume.” We only see the energy that was consumed.
Because of that, many energy efficiency projects today are still based on so-called “deemed savings”. That is a nice way of saying the savings are based on engineering calculations done beforehand. We don’t know if these savings are truly achieved over the life of the project.
So we end up with two problems:
- We cannot directly measure avoided consumption, and
- We rely heavily on ex-ante engineering estimates.
These two feed into a bigger problem: if we want to reach our climate goals, we need to be accountable for the results of the projects we implement in buildings. In our field this is known as measurement and verification (M&V) of savings.
IPMVP – The Framework
Benedetto:
How do we solve this?
Fortunately, we have a protocol: IPMVP – the International Performance Measurement and Verification Protocol. It defines guidelines and a methodology for estimating energy and water savings in buildings and facilities.
The good news is that applying this protocol becomes easier as we:
- Get more high-granularity data from buildings, and
- Leverage algorithms like machine learning or AI to process consumption data.
So what does the methodology look like in practice?
In simple terms:
- We first establish the normal state of how the building operates before the energy efficiency action. This is our baseline period.
- We analyze that consumption and feed it to a machine learning model.
- The model builds a baseline model that can predict an adjusted baseline or counterfactual consumption in the future.
This counterfactual is “what the consumption would have been if no action had been taken.”
For example, if on a certain date we:
- Did not change HVAC settings
- Did not replace windows
- Did not implement any other efficiency measure
then we would have expected a certain amount of consumption. The model estimates that.
In reality, after the action, we measure a different consumption on the meter. The difference between the predicted adjusted baseline and the metered consumption is the saving.
If you imagine a chart:
- First you see measured baseline energy in the baseline period.
- Then after the installation period, the model predicts the adjusted baseline in the reporting period.
- The gap between the adjusted baseline (what we would have used) and the actual metered consumption (what we did use) is the verified saving.
Case Study – 10,000 Municipal Buildings in Denmark
Benedetto:
Our first implementation of this approach at Ento was not directly as a product feature, but as a project for a large Danish organization that manages municipal buildings.
They asked us to verify whether energy-saving actions taken during the energy crisis – roughly from September/October 2022 to February/March 2023 – had actually made a difference.
During that winter, many energy-saving measures were implemented in municipal buildings across Denmark. The question was: Did they work?
We said yes, because:
- We already had the ability to seamlessly integrate data from many buildings into our platform.
- We could run the necessary calculations at scale.
So for each of those 10,000+ buildings, we:
- Calculated a baseline period.
- Projected the adjusted baseline for the crisis winter.
- Compared it with the actual measured consumption.
- Estimated whether savings had occurred and by how much.
- Aggregated the results.
One particularly interesting chart from that project showed the distribution of savings across the entire portfolio:
- On the x-axis: percent savings relative to each building’s total consumption.
- On the y-axis: number of buildings.
The distribution was skewed to the right of zero, meaning that as an aggregate, the portfolio saved energy. Of course, there were buildings that saved little or none, and some that saved a lot, but overall the average was clearly positive.
Another insight: we often assume that focusing on the biggest buildings drives most savings. But in this portfolio, many small buildings were actually driving a large share of the total savings. That was a surprising and important finding.
From Project to Product – Ento’s M&V Tool
Benedetto:
After that project we asked ourselves:
“This was a very interesting analysis. How can we make this available to many more municipalities and building owners, on demand, inside the platform?”
We started by looking at how M&V is typically carried out in the industry. There are a few common pitfalls:
- Data collection is long and cumbersome. You have to gather data from different systems, often manually.
- You’re not always confident in the quality of the data.
- The analysis is usually done in a manual Excel sheet, which:
- Takes many hours to set up
- Is fragile – it can be lost or corrupted
- Even after all that, you often feel uncertain about the reliability of the result.
We felt we could improve on this, because:
- We are directly connected to utility data, which is the same data customers are billed from. This is the source of truth.
- Data is automatically injected into our platform every day, so there’s no long manual data collection process.
- We can remove the manual Excel work by letting users launch verifications through a simple user interface.
- The results follow IPMVP, an internationally recognized protocol, so the savings are certified and trustworthy.
Technical Approach – Models and Features
Benedetto:
I’ll briefly touch on the technical side.
