Reducing consumption, and knowing that you did

Lists of energy-saving measures all look alike, and most of them are correct. The problem is that they are generic: they say what works on average, not what works here. So priorities end up set by whoever argues most confidently, and the declared saving stays the one in the quote.

Why the declared saving is never checked

Checking means comparing what happened with what would have happened without the intervention. The second half of that comparison exists nowhere: nobody ran the plant in parallel without the change.

In its absence two substitutes get used, and both are weak. The first is last month, which had different production and different weather. The second is the same month last year, which has the same flaw with a year’s delay. When the result comes out favourable it is credited to the intervention; when it does not, an explanation is found. Neither case verified anything.

With measurement and a baseline the missing comparison can be reconstructed: not the plant without the intervention, but the consumption that plant would have had under the same conditions. It is an approximation, and it is the only defensible one.

Rank by weight, not by ease

Split consumption lines the loads up by itself. The order is usually a surprise: the items people argue about most are rarely the ones at the top, and the first three explain more than anyone expects.

Ranking by weight has a second, less obvious and more useful effect: it makes the size of the unmeasured part visible. If the residue between the supply point and the sum of the known loads is a fifth of the total, the first action is not an intervention, it is one more meter.

Quantify each measure before doing it

The history makes possible something a quotation does not: estimating what a change would have done over the period already recorded. Not over a typical day, not over a catalogue profile — over the actual hours of the year just gone, with the shutdowns, short shifts and odd weeks that really happened.

It is the quickest way to discover that a technically excellent measure returns little because the load runs two months a year, and that a mundane control change returns a lot because that load is on all the time.

Some decisions repeat every day

Which source is cheaper in this hour, when to start a shiftable load, what duty to hold a plant room at. Small choices that come back identical, and their value lies in the frequency rather than in any single occurrence.

They are also the decisions taken worst without data, because memory keeps no record of the hundreds of times the choice was made out of habit. The most concrete case is switching source on hourly cost: the measurement says when it pays, and the choice stays with whoever runs the plant.

Verify the result against the baseline

After the work, the comparison is with expected consumption under the same conditions, not with last month. It is the only way to tell a saving from a mild winter, or to notice that a real saving was swallowed by an increase in production and should be counted anyway.

The baseline is frozen beforehand. Rebuilding it afterwards, with the post-intervention data already in it, produces a comparison that always agrees with whoever built it. The reference period, the normalising variables and the exclusions are written down before starting, and then left alone.

What is left at the end

• Loads ranked by weight, with the measured consumption that justifies the ranking.
• An estimate for a measure, computed on your own history rather than a sector average.
• The saving verified afterwards, normalised for conditions.
• Recurring decisions backed by a figure rather than by habit.

The fourth is the one that keeps returning long after the first three have been used once.