In brief
- A before-and-after comparison is not proof: weather, occupancy and production all move consumption on their own.
- A credible saving compares the meter with a baseline predicted for this year’s conditions, built by regression on at least twelve months of data.
- IPMVP, ASHRAE Guideline 14 and ISO 50015 set the method, and statistical tests such as CV(RMSE), NMBE and the p-value decide whether a result can be believed.
- HarmoniQ’s 10–20% kWh saving is modest enough for weather to mask, so every installation is metered, normalised and validated, and the guarantee is written against kWh at the meter.
The simplest way to measure an energy saving is to compare consumption before and after the intervention. If a facility used 1,000,000 kWh last year and 920,000 kWh this year, did the intervention save 80,000 kWh?
Not necessarily. This year might have been milder, with less air conditioning in summer and less heating in winter. Occupancy might have fallen. A production line might have shut down, or a new machine been added. Any of these could explain the change, wholly or in part, without the intervention having done anything at all.
A credible saving has to answer a harder question: what would consumption have been this year, under this year's conditions, if the intervention had never been installed? The difference between that prediction and what the meter actually recorded is the saving.
The equationSaving = predicted baseline consumption − actual consumption ± adjustments
The predicted baseline is what the facility would have consumed under the actual conditions of the reporting period, its weather, occupancy and production, if nothing had changed. The adjustments account for anything else that did change, such as a line added or a shift dropped.
01Weather normalisation
In most buildings, weather is the largest single influence on energy use; in a plant, production sits alongside it. Weather normalisation is the process of removing weather's influence mathematically, so that a genuine saving becomes visible.
During the baseline period, typically the twelve months before the intervention, metered energy data is paired with weather data from a nearby station: temperature, humidity, solar radiation. A regression model is then built that describes how consumption at that specific facility responds to the weather.
After the intervention, the model is fed the actual weather of the reporting period and predicts what consumption would have been with nothing changed. The difference between that prediction and the metered consumption is the weather-normalised saving.
The method, step by step
- Collect baseline data At least twelve months of metered energy, paired with weather data from the same period. The baseline must cover the full range of seasonal conditions.
- Build the regression model Establish how energy use depends on its independent variables: temperature, humidity, occupancy, production. The model must pass statistical validation.
- Install the intervention Power-quality correction, a lighting upgrade, a change to the HVAC: whatever the energy conservation measure is.
- Collect reporting-period data Keep metering consumption under actual operating conditions, and keep collecting the same weather and operational data.
- Calculate the normalised saving Feed the reporting period's weather into the baseline model to predict what would have happened, then subtract actual consumption. The difference is the verified saving.
Without normalisation, a mild winter can make an ineffective intervention look like a success, and a hot summer can make an effective one look like a failure. Normalisation removes that noise and leaves the signal.
The figure below works the opening example through on an illustrative site whose real saving never changes. Make this year hotter or cooler than last: the before-and-after comparison swings, while the weather-adjusted saving holds.
Before and after
8.0%less
920,000 kWh this year, against 1,000,000 last year
The weather alone
+8.2%
This year’s weather, fed to the baseline model: 1,082,000 kWh
Weather-adjusted saving
15.0%
162,000 kWh below the adjusted baseline
This year ran 2 °C hotter than last. The meter reads 920,000 kWh against last year’s 1,000,000, an 8.0% drop, but the hotter weather alone would have raised use 8.2%, to 1,082,000 kWh. Against that adjusted baseline the saving is 15.0%, or 162,000 kWh.
02The verification frameworks
Three internationally recognised frameworks govern how energy savings should be measured and verified. They are complementary, not competing.
IPMVP: the International Performance Measurement and Verification Protocol
Maintained by the Efficiency Valuation Organization (EVO), IPMVP is the overarching framework used worldwide. It defines four options, chosen by the type of intervention and the data available.
| Option | Name | Method | Best for |
|---|---|---|---|
| A | Retrofit isolation: key parameter | Measure the most critical parameter, estimate the rest | Simple, single-system changes, such as lighting |
| B | Retrofit isolation: all parameters | Measure every parameter of the affected system | Complex single systems with variable loads |
| C | Whole facility | Analyse utility meter data by regression | Several measures, whole-building savings |
| D | Calibrated simulation | A calibrated energy model | Where baseline metering was not available |
Option C is the natural fit for power-quality correction, because its effect is spread across the whole facility: lower I²R losses in every cable, cooler transformers, less current through every board. Distributed effects are best captured where the utility measures them, at the main meter. The direct counterpart is retrofit isolation, Options A and B, which is where the on/off test sits: the effect measured at the affected system, minutes apart. The two methods are the two halves of the same standard.
