A combination of factors is rewiring the U.S. real estate industry. The restructuring of compensation for real estate agents, triggered by a legal settlement involving the National Association of REALTORS, is giving artificial intelligence an opening.
AI will likely change how you buy, sell, and finance a home. And perhaps even save you money.
Using AI to buy a house
When people get into house-hunting mode, most turn to home-search websites such as Zillow, Realtor.com, and Redfin. AI will likely soon play a larger role in assisting buyers on these sites, though one early mover that touts a combination of AI and human assistance is Homa.
Currently available only in Florida, Homa will be in Texas within weeks and expects to add California and Georgia by the end of the year.
Arman Javaherian, the founder and CEO of Homa, was a senior director at Zillow for six years.
“I basically built what’s called the Zillow agent platform, and this is the platform that connected the millions of home buyers with hundreds of thousands of real estate agents every month,” Javaherian told Yahoo Finance. “Zillow specifically makes 90% of their revenue from those buyers’ agent commissions.”
A lawsuit settlement is changing the industry
The National Association of REALTORS antitrust lawsuits settlements, which went live in August 2024, are changing the industry, Javaherian said. Home buyers have to sign a contract with a buyer’s agent before they can even tour a home. And agents often stipulate the full 3% commission, “because if they put anything lower, but the seller is paying more than that amount, they would have to forfeit it,” he added.
Acting as an AI-powered buyer’s agent, Homa takes 1% of the buyer’s commission and refunds the rest.
An example: On a $500,000 home, if the seller is paying a 3% commission, Homa takes 3%, or $15,000. Homa keeps $5,000 and refunds $10,000 to the buyer. That $10,000 can be used to lower the buyer’s closing costs, reduce the listing price, or buy down the mortgage rate.
An ‘Uber’ network of agents
Buyers can search for houses on the Homa website, and when they want to see a house in person, Homa uses a network of local showing agents to conduct the walk-through.
“It’s kind of like an Uber network of agents for showings. But it turns out that that’s one of the things buyers love the most,” Javaherian said. “They don’t feel guilty about taking up their agent’s time, which we heard a lot from our buyers. They can see eight homes in one day with eight different agents.”
The local agents are knowledgeable about school districts and neighborhoods. Meanwhile, Homa’s AI answers questions about the offer price and comparable sales and analyzes documents, such as disclosures.
Reducing possible conflicts of interest
When a buyer is ready to make an offer, Homa’s in-house agents provide pricing strategies and handle all negotiations. As the purchase moves to escrow, a dedicated transaction coordinator manages all the paperwork.
“With the traditional agent, the more expensive the house is purchased, the more money that agent is making,” and that can result in a conflict of interest, he added. Homa agents are paid a flat fee. “They don’t make more money if you pay more as a buyer.”
He noted that the real estate industry has been slow to change because brokerages “have very little control over the agents themselves. They’re all 1099 contractors. So if Compass goes and tells their agents, ‘Hey, we’re going to reduce our commission to 1%,’ all their agents will be like, ‘No way, I’m leaving. I’m going to Century 21.'”
Using AI to sell a house
The recent story of a Florida man selling his house with the help of ChatGPT made big news in March. Robert Levine of Cooper City, Fla., said using AI saved him tens of thousands of dollars.
“We repainted a couple of rooms in the house because ChatGPT said, ‘That’s where you’re going to get the biggest return on investment,'” Levine told NBC 6 South Florida.
ChatGPT helped him design marketing materials and the online listing and revealed how to get his home listed on the Multiple Listing Service (MLS).
“I think within the first 72 hours we had five offers on the home already,” Levine said. Within five days, he had a signed contract.
AI can also “virtually stage” a home with sample furnishings, enhance photos (sometimes to a misleading extreme), and research your local real estate market for pricing data.
Beycome is an “AI-guided,” flat-fee MLS listing service that claims you can save up to 6% by putting your house on the market with its platform. For $99, the company will list your home on search sites and MLS, with title and escrow services costing an additional $99.
Will AI replace real estate agents?
Will AI be the end of real estate agents? Homa’s Javaherian doesn’t think so: “We have real estate agents; we need real estate agents because only a licensed real estate agent can do that negotiation. Will their role change? Hopefully, with technology, yeah, their role can change, and as a result, home buyers can save.” And in the Homa model, agents can work from home.
Javaherian noted that outside of the U.S., the average total commission on purchases and sales is commonly about 2%. It’s 6% in America because of the previous structure of the real estate industry.
AI in mortgage lending
AI is also creeping into mortgage lending.
Vishal Garg, CEO of Better, proudly proclaimed, “We are an AI native company. We built this machine-learning rules engine called Tinman, starting in 2015, to basically take over every part of the mortgage process, and have what could be done by machines done by machines.”
Garg said large language models are layered on the Tinman platform as “orchestration,” which provides “confidence that all the math is right all the time.”
AI is hard to implement in the mortgage industry because “LLMs natively have about a 50% error rate on calculations. If you just take ChatGPT and say, ‘Hey, underwrite this loan according to a set of guidelines’ — they’ll get the math wrong 50% of the time. On something that’s complicated, like a DTI calculation, which might require a couple of recursive loops, you’re talking multiplying .85 to the power 14. So, literally, the error rate ends up being close to 70, 80, 90%. With a mortgage, you have to be 100% right all the time.”
Garg said Better has “zero errors” with the Tinman platform.
“The machine learning engine does all the calculations and rules application. The AI agent just orchestrates it. That’s why our cost to produce a mortgage is close to $2,000 alone, and everybody else is more like $10,000.”
