🧠 Zillow Tried to Replace Realtors

Subscribe | Contact | Meet The Team

Welcome back AI prodigies!

In today’s Sunday Special:

• 🕵️‍♂️Who and What

• 🏠How and Why

• 🧯Aftermath

Read Time: 5 minutes

🎓Key Terms

• Concept Drift: an evolution of the data's target variable (i.e., what the model attempts to predict) that invalidates the machine-learning model.

• Supervised Machine-Learning: an algorithm that predicts the future based on exactly what happened in the past through learning the relationship between input and output.

• Liquidity: when an asset can be quickly bought or sold without dramatically impacting its market price.

🕵️‍♂️WHO AND WHAT

In Fall 2021, Zillow fired 25% of its workforce, following a share price decline of nearly 70%. Meanwhile, the rest of the software industry was hiring at record levels, and the residential real estate market boomed as mortgage rates hit multi-decade lows. The U.S. recorded 6.2 million home sales per month, a multi-year high. As a facilitator of transactions, Zillow should have been thriving. It was. Zillow Offers, a predictive algorithm that found undervalued homes, allowed Zillow to flip homes through a cash offer, renovation, and rapid sale. For 18 months, Offers was the engine of Zillow’s growth, contributing 50% of its 2020 revenue. What happened?

Simply put, Zillow’s algorithms overestimated the value of the homes for which they paid. At one point, two-thirds of the homes that Zillow purchased were underwater. At the same time, Zillow aggressively expanded its purchasing program, acquiring more homes in Q2 and Q3 of 2021 than it had in the prior two years. Since holding inventory is costly, the company attempted to sell 7,000 houses in November 2021 to recoup $2.8 billion.

🏠HOW AND WHY

The exact causes of the Zillow Offers meltdown are not public. However, Zillow’s algorithms failed to adjust for the housing market cooldown. In machine-learning terms, concept drift occurred. Supervised machine-learning models, like humans, often erroneously assume that the past equals the future. When predicting home prices, this model presumed that a similar home profile (e.g., square footage, bedrooms, bathrooms, amenities, etc.) in 2020 would be priced similarly in 2021. However, several factors beyond the purview of Zillow’s 100 pricing analysts produced an unforeseen market environment.

1. Management Overreaction: in Q1 of 2021, Zillow’s most AI-forward competitor, Opendoor, bought more homes. In response, management added a “gross pricing overlay,” hiking offers by as much as 7% to woo sellers.

2. Rising Carrying Costs: Zillow’s inventory dramatically increased due to longer renovation times during the pandemic’s lumber and contractor shortages. As a result, they paid more property tax, insurance, and property management fees.

3. Mean Reversion: Targets in pandemic hotspots, including Sunbelt cities like Phoenix, Austin, and Nashville, had already experienced substantial price increases in 2020 and early 2021. When the COVID-19 vaccination rate reached 50% in May 2021, the purchasing rate in those areas declined, whereas the algorithm continued to expect increases. As a result, Zillow paid, on average, $65,000 more than the median home price for Phoenix targets.

4. Unproven Business Model: Zillow relied on selling target properties at a premium to recoup the value of its offer and carrying costs. However, only 10% of buyers who accepted offers sold their homes to Zillow.

Despite making up 50% of revenue, Zoom Offers failed to turn a profit. In the words of Zillow CEO Rich Barton, Offers turned the once-profitable housing platform into a leveraged trader.

🧯AFTER MATH

Could Zillow have averted disaster? Likely not. Fine-tuning machine learning models in real-time allows them to reflect changing market conditions, and competitors like Opendoor detected the cooling housing market and reflected this by hardcoding downward adjustments into their models. However, the limitations of algorithmic home-buying are more fundamental.

The business model relies on speed, requiring high liquidity or significant demand to make selling easier. Even in markets with high liquidity, such as coastal California, the model fails to account for nuances that buyers consider, such as noise level, whether surrounding houses are well-kept, or architectural style. And even if the model captures critical decision factors, the speed of sale required to minimize carrying costs and turn a profit requires Zillow to forgo traditional quality checks like walk-throughs and inspection processes. A fourth limitation is that sellers with flawed properties may feel incentivized to solicit an offer because the average price determined by the algorithm is still significantly higher than that of a local, knowledgeable buyer.

Therefore, home-buying algorithms show the most promise in liquid markets, with cookie-cutter, newer homes less likely to require serious repairs. Nevertheless, Zillow’s iBuyer debacle shows the limitations of programming solutions for nuanced, human-intensive decisions. For now, most real estate agents are safe from AI replacement.

📒FINAL NOTE

If you found this useful, follow us on Twitter or provide honest feedback below. It helps us improve our content.

How was today’s newsletter?

• ⭐️⭐️⭐️⭐️⭐️Nailed it!

• ⭐️⭐️⭐️Average

• ⭐️Epic fail.

❤️AI Pulse Review of The Week

🎁NOTION TEMPLATES

🚨Subscribe to our newsletter for free and receive these powerful Notion templates:

• ⚙️150 ChatGPT prompts for Copywriting

• ⚙️325 ChatGPT prompts for Email Marketing

• 📆Simple Project Management Board

• ⏱Time Tracker