Meet Your New Co-Worker

Print Friendly

How real estate companies are increasingly using machine learning systems, including AI, to work alongside human employees to achieve things that neither could do on their own.

Across the United States, in leasing offices and executive suites, real estate professionals spend more and more of their time working with robots—excuse me—machine learning systems. It’s easy to be overwhelmed by the grand claims and the flood of jargon used to describe the technology: artificial intelligence (AI), large language models, neural network computing.

But the number of real estate applications that use machine learning is already impossible to ignore. Sometimes it seems as if the leasing chatbots that helped property managers during the pandemic may be about to replace their company CEOs. One real estate company is even using ChatGPT to help write its investor reports for the first quarter of 2023.

But what AI can do is still different than what a human can do—at least for now. Real estate companies are finding opportunities in these differences to have human employees and machine learning applications work together to accomplish things neither could do on their own in the limited time available.

Mastering the Investor Report

Apartment investor DB Capital Management is using ChatGPT, the now-famous AI chatbot developed by technology firm OpenAI, to help write its report to its investors for the second quarter of 2023.

Machine learning can absorb a huge amount of information and find correlations and patterns within that data. The system also can apply those patterns to data it has never seen before in ways that feel new.

Writing large sections of DB Capital’s report may seem like a magic trick, but it’s just the latest surprising feat accomplished by an AI system since several became available for the public to explore.

To help the AI write its quarterly report, DB Capital fed ChatGPT several years of its past reports. ChatGPT also received the latest facts and figures on the more than 20 apartment properties in the company’s portfolio taken from the Entrata property management software system.

In the past, DB’s regional vice presidents and associates would take two weeks at the end of April to craft a first draft. Considering the salaries of these executives and their staff, it cost the firm tens of thousands of dollars every quarter.

Now this drudgery will be handled by AI, with DB Capital spending $50,000 to $100,000 to create its system. “It’s going to pay for itself pretty quickly,” says Brennen Degner, CEO of DB Capital. “There are sections of the report that it does very well. The goal is for ChatGPT to get us 75% to 80% there, then a person can get it to 100%.”

There are still some things that ChatGPT can’t do. Its creations may appear to be new, but they are all based on systems and patterns the system has seen before.

DB Capital’s regional vice presidents will probably still have to write their updates on how value-added renovations are proceeding at individual apartment properties. These projects are unique enough to pose a challenge to the AI. Degner himself as CEO will continue to write an introduction.

The company now may even be able to hire a writer to put the finishing touches on its quarterly reports, once ChatGPT has created a first draft and the executives have added some of their own analysis. According to DB Capital, it has never hired talent like this to help with its quarterly reports.

“At that point you have already saved so much money on your quarterly payroll for your vice presidents, a lot of other things become possible,” says Degner.

The company also could use the system to provide reports to its investors more often. “Why not monthly reports?” he adds.

Humans Still Needed

Companies that use machine learning must keep a close watch to catch inaccuracies as the systems splice together words and numbers. Depending on what material the systems have been learning from, they can appear to make up their own facts.

For example, Picket Homes is creating a new property management software system for companies that manage single-family rental homes. Machine learning allows the system to “write” descriptions of single-family houses being offered to rent.

“We have to tell it to not fabricate or say something that it could not verify,” says Q Shay, co-founder and chief technology officer for Picket Homes, a technology company based in Seattle. “Just including that prompt improved the accuracy. … But it still had stuff that was wrong.”

To catch these problems, “a human being is always in the driver’s seat,” says Shay. “I don’t see any version in the near future where a human being is not hitting the send button.”

A Leasing Assistant

On Sundays, the leasing offices are now closed at stabilized apartment communities operated by Camden Property Trust.

The REIT can keep its leasing offices closed on Sundays and maintain its occupancy goals, thanks to technologies like virtual leasing agents powered by machine learning.

“They can give their team a day off,” says Tyler Christiansen, CEO of Funnel Leasing, the technology company that created Camden’s automated leasing agent. “It’s perceived as a perk—the on-site teams have that extra day to go to church or go to a party or whatever it may be.”

Virtual leasing agents can answer questions asked by prospective renters at an apartment community on its website either via email, text, or chat window. Machine learning allows these systems to recognize the questions asked by potential residents and pair them with the appropriate answers.

“The residents ask kind of the same questions. When is the pool open? When can I go to the clubhouse?” says Shawn Mahoney, senior advisor at RET Ventures, a real estate technology venture capital firm, and the former chief technology officer for GID. “Artificial intelligence can take a load off of that.”

