Before you accept a proposed reduction figure, understand the measurement basis behind it — and the five questions that separate a verifiable claim from a number.
Since September 2025, BCA's Mandatory Energy Improvement (MEI) rule has required energy-intensive existing buildings in Singapore to audit their consumption and improve it — with verifiable data. The regulated electricity tariff for the second quarter of 2026 is 27.27 cents per kWh before GST. Carbon tax has moved from S$25 per tonne in 2024–25 to S$45 in 2026–27, with a S$50–80 target for 2030. And cooling accounts for 40–50% of a typical commercial building's electricity.
Put those together and you get a specific kind of meeting. A supplier sits across from you with a proposal, and the proposal contains a number. Twenty percent. Twenty-five. Thirty.
The number is the easy part. The difficult part is that a cooling saving is not something you can weigh. It is the difference between two measurements taken at different times, under weather you do not control — and the size of that difference depends enormously on how those two measurements were chosen.
This article is about that choice: what a defensible comparison actually involves, what happens to the headline figure when the basis changes, and what to ask when a proposal lands on your desk.
Three things decide how much electricity an air-conditioning system draws at any given moment:
None of these hold still. A retrofit is installed on a particular day, which means the “before” period and the “after” period are, by definition, different weeks of weather. If the after-period happens to be milder, consumption falls whether or not the retrofit does anything. If it happens to be hotter, a genuine improvement can be buried.
So a reduction percentage is never simply a fact about the equipment. It is a fact about the equipment and about how the two periods were selected. A supplier who hands you a percentage without describing that selection has not given you a measurement. They have given you an outcome.
We can show what this does with our own data, from a single supermarket site running a 10 HP outdoor unit.
Same site. Same equipment. Same underlying data set. Eleven percentage points apart, purely because of which hours were counted.
Neither figure is wrong. The matched-condition number answers “what does the retrofit do when other variables are held still?” The all-condition average answers “what happened over the whole period, weather included?” They are answers to different questions, and a proposal that quotes one without naming which one is not telling you enough to evaluate it.
We publish both. On our Energy page the case card carries the matched-condition figure and a footnote states the all-condition average alongside it. We would rather explain a spread than present a single number you cannot check.
The same discipline applies to the group figure. The 22.23% we publish as an ESS average is a simple, unweighted mean of six individual case results, each measured under its own site conditions; for the re-verified site described above, the value entered into that mean is the average of its matched-condition result (27.4%) and its all-condition average (16.3%). It is a summary of separate measurements, not a single controlled experiment — which is why, whenever one verified figure is needed rather than an average, we cite an individual case directly.
Here is the procedure our verification work uses. It is not complicated, but each step exists to remove a specific way of being wrong.
The consequence is worth stating plainly: this method is designed to produce a conservative, reproducible, audit-ready lower bound, not a best case. That is what makes the resulting figure usable in the kind of documentation MEI and Green Mark ask for — and it is also why our published numbers are lower than they could be.
A pharmaceutical plant. Metered consumption fell from 1,693.5 kWh to 1,253.9 kWh across the comparison, which is a 26.0% reduction as measured. Normalising the residual condition difference through the cooling characteristic table brings the reported figure down to 25.3%. The measured current at the unit moved from 4.4 A to 3.3 A over the same comparison.
A supermarket, 20 HP. This one is more instructive, because it did not improve uniformly:
The band where the result was negative is the part most suppliers would remove. We publish it, because a saving that only appears at aggregate level is a saving the reader should be able to interrogate. It also tells a facilities manager something operationally useful: the benefit concentrates where the compressor is working hardest.
The physical mechanism is measurable too. Across the retrofit panel, refrigerant temperature falls from 53.6 °C at the inlet to 49.4 °C at the outlet — a 4.2 °C drop measured on site. That additional sub-cooling of 3–5 °C after condensing is what reduces the refrigerant mass flow required for the same cooling, and compressor power follows the mass flow. It is a small number, and it is the reason the larger ones exist.
Separately, the panel itself has been through third-party pressure testing: 5.0 MPa held for 24 hours with no anomaly, burst at 11.0 MPa giving a safety factor of 2.2, and no corrosion under salt-spray exposure. Those figures answer a different question from the energy ones, but it is usually the next question a facilities manager asks. The full set is on our Proof Hub.
None of these require you to be an HVAC specialist. They are the questions whose answers determine whether a number can be defended in an audit.
Why it matters: as shown above, the same site can produce figures eleven points apart depending on the answer.
A strong answer names the basis without hesitation and can give you the other figure as well.
A weak answer treats the question as a technicality, or produces only the more favourable number.
Why it matters: if the periods were simply “the month before” and “the month after”, the figure includes whatever the weather did.
A strong answer describes an explicit pairing rule — same hours of day, a stated tolerance band on outdoor temperature and humidity.
A weak answer is “we compared the bills”.
Why it matters: pairing never removes every difference. What happens to the residual is where a number is either normalised or quietly inflated.
A strong answer names the normalisation method and the source of the correction data.
A weak answer is that the remaining difference was negligible, with nothing to support it.
Why it matters: a reduction achieved by running the system less, or by letting the space sit warmer, is real on the bill but is not what the equipment did. If those are folded in, you cannot attribute the result — and you cannot repeat it.
A strong answer lists what was left out and why.
A weak answer is that nothing was excluded.
Why it matters: results from older equipment do not transfer directly to new equipment, and a retrofit almost always has warranty consequences. Both facts change your decision.
A strong answer states the age of the tested machines and is direct about warranty implications, including the unwelcome parts.
A weak answer avoids the warranty question.
If a supplier cannot answer questions 1 to 3, you do not have a measurement. You have a marketing figure with a decimal point.
Applying the same standard to ourselves:
Those constraints are not caveats bolted onto a sales claim. They are what makes the rest of the numbers worth reading — and they are the same questions we would want you to ask us.
If you want to see the method applied end to end, the case data and the third-party test results are collected on the Proof Hub, and the retrofit itself is described on the Energy page.
Want this method applied to your own site?
Email contact@ecoism.biz
or start here.
