Skip to content

Methodology

How we model a cleaner hour.

The homepage says the 35% reduction figure is a modeled estimate. This page is the rest of that sentence: the real data sources, the calculation, and where it still falls short of a measured result.

01 / What “Modeled” Means

What “modeled estimate” actually means

The “up to 35% reduction” and “~$350/yr” figures on the homepage are outputs of a simulation, not measurements from real households. We take a household's baseline hourly electricity use, hold it against real grid-carbon-intensity data, and calculate what emissions would look like if flexible loads, such as EV charging, a dishwasher or a water heater, shifted into cleaner hours instead. No household has run this yet, so there's no measured result to report alongside it.

That distinction matters more to a utility partner or researcher than to a homeowner deciding whether to join a waitlist, so it gets one line and a footnote on the homepage. Here is the rest of it: the actual data sources and the math behind that estimate.

02 / Data Sources

Published data for a modeled result

We use these published sources to build the model. No GreenAI field deployment has measured the result.

01Real grid demand & generation

CAISO Today's Outlook

CAISO publishes hourly demand and fuel-mix history for California. The homepage chart uses hourly average MW from June 15 2026 and clamps small negative overnight solar readings, which reflect sensor noise, to 0 for display.

caiso.com/outlook/history

02Household load profiles

NREL ResStock

NREL ResStock provides simulated hourly electricity-use profiles for representative U.S. homes. The model uses a profile to build the baseline for EV charging, HVAC, water heating and appliances.

resstock.nrel.gov

03Hourly carbon intensity

Electricity Maps

Electricity Maps estimates grid carbon intensity by hour. The model compares each candidate schedule with those estimates and selects hours with lower carbon intensity.

electricitymaps.com

03 / The Calculation

The calculation in four steps

  1. 01

    Baseline load profile

    The model takes a household's hourly electricity use from a NREL ResStock profile, including EV charging, HVAC, water heating and appliances.

  2. 02

    Join to carbon intensity

    The model matches each profile hour with Electricity Maps carbon intensity for that hour.

  3. 03

    Search shifted schedules

    The simulation moves flexible loads to lower-intensity hours within each device's comfort window and deadline. An EV must still charge by 7 AM, and the water must stay hot.

  4. 04

    Compare baseline vs. shifted

    The model recalculates emissions for the shifted schedule and compares that total with the baseline to estimate the reduction.

The model multiplies load by carbon intensity for each hour h, then sums the hourly results:

emissions = Σh load(h) × intensity(h)

The model divides the difference between baseline and shifted totals by the baseline:

reduction = (baseline − shifted) / baseline

04 / Assumptions & Limits

Assumptions and limits

The honest version of this page names what the model does not yet account for, alongside what it does.

  • 01A single simulated household archetype and climate zone stand in for the baseline load profile. It is not an average across home types or regions, so real results will vary by home.
  • 02The load and carbon-intensity series behind this estimate are historical snapshots rather than live or year-round data. The homepage chart is a single real day, June 15 2026.
  • 03Each simulated device is scheduled against one fixed set of comfort and deadline assumptions, not something you can see or tune in this estimate.
  • 04No field validation yet. Every figure here comes from simulation against historical published data, not from measured bills or emissions in real GreenAI households.

One caveat worth stating plainly: the homepage chart uses real CAISO (California) grid data, while the homepage's “~$350/yr” figure is an average New Jersey savings estimate, drawn from a different grid. California's data is the best available proxy today, and New Jersey-specific modeling is in progress. Until it lands, read the two figures as illustrating the same mechanism, shifting load to cleaner and cheaper hours, rather than as one consistent regional model.