We use tree-based models, specifically gradient boosting machines, because they are:
- Fast
- Accurate
- Interpretable
We predict hourly energy consumption – not just daily or monthly. For each hour we try to estimate what the adjusted baseline would have been without the action.
The models use several types of features:
- Weather features:
- Outdoor temperature
- Solar irradiance
- Wind
- Precipitation
- Calendar features:
- Hour of the day
- Day of week
- Week of the year
- Bank holidays
- Manually defined holidays
- Special events:
- A COVID feature to handle the impact of lockdowns and changed occupancy patterns.
Every building is unique, so we use automated feature selection. The model decides which features are informative. For example, wind speed might be important for some buildings (e.g. very leaky buildings) but not for others.
We also tackled some common M&V challenges:
- Automatic period detection
We developed an AI-powered algorithm that analyzes the time series and automatically detects:- Baseline period
- Installation period
- Reporting/verification period
- Stopping conditions for verification
We stop a verification when one of three conditions is met:- We reach 365 days of verification
- Another action is implemented on the same meter
- The building changes to a different operating mode/segment (for example, a major change in usage pattern)
- Annualized savings from shorter periods
Sometimes savings are only verifiable for a short period (for example, because something else changed afterwards). We then extrapolate to annualized savings, based on:- The consumption already observed
- The savings already observed
- The expected remaining consumption for the year in that specific building
- Uncertainty and confidence levels
We compute 95% confidence intervals that reflect:- Model quality
- How well we understand the building’s behavior
If the model struggles – maybe because crucial features like occupancy are missing – the uncertainty range is larger. If the building is easier to predict, the range is smaller. This gives a more honest and risk-aware picture of the savings.
The key point is: we don’t just give you a number – we also tell you how uncertain that number is.
Live Demo – Savings Tool
Benedetto:
Let’s look at how this works in the platform.
[Screen share – Ento platform, Savings tab]
When you log into the platform and go to the “Savings” tab, this is what you see. For this part of the demo, we’re using real data from a Danish municipality (Horsens Kommune), who kindly allowed us to show their anonymized results.
At the top, you have an overview that shows, year by year, how your savings have developed from specific registered actions. For each action, we calculate annualized savings. Even if we can’t verify for a full year, we still annualize based on the verified part.
All of this feeds into:
- Total annualized savings across your history
- CO₂ avoided
- Monetary savings (for example, around 4 million DKK here, roughly 500,000 EUR)
- For water actions, we also show the volume saved in cubic meters.
Below, you see a table of actions:
- Action type (e.g. light optimization, ventilation settings, etc.)
- Energy type (electricity, district heating, solar PV, water, district heating penalty)
- Responsible person or energy manager
- Whether the action actually reduced consumption or not
- How long the verification has been running
- Annualized savings
- Comparison to expected savings, if you entered those
- Payback time, if you registered the cost
- Tags for grouping and filtering
Let’s open one concrete action.
[Opens an action]
Here we see:
- Action title
- Annualized savings
- Total verified savings
In this case, the action has been running for a full year, so the annualized and total verified are the same. If we had only verified for six months, the total verified would show the six-month figure, and the annualized savings would show the extrapolated full-year value.
The chart shows:
- Baseline period (before the action)
- Reporting period (after the action)
- The point in time when the action was implemented
- Other actions on the same site
- Excluded periods automatically detected and shaded (for example, outliers or mode changes)
This is where the AI-based segment detection comes in. We look at the time series and detect different operating modes over the building’s history. In this example, we see five different segments. The baseline period is the segment that was active when the action was implemented, and verification continues until one of the stopping conditions is met.
Below the chart, we summarize:
- Dates for baseline, installation and reporting periods
- Number of days verified
- Annualized savings
- Total adjusted baseline consumption – the total the building would have consumed if the action had not been implemented.
From that, we compute for example “32% savings compared to the adjusted baseline”.
You can also switch between daily and hourly views. The hourly view is powerful because it shows when you’re saving energy:
- Is it mainly during base load?
- Or mainly at peak hours?
This is important because saving during peak hours often has higher cost and CO₂ impact.
You also see details like:
- Cost of service, if entered
- Return on investment
- Meter type
- Energy manager responsible
Verifications – Comparing Periods (e.g. 2023 vs 2022)
Benedetto:
We realized that “actions” are just one side of the story. Sometimes you don’t want to tie savings to a specific action. You simply want to know:
“How did this building perform in 2023 compared to 2022?”
