Jenny Morant has had a tough year. In Manhattan Beach, California, where she works as a real estate agent, it’s been impossible to sell a house. What’s driving the market right now? “We’re down to divorces, maybe marriages, maybe babies, and job relocation,” Morant says.
Interest rates are largely to blame: After the Fed raised rates to 5.5% in July and then held them there, the mortgage rate soared. In October, the national average for a 30-year fixed mortgage was 7.86%. In California, it was more than 8.5%. Buyers who need a mortgage don’t want to pay that kind of premium. Potential sellers are reluctant to put their houses on the market knowing that they might not be able to afford to buy a new place. Even all-cash buyers are constrained by the lack of inventory on the market.
Compass, Morant’s brokerage and the most tech-forward real estate firm in the U.S. as well as the largest by sales volume, thinks it has the answer to these challenges: artificial intelligence. Its cofounder and CEO, Robert Reffkin, has long envisioned machine learning as a transformative tool for one of the most real-world businesses there is. “For years, we’ve had a focus on AI,” he says. “Reducing the amount of steps for different stakeholders—primarily agents—is the way we measure our success. AI is making us more efficient.” Then OpenAI’s ChatGPT turbocharged that effort. “We recently launched the beta for our integration of OpenAI’s ChatGPT-4 capabilities directly into the Compass platform,” COO Greg Hart told investors in August.
In the last year, a lot of companies have been talking like Reffkin and Compass. But to date, machine learning has never helped Compass turn a profit. In 2022, it lost $601 million on $6 billion in revenue. In the first half of 2023, revenue declined 28%, though Compass has narrowed its quarterly deficits and was cash-flow positive for the first time in the second quarter. Compass’s shares, meanwhile, have tumbled from a height of $18 when it went public in April 2021 to just over $2 in late October.
This past summer, Reffkin redoubled his commitment to harnessing AI to remake realty, organizing a company-wide hackathon to develop new generative AI ideas that could help its more than 28,000 agents in nearly 500 locations enlist Compass AI—as the company calls it—for everything from marketing listings to completing the paperwork on a sale. “We asked our agents for the best ideas to integrate AI into the [Compass] platform,” Reffkin says. “We want to know what’s interesting to them.”
Industry observers, like investors, remain skeptical. “If you think about all the human activities and all the industries and all the businesses out there,” says Mike Delprete, a real estate tech strategist and scholar-in-residence at the University of Colorado Boulder, “real estate is probably last—or next to last—on the list of those that are going to be revolutionized by AI.” The business of selling houses, after all, is inherently messy, emotional, and human. Can AI really generate what Compass needs?
Compass has always pitched itself as more than a luxury real estate brokerage. It raised $1.5 billion as a startup on the premise that it was actually a tech company. At its inception more than a decade ago, Reffkin imagined that Compass could be a sort of Bloomberg terminal for real estate: It would create a suite of data-driven tools that agents across any brokerage could purchase to give them an advantage.
The company spent most of the 2010s building and talking up its technology platform as a way to woo the country’s best agents and brokerages to join Compass. Then, in 2018, it unveiled “Powered by Compass,” a version of the platform that could be licensed by other agents, beginning with some 200 in Massachusetts. Compass saw it as a new line of business that could accelerate its takeover of the industry.
But Compass’s own brokers rebelled. Agents didn’t want to lose exclusive access to Compass’s tech, and they feared becoming ancillary to the business if it transformed into mostly a software company selling marketing and analytics tools. The initiative was dead two weeks after it was announced.
In 2020, Compass released an AI-centric product called Likely to Sell (LTS). The tool compiles data from property prices in different areas to figure out who might be convinced to sell their home in the next 12 months. Brokers can even use a prewritten template to craft nudging emails. “No other brokerage has anything like Likely to Sell,” Reffkin said during his August 2023 earnings call. “It’s generated more than $100 million of incremental revenue in the last two years.”
What Reffkin didn’t say was that Compass generated approximately $12 billion in revenue during that period, meaning that Likely to Sell contributed less than 1% to the business. “Likely to Sell is a really neat tool,” says Demetra Kentrotas, chief of staff to two New York–based agents who have been with Compass for a couple of years. “But in terms of actual prospecting, [LTS] can help but it takes a lot of work. So [agents] still rely on traditional methods for the most part.”
Compass has spent approximately $1.5 billion on its technology stack, according to Reffkin, and he says the company anticipates spending $100 million annually on R&D. Not all of that is AI-focused, but Compass has also been bullish on features that use AI (with some human assistance) to review compliance paperwork before a property can be sold. “The impact of these tools is starting to be felt,” Rory Golod, Compass’s SVP of growth and communications, told The Real Deal last year. “That will eventually manifest itself in financial outcomes.”
Even as Reffkin and his leadership team tout Compass’s technology, the company acknowledged last year that it had entered a new phase, trimming its tech team from 1,500 people to about 700. Curiously, Compass “decided to part ways” in August 2022 with one of its biggest AI evangelists, CTO Joseph Sirosh, who previously ran AI at Microsoft before spending almost four years at the company. A Compass spokesperson says, “Last year, Compass realized its mission to build an end-to-end platform and made the decision to retain members of the product and engineering team deemed appropriate to sustaining and enhancing the platform.”
