An automated valuation model (AVM) is software that estimates what a home is worth from data, without anyone visiting the home. It is used mainly in America, where lenders, insurers and property websites value millions of homes this way. The best-known example is the estimated price that one large property website shows next to almost every American home.
How does it work? The model starts with comparable sales, called comps: recent sales of similar homes nearby. It then adjusts for the differences between each comp and the home being valued, such as the floor area, the number of bathrooms, the age of the house and the size of the plot. It also reads public property records, the listing's photographs and, more and more, satellite and aerial images, which show the roof, the plot and any swimming pool. Reading images of places in this way is called geospatial AI. With each value, the AVM gives a confidence score, which says how sure the model is.
Consider a lender in Ohio, in the United States. A home owner applies to refinance her loan, which means taking a new loan to repay the old one at a lower interest rate. The AVM finds six sales within one mile in the last six months, at $285,000 to $330,000. Her home is a little larger than most of them, so the model values it at $312,000, with a likely range of $298,000 to $326,000 and a high confidence score. The range is narrow, and the loan is small compared with the value. So the lender's credit decision engine accepts the AVM value, and no appraiser visits. An appraiser is the American name for a valuer. If the confidence score had been low, the lender would have ordered a full appraisal.
Who uses an AVM? Lenders use it in underwriting, to decide how much they can lend against a home. Mortgage servicers use it to see what a home in default is worth today. Insurers combine it with property risk data, such as flood risk, to set the premium. Home owners use it to check their property tax assessment.
What can go wrong? An AVM is only as good as its data. If the model is wrong about one neighbourhood or one kind of house, the same error repeats across thousands of loans at once. Unusual homes, and homes with few recent sales nearby, get low confidence scores.
In India, AVMs are used much less. Before a home loan or a loan against property, lenders usually send an approved valuer, an engineer or surveyor on the lender's list, to inspect the property in person. One reason is data. In America, the sale price of almost every home is a public record. In India, the price registered on the sale deed is often lower than the price actually paid, so sale records are a weaker guide.