40 Years Later, AI Finally Finds Housing

Such are the vagaries of AI. It’s analogous to the old computer science whipsaw, “garbage in, garbage out,” or nonsense input produces nonsense output. Now some folks think what I write is nonsense, but beyond that, you have to be specific; the more specific, the better. Otherwise, you might unintentionally murder someone, me being a case in point.

Anyway, AI is all the rage right now. And it’s moving rapidly into the housing sector. But it is certainly not new. Forty-odd years ago, I ran into an Ohio newspaper publisher who had created a program that would write ads for real estate agents, the premise being that most agents were lousy copywriters. An agent would enter the pertinent details and out popped ad copy.

Here’s an example: “REGAL HOME: VIP residence with parkside charm. Remodeled, stone-pillared Colonial on 18 acres, first-owner pride, 18 fireplaces, master-suite, modern kitchen, 12 BR/12 baths, pantry, finished basement, swimming pool. Near bus. Fencing, formal dining rooms, garden, wood paneling, balcony, servants quarters. FHA-VA $30 million.”

‘More Panache’

In case you haven’t guessed, the description is for the White House. It’s not very scintillating, but it did the trick, relieving the fictional listing agent from an arduous chore. Even though newspaper advertising isn’t what it used to be, AdWriter is still hanging on, writing copy and such for about a dozen or so companies. “It was built for print,” says Ken Douthit, the son of AdWriter’s founder, Hal Douthit.

But other outfits have come to the fore recently that offer a little more panache than the Ohio company’s original work. After a paragraph identifying the property as the White House, here’s the beginning of Restb.ai’s 335-word listing description for the presidential mansion:

“This stunning Neoclassical-style building features a striking white exterior, adorned with grand columns and intricate detailing. The White House comprises 132 rooms, including 35 bathrooms, spread across six levels, providing a mix of formal and informal spaces to cater to the diverse functions of the presidential office.”

Restb isn’t new, either. It’s been around for eight years or so. But it is just now publicizing its technology. At the same time, a string of artificial intelligence-powered technologies have moved into the housing space recently, all aimed at making it easier for agents to automate the time-consuming task of writing property listing narratives, among other things.

There are chatbots such as ChatGPT from OpenAI, Google’s Bard, and Microsoft’s Bing, among others. All work a little differently, and all leave something to be desired. At best, they may require the agent to tweak the results. But at worst, what they spit out is not worth the few seconds it takes to view the results. It’s called a “hallucination” in technology lingo.

Beware The Risks

AI is a “very powerful tool as far as being creative,” Tony McGrory of the BTG Group, which is building tiny homes in New Mexico, told me. “But there are a number of downside risks.”

One, obviously, is accuracy. While some descriptions “can be plausible,” McGrory told me, “they can have only an element of truth. If the information isn’t verified, you can very quickly transmit misinformation.” The product “has to be absolutely verified,” agrees Long & Foster agent Hill Slowinski in Maryland, who uses ChatGPT. “I had to put a lot into (the narratives) to customize them,” he told me. “But it gives me a logical outline and pointed out some things I hadn’t thought about.”

ChatGPT responds to questions, prompts, and a list of agent-provided features. Within seconds, it culls vast amounts of data from the Internet, books, articles, and websites, and produces listing descriptions. Restbi, on the other hand, is not a bot, at least not in the true sense of the word, Chief Product Officer Nathan Brannen explained to me. It doesn’t chat, either.

Rather, it uses computer vision to scan photos submitted by the listing agent. It extracts details from the images and information from data providers like CoreLogic and Black Knight plus tax records and school districts from public sources. The technology then writes a complete description that automatically populates the agent’s particular multiple listing service platform.

The program can identify more than 300 features, taking the story far beyond the details noted in a typical listing. And it can spin out write-ups in more than 50 languages. Agents can even select different styles or tones – whimsical, for example, simple or straightforward – and, like AdWriter, can write to any word count an agent desires.

Of course, AI has found its way into the mortgage space, too. But even though it has only recently found its way into the public consciousness, it’s been around for some time. For example, Sun West Mortgage’s AI system MORGAN was developed “over decades,” says CEO Pavan Agarwal.

MORGAN is a 24/7 personal assistant accessible via chat that delivers guaranteed responses to borrowers within minutes. Better yet, it can handle even the most complex loan scenarios, eliminating the paperwork and associated time and costs many lenders dread. But at its core, it “delivers a very accurate and complete (underwriting) analysis” in minutes rather than weeks, thereby offering every borrower an equal shot at financing.

Then there’s Ocrolus, which uses AI-driven document automation for faster and more accurate lending decisions. The Manhattan-based company combines state-of-the-art AI from across the sector to process bank statements, pay stubs and the like without human involvement and produce a single solution so lenders can act with confidence. The system is enabled by its massive document data set built, the company says, from processing hundreds of millions of document pages.

Ocrolus most recently integrated OpenAI’s GPT “embeddings” into its set of technologies, including subsystems from Amazon and Google and its own proprietary deep learning architectures. “Most documents we receive are processed with no human interaction,” says Zach Haehn, senior vice president of engineering. “By expanding our technology platform, we’re creating more efficient ways to deliver the accurate data and analysis lenders require to make more informed decisions in the fastest way possible.”

And let’s not forget Radian and the mortgage insurance company’s AI-powered “Homegenius” search portal. Using image recognition technology adapted from what is being developed for driverless cars, the system analyzes listing photos to assess room conditions and a property’s overall value.

Meanwhile, back at the ranch, Zillow and Redfin have recently incorporated ChatGPT into real estate listing sites. The plugins allow would-be buyers to describe the homes they are looking for in a conversational way. Then, in a blink of an eye — or a flash in the pan? — the chatbots will pull up relevant listings.

Just as the bots save agents time in writing listing descriptions and ad copy, the Zillow and Redfin bots are intended to save buyers hours researching and scrolling and clicking on listings, most of which aren’t what they are looking for, anyway. Says Redfin’s Vice President of Product Ariel Dos Santos, the plugins can suggest houses and neighborhoods that may not have been found with a conventional map-based search.

Hype? Perhaps. That’s why real estate strategist Mike DelPrete is taking a cynical stance, at least for now. “It’s so difficult to separate the signal from the noise,” he advised agents on a recent podcast, “that all of it should be treated with a bit of skepticism ... It needs to be ‘eyes wide open’.”