That’s exactly what our “Verifications” feature does.
In Verifications, you’re not registering a specific action. You’re just comparing a baseline period and a reporting period.
For example, a customer might want to:
- Compare 2023 vs 2021
- Or follow a mandated scheme, like 2023 vs 2019, depending on reporting requirements.
Let’s open one verification.
[Opens a verification for a specific site, district heating]
Here, the customer wanted to compare 2021 with 2023 for a district heating meter. The verification shows:
- Annualized savings (e.g. 72 MWh)
- Total verified savings for the period so far (e.g. 60 MWh over 334 days)
This verification updates daily as new utility data arrives, until it reaches 365 days. You can also switch to hourly view and see at what times the savings are happening.
Importantly, you can run verifications at different levels:
- Single site
- All sites of an industry (e.g. all schools, all supermarkets)
- Sites with a given tag (e.g. all supermarkets renovated in 2022)
- Or all sites in your portfolio
When you create a verification, you can choose:
- A single site
- An industry
- A tag
- Or leave it empty to verify the entire portfolio.
Creating a New Verification (Demo User)
Benedetto:
To avoid interfering with real customers’ data, I’ll switch to a demo user to show how you create a new verification.
[Switches to demo environment]
Let’s say we pick a site called “Bank High Street Branch 3” and we want to compare 2023 with 2022 for electricity.
We:
- Select the site.
- Choose electricity as the energy type.
- Set the baseline period as 1 January–31 December 2022.
- Set the reporting period as 1 January–31 December 2023.
- Click “Launch verification.”
This sends the calculation to our cloud computing platform. When it’s done, the verification appears in the list, just like the others.
I already have some portfolio-level verifications prepared here:
- 2021 vs 2020 on 38 sites, where we actually saw an increase in consumption, especially due to a higher base load early in the year.
- 2023 vs 2022, where we see a clear decrease in consumption, showing savings across the portfolio.
These portfolio verifications allow you to track performance across your whole portfolio over time.
Registering an Action – Example from a Bank
Benedetto:
Now let’s look at how you register a saving from an action.
We’ll stay with the demo bank buildings. These are real banks that actually achieved considerable savings after using Ento.
For one specific bank site:
- Before, the annual consumption was about 58 MWh.
- They discovered that the HVAC was running while the building was closed (a pure operational issue).
- By fixing the operating hours, they reduced consumption to around 38 MWh per year – a very significant saving with no major investment.
To register this action, the customer can:
- Go to the site and open the “Register Savings” dialog.
- Give the action a title, e.g. “Fixed HVAC operating hours”.
- Enter the implementation date.
- Enter the cost – in this case almost zero, maybe just a small service fee.
- Choose “Detect periods automatically” (default).
If they prefer, they can also:
- Manually define the baseline start and end dates.
- Define an installation period (e.g. if the building was closed during a retrofit).
- Ask the system to automatically exclude outlier segments.
I’ve already created two savings for this site to illustrate:
- About 6,000 EUR/year saved by fixing operating hours.
- About 4,000 EUR/year saved by an HVAC efficiency upgrade (e.g. new motors).
Our segment detection captures how the building changed over time:
- First, it operated normally.
- Then something went wrong – consumption increased.
- Thanks to Ento, they noticed it and fixed the hours – consumption dropped again.
- Later, they did the HVAC upgrade – it dropped further.
- Recently there may be a new issue starting, which they should now investigate.
So the tool doesn’t just quantify savings; it also helps you spot new problems.
Back to the New Verification
Benedetto:
Let’s quickly check if the verification we created earlier for “Bank High Street Branch 3, 2023 vs 2022” is done.
[Opens verification]
Yes, it’s done. Here we see that:
- The building has already been reducing consumption over previous years, e.g. through several actions.
- The verification now shows how 2023 compares to 2022 – so you can track this individual building’s performance over time.
At the same time, portfolio-level verifications let you see how your entire portfolio is performing, year-on-year.
That was a lot of talking. I hope it was clear how the tool works and how it can impact your organization, by making it much easier to report and to get insights into your savings.
I think we’re ready to move into the Q&A.