When AI can’t answer a question, the interaction is forwarded to a human leasing agent; however, that happens less and less often.

“We can handle 80% of the queries,” says Funnel’s Christiansen. That’s a big improvement from just a few years ago when Funnel created its first virtual leasing agent. “We could only accurately respond to 30% of the queries.”

These virtual leasing agents easily pass the so-called Turing test, which asks if a machine can engage in a conversation with a human without being detected as a machine.

“Many times we have customers who come in and say, ‘I’ve been communicating with Grace, I would like to meet her,’ even though it says clearly this is a digital leasing agent,” says Jamie Gorski, chief experience officer at GID-Windsor Communities, which uses its own AI leasing agent created by Leasehawk.

Funnel’s Christiansen remembers a time not long ago when human leasing agents were skeptical of these virtual leasing agents. “We don’t get any pushback from them any more saying that they don’t want that,” he says.

These tools became especially important during the pandemic, when many leasing offices shut their doors to slow the spread of the disease.

Today, chatbots and virtual leasing agents remain important as apartment companies struggle to hire.

“Everyone is impacted by the labor shortage, so it’s taking longer to fill open positions,” says Jeremy Brown, vice president of marketing for ZRS Management. “If you can have a bot answer a call when you have someone out on leave or you just haven’t found the right person for that position, it will certainly be helpful in the interim period.”

ZRS counted up the email responses sent by its virtual leasing agent. Assuming that a human would take three to five minutes to respond, the virtual agent has done work that it would take humans 86,000 hours to accomplish.

Thanks to these virtual agents, potential renters are also able to ask questions, receive answers, and even schedule appointments after regular business hours when leasing offices are closed.

“Maybe 2 a.m. is the only time you are able to look for an apartment,” says Jessica Galik, multifamily technology solutions lead for Cushman & Wakefield.

ZRS Management found that 25% of its appointments with potential renters were booked by its digital agent—and of those, nearly half, 43%, were booked after regular business hours.

However, companies are still committed to having human leasing agents work alongside their virtual counterparts. “We are not pulling back from our investment in people,” says Galik. “The chatbot is there to give the leasing agent back some time.”

Some firms may even be able to hire staff who work after regular hours to answer texts and emails that the virtual leasing agents need to hand over. “Across the industry there will be extended hours and also remote positions,” says Gorski. “You can provide more coverage and ensure a seamless handoff.”

Machine Learning Really Listens

Apartment companies can also find meaning in the emails and texts from their residents that pile up and sometimes get ignored.

Travtus, a technology company based in London, has created a machine learning platform called Adam that can absorb a flood of inbound communication—emails, text messages, and phone messages—that washes through leasing offices and community inboxes. The system then finds patterns between them that often turn out to be meaningful. They can even point out problems that property managers can fix to improve their business.

For example, Adam found hundreds of emails from residents asking to use the printer in a property management office. These printouts may contain confidential information and may sit for hours in the printer tray before a resident comes to claim them.

Once property managers understand how much residents like using their printers, they can put a printer in a common area or in some other place where managers aren’t responsible for the documents.

The machine learning system can also identify residents who might not renew their leases. Adam flagged one resident with an expiring lease who asked for a lower rent and also had a history of calling the management office to complain about noise from the upstairs neighbor. Armed with this information, the manager declined the ask to lower the rent but offered the resident a chance to move to a top-floor apartment away from the noise. The resident quickly accepted.

“You don’t need to read people’s minds—you just need to read their emails,” says Tripty Arya, founder and CEO of Travtus. “A human could definitely do the same thing if they really cared to find out enough about it—it just probably would take longer, and the important information could easily get lost in the volume of data.”

Keeping Chatbots in Line

Apartment companies are also careful to restrain virtual leasing agents so they don’t break fair housing laws.

“If you allowed the software to answer whatever way it saw fit, the AI may answer based on the knowledge that it has, completely absent of the concept of fair housing laws,” says Funnel’s Christiansen. “AI tells you how it sees the world—not how it could, should, or would be,” he adds. “If you were to plug GPT technology in leasing software today without guardrails and you were to ask it any sort of sensitive question, it could absolutely answer in a way that is inappropriate or illegal.”

Funnel’s virtual leasing agent is given a large dataset of interactions to help it understand the questions. Then Funnel gives the AI a much shorter list of scripted responses to choose from as it responds. If none of the scripted responses seem to fit, it refers the question to a human leasing agent.

“There is never anything that comes out of our AI that should surprise anyone,” says Christiansen.

Source: Multifamily Executive