Earlier this year, Compass reportedly started to train its agents to use ChatGPT. (The company has learned from past experience not to cut out brokers from its tech decisions.) First up was a plug-in that promised to write property descriptions.
Golod demoed the tool for me, showing how an agent might type in the prompt “write a description for a sunny three-bedroom in Tribeca with a large primary bedroom” and enter details of the property. Within seconds, the application returned a description: “Three-bedroom on the corner of Duane and West Broadway with a huge primary bedroom that is very sunny!”
No wonder, perhaps, that agents had other ideas they’d like to see.
On a Tuesday morning in mid-August, eight teams of engineers—145 people in all—are gathered across a handful of conference rooms in Compass’s Fifth Avenue offices in New York and on Zoom for the AI hackathon. They’re midway through a 48-hour sprint, and the scene is relatively serene. It’s nothing like in the movies. I don’t see discarded food wrappers littering rooms or hoodie-wearing employees huddled sleeplessly together. Instead, many participants are remote and conferencing in to antiseptic, white meeting spaces. A marketer breaks the silence, making sure the engineers know that, unlike them, she doesn’t plan on staying up too late.
Patrick Hulce, a senior staff software engineer based in Dallas who flew in to be here, worked until midnight the night before on a prototype of a new GPT-powered marketing dashboard. Called Ignite, it would give agents AI-generated action items as well as content recommendations for each property. If it works, agents would no longer need to develop custom marketing plans. The AI could decide which types of flyers and emails they should send, and in which publications they should advertise. When I ask Hulce about his progress, he laughs nervously. “The deadline is midnight, so I guess I’ll sleep tomorrow. It’s a little bit exhilarating, even though it’s stressful.”
Elsewhere in the office, a rival team is working to improve the text generator that Golod showed me. The update they’re working on—project name LoREm Ipsum—will let agents select a tone (transparent, friendly, realistic, etc.) and generate copy in that vein. A third team is building an application that would check documents that agents submit for errors, eliminating the typical back-and-forth between brokers and the accounting department—and helping them get paid more quickly. Agents occasionally hop on the phone with teams to talk about features they would like to see added to the prototypes. (Compass execs declined to let me meet the other teams, saying it wanted to give them “a safe space.”)
Two days later, after a company-wide Zoom presentation of the teams’ demos, the results are announced. Unsurprisingly, the three projects that Compass steered me toward are among the winners. The winners receive a $3,000 prize and the chance to further refine their projects. Compass hopes to begin by incorporating some of this tech into its platform in the coming months.
I catch up with Hulce, who’s tired but happy. His Ignite project garnered some awards, and he’s hopeful that it will eventually be built and deployed. But LoREm Ipsum was the agent’s choice honoree. “I think many of the agents didn’t know about it before and were actually amazed by the product,” he says.
For all of Compass’s AI-related efforts, the company has achieved much greater efficiency gains the old-fashioned way: through layoffs and cost-cutting. Three rounds of layoffs in the past two years as well as moving some labor offshore helped it to trim operating expenses by $128 million in the second quarter of 2023 compared to the prior year, allowing Compass to become cash-flow positive. “All they do is cut their expenses,” says Delprete, the global real estate tech strategist. “They keep firing people until they’re break-even.” Compass has also started charging agents to use certain services, such as DocuSign. And the company’s biggest bet to grow revenue and improve margins—its own mortgage, title, and escrow services, which Reffkin has mentioned on several earnings calls—has little to do thus far with artificial intelligence.
Morant, the California broker, says Compass’s tech stack is useful but notes that it has functioned as a way to convince brokers to join the company rather than change how they work once they’re there. Reffkin himself has highlighted Compass’s platform as a recruiting and retention tool. “Compass AI, for now, is largely about making agents happy,” Hulce says. “If you’re the best place to be for an agent, more agents will come to Compass, more agents will stay at Compass, more transactions will happen at Compass. But that’s not quite as quantifiable when you look at the bottom line.”
The backlash to Compass’s 2018 licensing deal made it clear who is really in charge at the company: the agents. No amount of technology can replace them. Compass has a lot of valuable data, and AI can help brokers sift through it. But sales—especially on the high end of the residential real estate market, where Compass has made its name—come down to the personal relationships that agents cultivate. An automated email sent out based on Likely to Sell data may sometimes work, but maintaining relationships with clients past and present and getting word-of-mouth referrals is often what seals the deal.
Because Hulce was one of the hackathon’s organizers, he won’t receive a share of the $3,000 cash prize awarded to his team. That’s fine with him: He cares more about the bragging rights. When it comes to AI and real estate, maybe Reffkin feels the same way too.
login
who are we?
contact us
qosheapp
Toi Staff
Gideon Levy
Belen Fernandez
Andrew Mitrovica
Mort Laitner
Rami G Khouri
Ali Fathollah-Nejad
Nikkei Editorial
visit website