Q&A
Maria:
I’m going to jump in. The first question is not so much about IPMVP but more about the platform. Someone asks whether we offer an API to other software vendors. Is that something Ento is doing? Do you want to take that one or should I?
Benedetto:
You can answer if you want, otherwise I can.
Maria:
I’ll take it. Short answer: not right now. That’s not something we currently offer, but it’s on our radar and something we’re discussing internally.
The next question is:
“How do you extrapolate annualized savings from shorter verification periods? Is there a minimum duration needed to calculate annual savings?”
Benedetto:
There is no strict minimum duration. Of course, the quality of the prediction improves as the verified period gets longer.
Conceptually, we:
- Look at the actual consumption we have already observed.
- Look at the adjusted baseline for that same period.
- From that, compute observed savings for the verified period.
- Estimate how much consumption is still “missing” for the rest of the year (for that specific building).
- Extrapolate how much savings we can expect on that missing consumption.
So the annualized savings are based on:
- The savings we have already measured, and
- The predicted remaining consumption for the year.
If we’ve only verified, say, three months, there is more uncertainty than if we’ve verified eleven months. That’s reflected in the uncertainty ranges we show.
So, in short, we use the building’s real consumption and detected savings so far, and then extrapolate to a full year based on how much consumption we expect is still missing. The longer the verified period, the lower the uncertainty.
Maria:
That was clear. We’ll leave it at that, and if anyone wants more technical detail, let us know in the chat.
Next question:
“How do you manage an action that saves both district heating and electricity? Is this process automated?”
Benedetto:
Right now, it’s not fully automated. At the moment you need to:
- Create two actions:
- One on the district heating meter
- One on the electricity meter
In the future, we plan to allow you to:
- Select multiple meters when creating a single action,
- And then automatically launch the necessary verifications for each meter.
That’s part of our roadmap. We started building this tool only a few months ago, and we have quite a few improvements planned.
Roadmap and Future Improvements
Maria:
You mentioned improvements in store. Could you elaborate a bit on what’s coming?
Benedetto:
Yes, sure. I actually have a slide for this.
[Shows roadmap slide]
The main things we’re working on now are:
- Expanded info for aggregated verifications
- Today, when you open an aggregated verification (e.g. all sites), you mainly see total savings.
- In the future, we want to add more graphs – for example, a portfolio distribution similar to the case study chart I showed earlier.
- We also want to show which sites contribute how much to the total savings, so you can easily see:
- Which sites are leading the savings
- Which sites are lagging and might need attention
- Model inspection and explainability
- This is aimed at more technical users who want to dig into the models.
- You’ll be able to see:
- The relative importance of each feature (weather, calendar, COVID, etc.)
- Whether we use certain features at all for a given building
- Even hour-by-hour contributions of different features to the prediction
- This helps “geeky” energy managers really understand what drives energy use in each building.
- Matching hourly savings with hourly price and carbon intensity
- We already calculate hourly savings.
- Today, we multiply them by an average price and an average grid carbon intensity.
- We want to go further and match them with:
- The actual hourly electricity price from your utility contract
- The actual hourly grid carbon intensity, from external data sources
- This matters because saving 1 kWh at noon on a sunny day is not the same as saving 1 kWh at night when a coal plant might be running.
So these are the three big areas. But we also constantly incorporate customer feedback, and if something else provides higher value, it can jump ahead in the roadmap.
Maria:
We got some clapping emojis for that, so that’s clearly appreciated.
Next question:
“If multiple actions take place within a short time period, how do you make sure they don’t overlap in the same verification?”
Benedetto:
That’s a really good question and honestly one of the hardest problems in M&V.
There is no easy solution. From a logical point of view:
- When you implement an action, then implement another action one or two months later, the new baseline for the second action should be that “two-month” state.
- But if your new baseline is only 1–2 months long, you lose seasonality. You don’t know how the building behaves in winter vs summer, etc.
We handle it in several ways:
- We stop the verification of an action when a new action is registered on the same meter. That way, we don’t keep attributing savings to the old action that are actually due to the new one.
- For the baseline, there are different strategies, and sometimes the practical answer is to merge actions:
- If actions were implemented very close in time (e.g. within 1–2 months), it can make sense to register them as a combined action.