ASHRAE Guideline 14
ASHRAE Guideline 14-2014, Measurement of Energy, Demand, and Water Savings, gives the detailed statistical criteria that IPMVP refers to. It specifies how to build regression models, which validation tests they must pass and how to calculate the uncertainty of a saving. It is the statistical backbone behind IPMVP's framework.
ISO 50015
ISO 50015:2014, Energy management systems: Measurement and verification of energy performance, is an internationally recognised M&V framework that complements both IPMVP and ASHRAE. It matters most to organisations running an ISO 50001 energy management system, because it gives them the method for demonstrating continuing improvement in energy performance.
03What makes a result credible
A saving is only as good as the statistics behind it. These are the tests that decide whether a result should be believed.
| Metric | What it measures | Threshold to look for |
|---|---|---|
| CV(RMSE) | How well the model fits the data; lower is better | 15% or less for monthly data, 30% or less for hourly |
| NMBE | Systematic bias in the predictions; closer to zero is better | Within ±5% for monthly data, ±10% for hourly |
| p-value | The probability that the result is down to chance | Below 0.05, ideally below 0.01 |
| R² | The share of the variance the model explains | Above 0.75, though CV(RMSE) says more |
| Cohen's d | Practical significance: the size of the effect | 0.5 is a medium effect, 0.8 a large one |
| Fractional savings uncertainty | The uncertainty band around the reported saving | Below 50% of the saving, at 68% confidence |
Statistical and practical significance
A result can be statistically significant, with a low p-value, and practically insignificant, with a tiny effect. Equally, a large effect with weak statistics may be real, but the data is not yet conclusive. Credible M&V needs both: a low p-value, so the saving is not down to chance, and a meaningful effect size, so the saving is large enough to matter.
Statistical significance
Not down to chance
A low p-value: the saving is unlikely to be an accident of the data.
Practical significance
Large enough to matter
A meaningful effect size: the saving is big enough to be worth having.
04Common pitfalls
Pitfall 1
Cherry-picking the baseline
Choosing an unrepresentative baseline period, such as a year of unusually high consumption, inflates the apparent saving. The baseline must cover the full range of normal operating conditions: every season, typical occupancy, representative production.
Pitfall 2
Too little data
A weather-sensitive facility needs at least twelve months of baseline to capture every season. A shorter baseline risks missing a heating or cooling season altogether, and producing a model that fails outside the conditions it was built on.
Pitfall 3
Ignoring non-routine events
Changes in production, equipment, occupancy or operation between the baseline and reporting periods must be identified and adjusted for. Otherwise the saving is misattributed: a factory that cut production will show lower consumption whatever was installed.
Pitfall 4
Overfitting
Too many variables for the data available gives a model that fits the past perfectly and predicts poorly. A good regression model is parsimonious: it uses only the variables that genuinely drive consumption.
05Why this matters for power quality
HarmoniQ typically reduces the energy a facility consumes by 10–20%, in kilowatt-hours, for the same output. That is a genuine and valuable saving, but it is modest enough that weather, production changes or seasonal shifts can mask it or exaggerate it.
Without rigorous M&V, a facility manager has no way to know whether a drop in consumption came from the correction, a mild winter or a production slowdown. With it, the saving is isolated, quantified and statistically validated, which gives confidence to everyone from the facility manager to the CFO to the board.
What HarmoniQ provides
Every HarmoniQ installation is metered before and after, with revenue-grade equipment.
We build a weather-normalised regression model of the site and validate it against recognised statistical thresholds, following IPMVP, ASHRAE Guideline 14 and ISO 50015. Our savings reports are fully documented and independently verifiable.
The same discipline is why our minimum performance guarantee is written against kWh measured at the meter, not against the bill. A bill moves with tariffs and charges outside our control; normalised consumption is what the intervention changed. So when we report a saving, it has been weather-normalised, statistically validated and checked against international standards. It is not an estimate. It is a measurement.
