Energy Estimation Methodology

Availability: estimation is an optional platform capability and may not be included in every subscription. Portfolios without estimation enabled report observed meter data only.

Estimation is performed on the data available at the time of reporting; estimates are refreshed as new observed data is received.

Cambio’s estimation policy is applicable for standing operational assets. Where utility data contains gaps due to temporal interruptions or spatial coverage limitations, Cambio applies a systematic, deterministic estimation methodology to produce complete, building-level energy consumption profiles. Data completeness assessments and estimations are performed independently for each energy stream (electricity, fuel) at each property and ahead of any analytics, calculation of GHG emissions, or data export.

All estimates:

Estimation is deterministic and weather-adjusted: it responds to observed local climate conditions and building-specific or cohort-derived consumption patterns rather than fixed benchmark values or static intensity assumptions. Data quality issues are handled through the alert and validation workflow, not inside the estimation logic. The underlying model is a five-parameter piecewise linear (change-point) model that captures the relationship between outdoor temperature and energy consumption across three operating regimes: heating, baseload, and cooling.

Estimation is triggered independently for two model types: Electric and Other (fuels + district energy, combined).

Missing meter type. If a property is expected — per building onboarding — to have a given meter type and none exists in the data at all, estimation is triggered for every day in the reporting period for that model type. This check runs before, and independently of, the check below.

Area × Time Coverage. For each day in the reporting period, a coverage score is computed per energy subtype (Electric, Fuels):

Based on available onboarded data at the time of estimation, the system applies one of three estimation pathways based on data availability and quality, described below. The applicable pathway is determined for each property at the time of estimation based on available historical data.

Applied when a property has ≥ 9 months of historical data spanning multiple seasons and passes quality validation gates of having > 75% spatial coverage.

A set of property-specific change-point regression models (electricity and a fuel model if applicable) is developed using:

The model learns heating and cooling balance points and relates consumption to heating and cooling degree days. This approach captures building-specific characteristics, including:

Model quality is monitored against ASHRAE Guideline 14 metrics: Normalized Mean Bias Error (NMBE) and Coefficient of Variation of Root Mean Square Error CV(RMSE). This pathway is always preferred when data sufficiency criteria are met

Applied when property-specific data are insufficient or fail coverage thresholds. Consumption is estimated using averaged parameters from a cohort of similar buildings, selected based on:

Models with high CV(RMSE) scores are not eligible for use in the cohort analysis.

Cohort size has a minimum threshold of 5 buildings, selected from a database of millions of building energy models. The methodology prioritizes exact matches (identical climate zone, asset class, comparable floor area) and selects all available exact matches if ≥ 5 exist. When fewer than 5 exact matches are available, it fills the cohort with the next most similar buildings to reach the 5-building minimum. Weather-normalized parameters are averaged across the cohort.

When cohort-based estimates are used for all fuel types, the total modeled EUI is calibrated to regional benchmark data. The benchmark alignment parameter (α) is set per client based on confidence in regional benchmark data quality. Where high confidence exists across all geographies, α=1 is applied, meaning estimates align fully to regional benchmarks. Single-fuel cohort estimates or other estimation methodologies do not incorporate benchmark blending.

The calibration formula is: S = (1-α)·P + α·X, where P is the model annual total, X is the benchmark annual target, and S is the benchmark-calibrated result. For α=1, this simplifies to S = X. A correction factor (k) is applied to preserve monthly weather-responsive patterns: k = S / P

This approach ensures estimates align with regional performance expectations while preserving the weather-responsive temporal distribution characteristics of the cohort model.

The resulting estimated Energy Use Intensity (EUI) is compared against regional benchmarking data for the same country and asset class. The estimate is then weighted toward this regional performance expectation to ensure alignment with observed market norms. The cohort approach uses validated consumption patterns from comparable assets that have met self-model quality thresholds on their own historical data and resulting coefficients to generate estimates applied to the subject property's observed weather conditions

For supported jurisdictions (UK, EU), cohort-based estimates are adjusted using Energy Performance Certificate (EPC) ratings. These adjustments apply a bounded multiplicative factor derived from country-specific EPC intensity tables to align estimates with jurisdictional performance classifications.

Once estimation is triggered for a given month and energy subtype and a model has been selected, the system applies a single gap-filling calculation, regardless of whether the coverage shortfall stems from missing floor-area coverage, missing calendar days, or both. The approach preserves observed data, maintains auditability, and ensures whole-building totals reconcile correctly.

This single calculation applies uniformly across "spatial" and "temporal" gap types: a day with no observed data reduces to the whole-building prediction standing alone (observed = 0), a day with partial floor-area coverage nets out only the unmetered remainder, and a day affected by both is resolved by the same subtraction — no separate logic branch is required.

Once estimation is triggered for electricity and fuels separately and model selection has occurred, the system applies one of three gap-filling strategies depending on the nature of the coverage shortfall. The approach preserves observed data, maintains auditability, and ensures whole-building totals reconcile correctly.

Electricity, fuels, and district energy are modeled independently. This means a single property may use self-model estimation for electricity and cohort-model estimation for fuels, based on data availability for each stream.

Total Energy equals the sum of all three streams and represents whole-building energy consumption

Every estimate includes complete provenance metadata:

When new observed meter data become available, overlapping estimates are invalidated and regenerated. All estimation logic is deterministic and reproducible given identical inputs.

Cambio uses benchmarks for two distinct purposes, and each benchmark is labeled by use case. Benchmarks used for variance alerts are review prompts against comparable records; benchmarks used for cohort calibration determine estimated values where actual consumption is missing or incomplete.