Energy Academy
Level 1/Monitoring & Targeting (M&T)
Targeting & Reporting10 / 12

Exception Reporting: Finding the Anomalies

How to spot when consumption is higher than it should be and investigate.

10 min read ยท Last reviewed July 2026

An M&T programme lives or dies by one weekly habit: comparing what each meter recorded against what the baseline said to expect, and investigating the meters that disagree. That routine is exception reporting. Done well, it is short, boring, and occasionally worth thousands of pounds in a single line. Done badly, it either cries wolf until nobody reads it or stays so quiet that faults run for months. The difference is a threshold chosen with a little statistical care.

Expected, actual, and how far apart is "wrong"

For each meter and period, the baseline model produces an expectation given the period's actual conditions. The residual, actual minus expected, will never be exactly zero: buildings are noisy. The question is how big a residual is worth a person's time.

The standard answer borrows from statistical process control. When the baseline was fitted, the scatter of the historical residuals gives a standard deviation, ฯƒ. A residual within ยฑ2ฯƒ is weather, noise, and ordinary variation; flag anything beyond it. At 2ฯƒ, roughly one clean period in twenty will false-alarm, which is a tolerable price for catching real faults quickly. Sites drowning in alerts can move to 2.5ฯƒ or add a persistence rule such as flagging only when two consecutive periods exceed the limit.

Worked example โ€” is this month an exception?
Given
  • Baseline model: expected gas this month (at its actual 350 HDD) = 43,000 kWh
  • Actual consumption: 49,500 kWh
  • Standard deviation of baseline residuals: 2,000 kWh
Find
Whether this month should be flagged.

Triage: from flag to finding

An exception is a symptom, not a diagnosis. A practical triage sequence for a flagged meter:

  1. Check the data first. Meter faults, estimated reads and unit errors produce spectacular exceptions. Rule them out before dispatching an engineer.
  2. Ask what changed operationally. New equipment, changed hours, a production trial, a heating season switch-on. Operations often know the answer in one conversation.
  3. Use the shape of the deviation. Half-hourly data narrows the suspect list fast: an overnight rise points to something left running or a control failure; a working-hours rise points to load or setpoint changes; a step change dates the event precisely, which usually identifies it.
  4. Close the loop. Record the cause and the fix against the exception. A programme that logs its findings builds a site-specific fault library, and next year's investigation starts from experience instead of scratch.

Low exceptions deserve the same attention as high ones. Consumption far below expectation sometimes means a genuine improvement, but it just as often means a failed meter, an estimated read, or a heating system that quietly failed in a way occupants will notice on Monday.

Only exceptions get attention, and every exception gets attention

The discipline cuts both ways. Meters inside their control limits get no meeting time, however interesting their charts look; meters outside them always get a named investigator and a deadline. This is what keeps the routine short enough to survive contact with busy people, and reliable enough that a flag actually means something.

Multi-site organisations add one more layer: a league table of sites by normalised deviation from baseline, which concentrates central attention on the outliers and creates a useful, mildly competitive pressure among site managers.

The final lesson in this course turns to communication: dashboards and reporting that turn all of this machinery into something the wider organisation can see, trust and act on.

Sources and further reading