- You can then treat the period in between as an installation period.
- You won’t get separate savings for each individual action, but you’ll get a reliable figure for the combined effect.
- We’re also working on R&D methods to improve this further:
- We want to avoid forcing customers to always merge actions.
- We’re exploring ways to leverage the large amount of time series data we have across many buildings to better isolate each action’s impact, even when actions are close in time.
From an IPMVP perspective, the traditional advice is often:
- Carefully define baseline periods so they’re not contaminated by other actions.
- If you only have a very short baseline (e.g. 1–2 months), that’s always going to be challenging.
This is an area where we hope advances in AI and more research will let us bring something new to the field. I wrote my PhD on measurement and verification, and I can say that no one has a perfect solution yet for multiple tightly clustered actions. We’re actively working on it.
Maria:
Thank you. Another question:
“What type of machine learning algorithms do you use to build the energy baseline?”
Benedetto:
We use gradient boosting machines, which are tree-based models.
They’ve consistently been shown to be:
- Very strong for tabular data like ours
- Capable of handling nonlinear relationships between features and energy consumption
- Relatively fast and interpretable compared to deep neural networks
For example, in a well-known Kaggle competition on energy consumption prediction, all the winning solutions used gradient boosting variants. That’s one of the reasons we chose them.
Maria:
Thank you. Next question goes back to the hourly price topic:
“When you say hourly prices, does that include grid/’shipping’ price? How will you handle different price structures from different utilities, and can you handle tax deductions?”
Benedetto:
We’re in the process of implementing this. The way we typically work is:
- For each customer, we set up custom integrations to their utility and contract structure as part of the onboarding process.
- We already do something similar for energy data; we’ll extend this to price and tariff data.
So when we onboard a new customer, we:
- Identify which utility they use
- Understand their tariff structure, grid fees, etc.
- Store that in our database
Then, any new verification or savings calculation can:
- Use the correct hourly price
- Include grid fees (“shipping price”)
- Include tax considerations if relevant and configured
In other words, it will be integrated into our existing onboarding and configuration process. Once set up, it’s automatic.
Maria:
Perfect. I don’t see new questions for the moment.
If you’re already using the platform and want to test this on your data, or if you’re here for the first time and want to try it, please feel free to reach out. We can help you get your data onto the platform and test this verification tool.
We do have one more question:
“Are there any types of buildings where the measurement and verification results were not accurate?”
Benedetto:
It’s less about a specific category of buildings and more about the data available for a given building.
We’re trying to explain a building’s consumption based on a set of external features. Sometimes, we’re missing a feature that’s really important for a specific building, for example:
- Highly irregular occupancy patterns that don’t follow usual calendar behavior.
Usually we use calendar features (weekday, time of day, etc.) as a proxy for occupancy. For many buildings – offices, supermarkets, banks – this works well. The models are quite accurate without explicit occupancy data.
But in some buildings, occupancy patterns are irregular and don’t correlate well with the calendar. In those cases:
- The model can’t fully capture the behavior.
- Prediction quality is lower.
- As a result, uncertainty ranges are larger.
However, in our experience across many tertiary buildings (offices, supermarkets, banks, etc.), the models work quite well with the standard feature set.
Closing
Maria:
We’re happy to hang around for another minute in case more questions come in.
Otherwise, the timing is quite good – we’re close to the end of the hour.
As mentioned earlier, if you don’t already have your data on the platform, you can:
- Send us an email
- Book a demo
- Get your data connected and test this verification tool on your own buildings
You can contact Benedetto or me directly, or you can reach out to our sales team, whatever works best for you.
If there are no more questions, I think we’ll say thank you so much for joining.
Any final words, Benedetto?
Benedetto:
Nothing major – just that we ended almost exactly on time, which is nice.
I hope the webinar was informative and that you now have a clearer idea of:
- How IPMVP-based verification works, and
- How our tool can make reporting and gaining insights into your savings much easier.
If you want to know more about the technical side, feel free to reach out. My email was on the first slide, and you can also find me on LinkedIn by searching for my name.
We’re always happy to have these kinds of conversations.
Maria:
Perfect. Thank you everyone for joining, and we hope you enjoy the rest of your day.
Benedetto:
Thank you. Bye.
Maria:
Bye.
.jpg)
.jpg)
