Two reports on the same building, covering the same quarter, can disagree by a wide margin on how much energy was saved. Neither report is lying. They have simply made different choices somewhere along a chain that runs from a current transformer to a percentage on a slide. Following that chain is the most useful way to understand what an energy management system really does.
Four links in the chain
- Measure: meters and sub-meters, with an accuracy class that is a specification, not a given.
- Store: interval data, timestamped and aligned, which is where most data quality problems appear.
- Compare: a baseline normalised against weather and activity, otherwise the numbers describe the winter rather than the works.
- Act: schedules, setpoints and load shifting, which is the only stage that changes a bill.
Measurement: the accuracy class is a choice someone made
An energy management system knows nothing that its meters do not tell it. Electricity meters are built to defined accuracy classes, and for transformer-operated static meters measuring active energy those classes are set out in IEC 62053-22:2020, covering classes 0,1S, 0,2S and 0,5S. The S classes matter because they hold their accuracy at low load and at poor power factor, which is precisely where a building spends its nights and weekends.
The instrument transformers feeding the meter carry their own error, and in a retrofit they are frequently the weak link rather than the meter itself. A billing-grade main meter paired with cheap sub-meters produces the familiar situation where the parts do not add up to the whole, and where the difference gets quietly labelled as unaccounted consumption.
Sub-metering is what turns a bill into information. A single incoming meter tells you the building used more than last month. Separating heating, cooling, ventilation, lighting and plug loads tells you which of them did, and that distinction is the entire difference between a report and a work instruction.
Storage: interval data, and the problems that live in it
Modern systems record at intervals, commonly every fifteen or thirty minutes, and it is the shape of that profile rather than the monthly total that carries the diagnostic value. A baseload that does not fall at night, plant that starts three hours before anyone arrives, a chiller running through a mild weekend: all of these are invisible in a monthly figure and obvious in an interval trace.
The recurring problems at this layer are unglamorous. Clocks that drift between devices, so half-hourly readings no longer align across meters. Gaps during a communications failure, filled in by an algorithm nobody documented. Points renamed during a maintenance visit, quietly breaking a report. Naming is a genuine industry gap here: the semantic layer intended to make points machine-readable across vendors, including the work under ASHRAE 223P, remains at proposed stage rather than published, so in practice a project still relies on a disciplined, human-maintained points list.
| Stage | What it produces | Where it commonly fails |
|---|---|---|
| Measurement | Readings per circuit or end use | Accuracy class and current transformers left unspecified |
| Storage | Aligned interval time series | Clock drift, gaps, renamed points |
| Comparison | Performance against a baseline | No normalisation for weather or activity |
| Action | Revised schedules and setpoints | Overrides applied locally and never removed |
Comparison: the baseline is where the argument happens
Raw consumption is almost never comparable year on year, because the weather, the occupancy and the activity of the building all changed. Normalisation is the step that removes those effects. Degree days are the standard instrument for the weather part, adjusting heating or cooling consumption to a common climatic reference so that a mild winter does not get recorded as an efficiency achievement.
ISO 50006:2023 is the guidance standard for this work within the ISO 50001 energy management framework. It sets out how to define an energy performance indicator and an energy baseline, and how to choose the relevant variables to normalise against, which in a commercial building usually means degree days plus a measure of activity such as occupied hours or floor area in use.
Where a claim needs to survive scrutiny from a landlord, a funder or an auditor, the reference framework is the International Performance Measurement and Verification Protocol, published by the Efficiency Valuation Organization. Its Option C compares whole-facility consumption before and after, which suits deep retrofits affecting many systems at once, while narrower options isolate a single measure. ISO 50015 covers the verification of energy performance improvement alongside it.
Action: what the system actually sends back to the plant
Analysis on its own changes nothing. The stage that moves a bill is control, and the interventions are less exotic than the software suggests: correcting run times so plant matches occupancy, widening dead bands so heating and cooling stop competing in the same zone, staging equipment so it runs near its efficient load, and resetting the setpoints that were overridden during a complaint eighteen months ago.
One distinction deserves care, because it is frequently blurred in vendor material. Shifting load to a cheaper tariff period reduces cost without reducing consumption. Reducing load lowers both. Both are legitimate objectives, and a system reporting them as one number will flatter itself.
Where an energy management system meets scheduled maintenance is where its data gets a second use. The same trend data that shows a fan running out of hours also shows it drawing more power for the same duty over a season, which is the entry point to condition-based maintenance in a property portfolio.
Common questions
Is an energy management system the same thing as a building management system? No, although they overlap and are often sold together. A building management system exists to operate plant. An energy management system exists to measure, compare and report on consumption, and it may or may not have authority to write commands back to that plant.
How long before the data is useful? Interval data starts revealing scheduling faults within weeks. A defensible baseline needs a full annual cycle, because the building has to be observed through both a heating and a cooling season before any normalised comparison means much.
Does the system need to control anything to pay for itself? Not necessarily. A monitoring-only deployment that exposes out-of-hours running and mis-scheduled plant often pays back on manual corrections alone, and it also tells you which automation is worth buying next.
Are machine learning models replacing this chain? They sit on top of it. A model trained on drifting sensors and misaligned timestamps produces confident output from poor input, so the measurement and storage layers still determine what any analytics layer above them is worth.
Is any of this legally required? In the European Union, non-residential buildings above defined output thresholds must be fitted with building automation and control systems under the recast Energy Performance of Buildings Directive, Directive (EU) 2024/1275. National transposition is what applies in practice, so the local implementing rules are the text to read.
From understanding the chain to running the project
Knowing how the measurement works is the easy half. Procuring, specifying and commissioning a system that keeps working is the other one.
Read our implementation guide for building energy management systems
Published in 2025. Updated on 13 August 2026. Sources: IEC 62053-22:2020; ISO 50006:2023; ISO 50015; International Performance Measurement and Verification Protocol, Efficiency Valuation Organization; Directive (EU) 2024/1275. Figures and thresholds are those published at the date of writing.
