AI in Real Estate Valuation and the iBuying Struggle
This article explores how Automated Valuation Models (AVMs) utilize artificial intelligence to estimate property values and examines why algorithmic home-buying programs failed when confronted with high market volatility. By analyzing vast amounts of property data, AI can generate instant home valuations, revolutionizing the real estate industry. However, as recent high-profile corporate exits have shown, relying solely on these algorithms to purchase and flip homes carries significant financial risks when economic conditions shift unpredictably.
How AI Powers Automated Valuation Models (AVMs)
Automated Valuation Models use machine learning and statistical modeling to determine the current market value of a home without a human appraiser. To do this accurately, AI processes several distinct layers of data:
- Property Characteristics: The algorithm looks at public tax records and Multiple Listing Service (MLS) data to evaluate square footage, lot size, the number of bedrooms and bathrooms, and the age of the home.
- Comparable Sales (Comps): The model identifies recently sold properties in the immediate vicinity with similar traits, adjusting the target home's value based on the differences in those sales.
- Computer Vision: Advanced AVMs use AI to scan listing photos. The software can identify whether a kitchen has granite countertops, hardwood floors, or outdated appliances, assigning a "condition score" to the property that impacts its final valuation.
- Neighborhood and Economic Trends: The AI analyzes hyper-local data, including school district ratings, crime statistics, walkability, and broader macroeconomic indicators like mortgage interest rates.
By combining these data points, an AVM can generate a valuation in seconds. Banks, real estate platforms, and consumers use these models as a baseline for home equity loans, refinancing, and setting listing prices.
The Rise of Algorithmic Home-Buying
The speed and perceived accuracy of AVMs gave rise to "iBuying" (instant buying). Companies like Zillow, Opendoor, and Redfin used these AI models to make instant cash offers on homes directly to sellers. The business model was straightforward: use the algorithm to buy a home at a fair price, make minor repairs, and quickly resell it for a small profit plus a service fee.
Why Algorithms Struggled with Market Volatility
While AVMs excel at estimating values in stable markets, algorithmic home-buying programs suffered billions of dollars in losses when the market became volatile. The failure of Zillow Offers in late 2021 highlighted the core vulnerabilities of relying on AI for high-stakes real estate investments.
1. Reliance on Backward-Looking Data Machine learning models predict the future based on historical data. In a stable market, sales from the past three to six months are reliable indicators of current value. However, when the real estate market experiences sudden shocks—such as the pandemic-induced housing boom followed by rapid interest rate hikes—historical data becomes instantly obsolete. The algorithms could not adjust fast enough to the sudden cooling of buyer demand, causing companies to overpay for thousands of homes.
2. The Nuance of Local Real Estate AI struggles with the hyper-local, qualitative aspects of real estate. An algorithm works well in a homogeneous suburban subdivision where every third house has the same floor plan. It fails in diverse neighborhoods where value changes drastically from one street to the next. Furthermore, an AI cannot smell pet odor, hear highway noise from the backyard, or sense a neighbor's dilapidated property—factors that heavily influence what a human buyer is willing to pay.
3. Adverse Selection Algorithmic buyers fell victim to a phenomenon known as adverse selection. Suppose an AI makes offers on 100 homes. It will naturally undervalue some, price some correctly, and overvalue others. Homeowners who received low or fair offers often rejected them, opting to list on the open market instead. Homeowners who received accidentally inflated offers jumped at the chance to sell. Consequently, the algorithms disproportionately acquired overpriced homes, loading company balance sheets with toxic inventory.
4. Supply Chain and Labor Bottlenecks The iBuying model requires flipping homes quickly to minimize holding costs like property taxes, utilities, and loan interest. During periods of high volatility, supply chain disruptions and labor shortages delayed renovations. While these homes sat waiting for repairs, the market shifted downward. The algorithms did not properly factor these operational bottlenecks into their initial pricing models, turning projected slim profits into massive losses.