“Back in 2018, I … thought how strange it was that when you bought a used car you got a Carfax report of its history, but when you spent a million dollars on a condo, you got hundreds of pages of noise,” says Thomas Beattie, CEO of 91原创-based OctoAI Technologies Corp.
Two years later, his company launched Eli Report, which has since produced over 25,000 strata document reports for about 7,000 users across North America. Some of this business is generated through partners such as eStrataHub and Strata Press, two centralized repositories of condo documents.
OctoAI is among a growing number of B.C. companies offering AI tools to streamline the purchasing process for real estate agents and buyers.
They all use some form of AI to produce summaries of strata document bundles in minutes, making them searchable and flagging important clauses for prospective buyers. In some cases, chatbots can answer pointed questions, and special assessments can be forecasted.
“One of the big challenges … with artificial intelligence is that when it doesn’t know [something], it’s very happy to make something up. We designed our systems specifically so that wasn’t a possibility,” Beattie said about Eli Report.
His tool has a proprietary system that uses extractive intelligence, sorts and categorizes issues and displays a side-by-side viewer, allowing easy reference to source material. This way, users can double-check the AI summaries, which are produced in under 10 minutes.
Meanwhile, traditional manual reviews can be time-consuming, expensive and prone to human error, according to Ashkan Tavassoli.
The CEO and co-founder of 91原创-based BLINQ Innovations Inc. sought out to automate these reviews, developing an AI-driven tool he said aims to see “the bigger picture” by moving beyond keyword searches.
“What we do is called semantical search, which is very similar to how manual reviews are done,” he said about his company’s StrataReports tool. “You basically read all the documents, understand the building as a whole, and then we write our report.”
StrataReports has a chatbot trained on a building’s data to help prospective buyers with any questions they may have, saving time that might otherwise be spent asking questions to Realtors.
“That’s kind of our secret sauce,” said Tavassoli.
The company has also partnered with Offerland, which provides a service estimating how much property investors can rent out their units for using live, comparable market data.
Another local company has recognized the lack of a standard format for strata document packages can be a challenge for real estate agents and buyers.
“The human error aspect is that most people are looking in specific places for information,” said Andrew Armstrong, principal and co-founder of 91原创-based The Real View (formally 1390729 B.C. Ltd.).
“If they don’t find it there, they assume it doesn’t exist.”
His company uses a technique known as retrieval-augmented generation to analyze strata document packages to produce a comprehensive report on key aspects of a building, ranging from its financial health to its pet policies.
Through optical AI, the company can analyze these documents, even if the PDF quality is poor.
If a real estate agent or buyer doesn’t find information where they expect to find it in the strata documents, The Real View’s tool examines every page of every document for every answer.
Manual reviews still advised
Although they are gaining in popularity, some services may not be fully prepared for the possibility of erroneous or incomplete AI summaries, which can create legal risk.
It’s why Eli Report, for example, carries errors-and-omissions (E&O) insurance, and has terms and conditions indemnifying the company against liability. The company regularly improves its AI training models to ensure nothing is missed, erring on the side of giving users more information rather than less.
Meanwhile, StrataReports co-founder Tavassoli said his company’s services are provided on an “as is” basis. He said his technology’s systematic approach makes it more accurate than manual reviews of strata document packages.
Even after using an AI tool, manual reviews may still be worthwhile for an additional layer of verification. But manual reviews could prove difficult for especially voluminous or disjointed document packages.
Tavassoli once saw a package consisting of 2,400 pages dating back to 2010.
Ultimately, reviews by humans and technology-led ones may both be vulnerable to some imperfection.
“You cannot guarantee 100 per cent on anything,” Tavassoli said.
