A demand reduction project produces two outputs. One is a lower bill. The other is a demonstrated saving that a finance function accepts.

The second is worth more, because it is what funds the next project. A saving that cannot be shown convincingly is treated, in every capital process that works properly, as a saving that did not happen.

The problem, stated precisely

You cannot measure a saving. You can measure consumption before and consumption after, but the difference between them includes everything else that changed in the meantime: production volume, weather, occupancy, shift patterns, other projects, a hot August.

A saving is the difference between what happened and what would have happened in the absence of the measure. The second of those never occurs and therefore cannot be observed. Estimating it defensibly is the entire discipline.

The two broad approaches

Isolate the equipment. Meter only the affected equipment, before and after. Precise, and it excludes everything else happening at the site. It requires sub-metering, and it only tells you about that equipment — which is a limitation when the question is whether the site's demand determinant moved.

Analyze the whole facility. Use the utility meter, with a mathematical model relating consumption to the variables that drive it — production output, degree days, occupancy — established during the baseline period and used afterward to predict what consumption would have been.

For demand work the whole-facility approach is usually the right one, because the demand determinant is a property of the whole site. A measure can reduce one piece of equipment's contribution and leave the site peak untouched, and only the site-level analysis will reveal that.

The standard reference for both, and for the conventions around them, is the International Performance Measurement and Verification Protocol.

The baseline is the deliverable

Everything rests on the baseline, and the baseline has to be established before the measure is installed. This is the single most common failure: the equipment goes in, the bills look better, and nobody can prove it.

Establishing a defensible baseline
  1. Twelve months of interval data at the billing interval, covering a full seasonal cycle: how to get your interval data.
  2. The corresponding independent variables for the same period: production output, degree days, operating hours, occupancy.
  3. A written description of how the facility was operated, including setpoints, schedules and anything unusual.
  4. A note of any period that should be excluded — a shutdown, a commissioning trial, a period billed on estimated reads.
  5. The relationship between consumption and the independent variables, quantified and documented while the baseline is current.
  6. Agreement, in writing and in advance, on which adjustments will be made and how.
  7. The specific determinants being tracked: energy, facility demand, on-peak demand, power factor. Demand work is judged on demand determinants, not on kilowatt-hours.

The sixth item is what makes the exercise credible. Adjustments agreed in advance are methodology. Adjustments introduced after the results are known are an argument, and a reviewer will treat them as one regardless of how sound they are.

Adjusting honestly

Adjustments are not a way of massaging a result; they are what makes the comparison meaningful. If production rose 12 percent after installation, comparing raw consumption understates the saving, and the correction is legitimate.

What makes an adjustment defensible:

  • The relationship it uses was derived from the baseline period, not fitted to the reporting period.
  • The independent variable is genuinely independent — production output rather than a figure that the measure itself influences.
  • It is documented before the reporting period begins.
  • It is applied whether it helps or hurts. An adjustment methodology used only when it improves the result is not a methodology.

Demand determinants need their own treatment

Verifying an energy saving and verifying a demand saving are different exercises, and the second is frequently attempted with tools designed for the first.

Energy is cumulative and averages well. Demand is an extreme value, and extreme values are noisier. A month whose peak fell by 8 percent may reflect the measure, or may reflect the fact that the day which would have set the peak happened to be cooler.

Three practical consequences:

Use more than one month. A single month is not evidence about an extreme value. Several months, compared against equivalent baseline months, are.

Compare like periods. August against August, not August against April.

Look at the whole upper tail, not just the maximum. If the measure works, the tenth-highest and hundredth-highest intervals should have fallen too. If only the single maximum moved, that is as likely to be weather as a measure.

That last test is the most informative one available and it costs nothing but a sort.

When the numbers disappoint

They sometimes will, and how that is handled determines whether there is a third project.

The useful response is diagnostic rather than defensive. The common explanations are specific and checkable: the measure addressed the on-peak determinant while the tariff also bills a facility demand — facility, on-peak and billing demand; a control strategy was disabled or drifted; restored load created a new peak just outside the window; or the site was billed against a ratchet floor throughout, in which case the reduction is real and simply has not reached the bill yet — ratchet clauses.

That last case is worth watching for, because it produces the confusing outcome of a measure working perfectly and a bill that does not move. The saving begins when the floor rolls out of the look-back window, and being able to say so with data is very different from saying it without.

Why to do it even when nobody asked

Verification is frequently treated as a cost imposed by an incentive program. It is more usefully seen as the thing that converts one successful project into a program of them.

A finance function that has seen one energy claim verified rigorously treats the next one differently. That is worth more than the incentive payment, and it is the reason the method should be designed alongside the business case rather than retrofitted: building the business case for demand reduction.