Definition
Ratio of AI-modeled tenant-level carbon attribution to actual metered consumption for the subset of tenants with sub-metering, extrapolated as an accuracy indicator for non-metered tenants.
The Measurement Gap
There is no metric that validates AI-modeled carbon attributions against physical measurement, which means tenant-level carbon charges and green lease adjustments are based on model outputs that may not correspond to actual energy consumption patterns.
Sustainability platforms produce tenant-level carbon attribution using allocation models: distributing building-level emissions across tenants based on square footage, operating hours, equipment inventories, and usage patterns. These attributions drive real financial decisions: green lease premium adjustments, tenant improvement funding for efficiency upgrades, regulatory penalty allocation, and ESG reporting at the tenant level. But the attribution model's accuracy is rarely verified against physical measurement.
The gap is structural. Building-level carbon reporting is validated against utility bills: the total number is checkable. Tenant-level attribution is a model output that divides that total among occupants. The division method introduces error that compounds when the model's assumptions about tenant behavior diverge from reality. A tenant with a dense data closet consuming disproportionate energy may be attributed average consumption based on square footage. A tenant with efficient lighting and minimal plug loads may be over-attributed. The errors flow directly into financial allocations.
Where sub-metering exists, it provides ground truth. CAAI uses that ground truth to calibrate the model's accuracy for the metered subset and extrapolate confidence for the non-metered majority.
Decision Supported
Are tenant-level carbon allocations accurate enough to base green lease adjustments, tenant improvement investments, and regulatory penalty apportionment on, or does the attribution model need recalibration before financial decisions are made from its output?
Formula
Plain Language
For every tenant with sub-metering, compare what the AI model said that tenant's emissions were to what the sub-meter measured. Average the error. The result indicates how much to trust the model's attributions for tenants without sub-metering.
Notation
CAAI = 1 − (1/N) × Σ |Attributed_emissions_i − Metered_emissions_i| / Metered_emissions_i
Variables
| Variable | Definition | Unit |
|---|---|---|
| Attributed_emissions_i | AI-modeled carbon emissions for sub-metered tenant i | kgCO2e |
| Metered_emissions_i | Actual carbon emissions calculated from sub-meter energy data for tenant i | kgCO2e |
| N | Number of sub-metered tenants in the portfolio | Count |
Units: 0 to 1. A value of 1.0 indicates perfect attribution accuracy. Track directional bias separately: persistent over-attribution or under-attribution for specific tenant types indicates a structural model flaw, not random error.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Are tenant-level carbon allocations accurate enough to base green lease adjustments, tenant improvement investments, and regulatory penalty apportionment on?