English is a tough ask to expect across the board — and a barrier pops up more often than you think in CRE. A 2022 report from the Census Bureau said that the number of people who spoke a language other than America's most common jumped from about one in 10 in 1980 to almost one in five in 2019. At the same time, 78% of the population speaks only English.
That means there’s a significant chance in any business, like those involved in CRE, that two people may not share smooth communication. Non-native English speakers often get uncomfortable and may not want to admit that a conversation goes beyond them, according to Jonathan Kroll, chief product officer of Visitt.
The company makes software for CRE building operations. They just released a fully-native, real-time translation called Live Translate. This tool is designed to work in their property management products. One helps operations teams. The other is an app that helps tenants engage with property management.
“It’s using AI in a property management and work order context to make translations accurate,” Kroll tells GlobeSt.com.
“There are many, many property teams that are bilingual.”
The problem is clear if two people don’t speak each other’s language at all.
“It really slows down the collaboration,” he says. That adds time to the operations team and third parties need to resolve problems. The problem compounds when tenants need to communicate with staff.
“In many cases, tenants feel the lack of communication,” Kroll says. “There is no adoption or training required. People are speaking in their native language.” It reduces friction and responses come faster.
At first, Kroll’s team tried to develop in-house solutions for various aspects of their products. Eventually, they found it better to use commercial software like OpenAI’s ChatGPT and Google’s Gemini.
Making use of the third-party generative AI packages varies. Each tends to be better for some applications rather than others. The company has had to experiment for each use of AI. (ChatGPT was the winner for translation.)
Preparation, which took about a calendar quarter, went beyond choosing the best-performing AI platform for translation. They had to train the system on the company’s specific language use.
“You need to provide the model with the context and instructions that were relevant to our world,” Kroll says. The specific language and other steps they take help suppress the potential for hallucination, in which a large language model makes up inappropriate and incorrect answers.
One other step is to keep trying new versions of the various LLMs available, as they all come out with new ones and Visitt needs to know when to change them or even vendors. Because of the structure, the company can fairly easily swap out LLM back ends.
“Our goal here is to really eliminate the language barrier in every aspect, so when it comes to daily operations, whether preventative maintenance,” he says.
Currently, the translation system only works for written communications, although AI could potentially address phone conversations.
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.