Energy Academy
Level 1/Monitoring & Targeting (M&T)
Baselines & Energy Signatures5 / 12

Establishing Your Baseline

What a baseline is and how to set one that's valid and useful.

10 min read · Last reviewed July 2026

A baseline is the answer to the question "compared with what?". It is a model of how the site consumed energy over a known reference period, expressed in terms of its drivers, so that any later period can be judged fairly against it. Get the baseline right and every subsequent claim, from a weekly exception report to a board-level savings figure, stands on solid ground. Get it wrong and the whole M&T programme inherits the error.

What makes a good baseline period

Four qualities matter, and they pull against each other slightly:

  • Complete. At least one full year, so every season appears. A baseline built from summer months will misjudge every winter.
  • Representative. The site operating normally: no extended shutdowns, no half-finished refurbishments, no COVID-empty buildings. If an odd event sits in the period, either exclude those weeks explicitly or choose another period.
  • Recent. The site as it is now, not as it was before the extension was built or the chiller replaced. Older data may be plentiful, but it describes a different building.
  • Documented. The period chosen, the data sources, the drivers used, any excluded weeks, and the fitted coefficients, all written down. A baseline nobody can reconstruct is a baseline nobody can defend.

Twelve months is the usual choice. Twenty-four is better where the data is clean, because two winters constrain the weather relationship far better than one.

From data to model

With the period chosen, fit the consumption against its drivers, exactly as the normalisation lesson set up. For a weather-led gas meter, that means regressing monthly (or weekly) kWh against degree days and reading off two numbers: the intercept, which is the base load, and the slope, the kWh each degree day costs.

Worked example — fitting a baseline by hand
Given
  • June: 140 HDD, 22,000 kWh
  • January: 350 HDD, 43,000 kWh
  • (In practice you fit all twelve months by regression; two points show the mechanics)
Find
The slope and base load of the baseline model.

A real fit uses all the points and reports how well the line explains the data (the R² statistic). For a weather-dominated gas meter, an R² above about 0.9 says the model is trustworthy; a low R² says something else drives this meter and the model needs another driver or a rethink. The energy signatures lesson covers reading these fits in detail.

CUSUM: making the baseline work for a living

Once a baseline exists, the most useful thing you can do with it is track the cumulative sum of differences, or CUSUM. Each period, compute actual minus expected consumption, and keep a running total. The power of the technique is in its shape: random scatter hovers around zero, but any systematic change bends the line. A downward bend dates the moment a saving began and measures its size; an upward bend catches a creeping fault long before any single month looks alarming on its own.

CUSUM is the workhorse chart of M&T. It turns a pile of monthly figures into a story with dates: "flat until March, then saving roughly 2,000 kWh a month after the burner service". That is language a finance director can audit.

The baseline is a contract

Treat the baseline as a fixed reference that changes only for documented reasons: a major plant change, an extension, a change of use. If the baseline drifts casually, savings become unprovable and the programme loses the credibility that justifies its existence. The measurement and verification course covers the formal rules; the discipline starts here.

Next, energy signatures: reading the fitted model itself as a diagnostic, where the slope and intercept each tell their own story about the building.

Sources and further reading