How Ampian AI works

The annual figure comes from a machine-learning model trained on the federal Residential Energy Consumption Survey: about 13,000 American homes and the electricity they actually used. Given your home’s size, age, equipment and household, it forecasts low, typical and high electricity consumption in kilowatt-hours for the year.

Engineering formulas then work through each use (heating, cooling, hot water, appliances, EV charging, a pool, solar) to spread that year across the months, using 30 years of local climate data. Rates from the utility providers that serve your ZIP code turn kilowatt-hour usage into dollars. If more than one provider serves your area at different rates, the range covers both.

Natural gas follows the same approach as electricity. A model trained on about 11,000 natural gas homes from the same federal survey first judges whether a home like yours uses natural gas at all, then forecasts low, typical and high annual natural gas consumption.

Formulas for each natural gas use you report (space heating, hot water, the stove, the dryer, a heated garage) set the winter-heavy monthly pattern. Usage is priced at your state’s average residential natural gas price over roughly the last three years, with extra room in the range for price swings.

Oil follows the same approach as electricity and natural gas. An AI model trained on oil-heated homes from the same federal survey forecasts low, typical and high annual oil consumption from your home’s size, age, foundation and heating equipment.

Local climate then spreads that year into a winter-heavy monthly pattern, and gallons are priced at the average for your state or region. Winters and oil prices both swing, so the range for oil is wider than for other utilities. We show that spread rather than hide it.

We forecast indoor use plus pool top-ups, because there is no reliable data for estimating other outdoor water use such as lawn irrigation. We go one fixture at a time: toilets (sized to the era the home was built), showers, faucets, laundry and the dishwasher, each scaled to the people in the home, plus a small allowance for leaks. For a pool, we add the water needed to replace what evaporates. The pieces are combined so the range reflects that they rarely all run high at once.

Gallons are priced at a combined water-and-sewer rate per 1,000 gallons: the latest published rates of local utilities in several large metro areas, and state or national survey averages where we can’t find a reliable public water and sewer rate.

We estimate what a full-service lawn company would charge for mowing through your state’s growing season, a few fertilizer and weed treatments, and seasonal cleanups. The number of visits comes from local frost dates, so a long Southern season costs more than a short Northern one.

Each visit starts from typical published rates, then gets adjusted for landscaping wages in your state and for the size of your lot and lawn. Condos and HOA townhomes don’t get this cost.

We look up annual snowfall at the weather station nearest your ZIP code and place the home in a light, moderate, heavy or extreme snow zone. Homes that average less than five inches a year, and condos, typically don’t have a snow removal cost.

Each zone has a seasonal price for plowing and de-icing from published service guides, adjusted to local wages and to the size of your driveway, which we take from your garage. The cost is spread over the months when snow actually falls.

We multiply the purchase price by your county’s effective tax rate, the share of home value that owners there actually pay in tax.

In states such as California, Florida and Texas, taxes for long-time owners are capped, which keeps the county average low. A sale removes the cap, so we raise the rate to what a new buyer would pay. Some neighborhoods charge an extra tax for local services like water, roads or new schools, and we add it where it applies. Where your area offers a residential or homestead exemption for primary homeowners, we subtract it from the estimate.

We start with what homeowners in your area pay for insurance on average, updated to today’s prices. If there isn’t enough local data, we use the county or state average instead.

Then we adjust for your home. A bigger home costs more to insure. A newer roof, or brick or steel walls, cost less. Where insurers are raising prices quickly, the range is wider.

Each major part of the home (the roof, the furnace or boiler, cooling, the water heater, major appliances, pool equipment) has a typical replacement cost and a typical service life. Costs are adjusted to local construction prices, and the roof is sized from your home’s footprint.

You enter how old each of these parts is. We work out how many years each has left, estimate what replacing it will cost by then as building prices rise, and spread that over the months remaining. A part already past its expected life shows up as a one-time “replace now” cost instead of a monthly set-aside.

We find exactly where the address sits on the map. That way, every check after it, like zoning and flood risk, is about the right home.

We look up the parcel’s current zoning at the moment you run the report: the zone code, its name and type, and what it permits. That covers accessory dwelling units (ADUs), multi-family housing, commercial and industrial uses, and short-term rentals.

Zoning codes are written for real estate professionals. The report adds a plain-English zone guide explaining what the designation means for an owner, with a link to the source code. Even as a first-time buyer, you can see what is and isn’t allowed for your home, so you can ask your agent or the county better questions.

Flood is the official FEMA flood zone for the property itself, the same designation lenders and insurers use, translated into Minimal, Moderate or High. Where FEMA hasn’t mapped the area, the report says Undetermined rather than calling it low.

Wildfire comes from the U.S. Forest Service’s hazard map of the surrounding landscape. Earthquake, wind, heat, cold, drought, hail and other hazards come from FEMA’s ratings for your neighborhood, which score each risk from very low to very high compared with other areas across the U.S.

We show the closest public schools, along with other well-rated schools nearby, including private and charter schools. Each one comes with its GreatSchools rating and how far it is from the home.

We also list supermarkets and public transit stops within five miles, closest first.

Zoning tells you what is allowed. Development news tells you what is actually happening. An AI search scans an approved list of reputable real-estate, business and transit news outlets, plus government sites, for stories about new projects from the past 12 months.

A story is kept only if its headline names your town, a neighboring town, the county or a nearby neighborhood, and the report shows up to five. Stories about canceled or denied projects stay in the list but don’t count toward the verdict.

Price history. The range of home-price change over the past 5 and 10 years, using your ZIP code where data exists and otherwise the metro area or state. The broader the area, the wider the range.

Rental yield. What a typical two-bedroom in the area rents for over a full year, divided by what homes there are worth. For example, $30,000 a year in rent on a $600,000 home is 5%. Below 5% is Low, 5% to 8% is Moderate, and 8% or more is High.

Neighborhood trajectory. Changes in income, home values and education across recent Census surveys, ranked against other neighborhoods in the same metro, give a High, Moderate or Low reading.

The verdict is calculated, not written. Each factor earns points on a fixed scorecard: price history, rental yield, natural hazards, nearby schools, new development and neighborhood trajectory. The total decides the verdict: Strong Buy, Buy, Consider, Neutral, Pass or Strong Pass.

Some safety rules override the score. If a natural hazard is rated very high, the verdict can’t go above Neutral. If there isn’t enough data, the report doesn’t show a verdict at all. Where a source has no data, the report says Not available instead of guessing.

AI then writes the plain-English explanations. Automatic checks remove any sentence that invents a figure or a place, softens a high climate risk, or argues for a different verdict.