Why does the number look the way it does?
Tracing a KPI back to the meter
The slide says –8%. Energy performance, year on year, and the arrow points the right way. Then someone at the table asks the only question that matters. Why?
There are two possible answers. The first is that the building got better. A control strategy was retuned, and ventilation started following actual occupancy. A heat exchanger that had been failing quietly for two winters was finally replaced. The second answer is that the counting changed. A tenant’s server room went behind its own submeter and left the calculation. Charging points were reclassified. A gap in February was filled with last February.
Both produce –8%. Only one of them produces a better building.
Ask where a figure like that came from, and few reporting teams can answer before the meeting ends. That is seldom anyone’s fault. The figure arrives without a biography. It sits at the end of a long chain of decisions, and most of them were made by someone who has since changed jobs.
The layers between the meter and a reported energy KPI
A reported KPI rests on at least five layers. Somewhere in a building there is a physical meter. A measurement point claims to know which meter that is and what it covers. A data point is assembled from one or several measurement points. A calculation rule decides what counts, what is excluded and how to normalise for degree days. The same rule decides what to estimate when the signal goes quiet. Finally, a report presents the result as a fact.
Every one of those layers is an interpretation. A KPI is an argument with numbers in it, and like any argument it can be sound, sloppy or quietly self-serving.
That would be manageable if the argument were written down. Usually it lives in a spreadsheet with four years of history and one author. Sometimes it lives in a naming convention that made sense to an engineer in 2019, or in the memory of the consultant who did the submetering. The organisation keeps the output and quietly loses the reasoning behind it.
Three ways an energy KPI moves while the building stands still
Estimation comes first. Meter data has gaps: lost connectivity, replaced devices, counter resets, readings that fail to parse. Every pipeline fills them somehow. Research is open about this. A 2026 benchmark of smart-meter forecasting models describes filling gaps by linear interpolation. Where that fails, it falls back on the household’s own mean consumption.
For training a model, that is sound practice. The trouble starts when the same value travels on, unflagged, into an invoice, an energy performance certificate or a loan covenant. An estimate that looks exactly like a reading has been promoted above its competence.
Boundaries come second. Which areas count, which tenants, which processes, which vacant floors? A new submeter goes in for a perfectly good operational reason. The denominator shifts underneath the trend line, and nobody thinks to break the series.
The third is the rule itself. Regulators, lenders and auditors change their definitions, and the usual response is to rebuild the model. The rebuild produces a number consistent with the new rule. It is also inconsistent with every number published before it. History quietly becomes incomparable with itself. That is the most expensive thing a reporting chain can do.
EPBD and CSRD: the definitions you report against keep changing
Whether your buildings stand in Copenhagen, Helsinki, Stockholm or Warsaw, the same directive applies. The recast Energy Performance of Buildings Directive should have been transposed into national law by 29 May 2026. No Member State managed it in full. In July, the Commission opened infringement procedures against all 27. Article 16 of the directive gives owners, tenants and managers direct access to building system data at no additional cost.
Sustainability reporting is shifting too. The Omnibus I directive, adopted in February 2026, narrowed CSRD scope to companies with more than 1,000 employees and €450m in turnover. Many property owners lost a legal reporting obligation as a result. Their lenders, investors and large tenants are still asking the same questions.
So the definitions you report against will keep moving. The chain beneath them can hold still. With that chain intact, you can replay your own history under a new rule. The board gets a like-for-like trend in the same week the definition changes, and the auditor gets the path back to the meters.
A defensible energy KPI needs origin, hierarchy and versioned calculation rules
Our work with customers starts further back than most expect: with the questions your business has to answer. Energy performance, sustainability reporting, loan covenants, tenant billing, operational follow-up. The metering plan follows from those questions. Each meter then has a reason to exist, and you know what that reason is.
From there, three things have to hold.
Every value carries its origin. Measured, estimated and corrected are three visibly different states, each labelled in the data.
The hierarchy from physical meter to data point is explicit. A submeter installed in March becomes a documented event, with a date and a reason.
Calculation rules are versioned. Say electricity for vehicle charging has to leave the energy performance figure. You change how the data point is composed, and the path back to the physical meters stays intact. So does every earlier version of the figure.
The honest objection is that this work is unglamorous and it is not free. Nobody has ever been promoted for improving data lineage. We make the case anyway. A missing chain costs you every time a rule changes, and rules now change often.
AI raises the stakes. A model will produce a fluent explanation of why consumption fell, assembled from correlation. It will be right often enough to be dangerous. Agents will soon read these figures faster than any person can question them, and provenance becomes the scarce part.
The opportunity is just as real. With clean lineage, AI can spot an anomaly in a heating curve, trace it to the meter that caused it and explain it in terms an auditor accepts. Automating energy data is safe exactly where the data can answer back.
Which brings us back to the room, the slide and the question. The right answer is short: minus eight per cent, three points of it from a retuned heating control, five from a boundary change in March, and here is the path back to the meters. It takes two seconds to say. Everything above is about what has to exist before you can say it.
