California city turns to AI to meet housing goals
kokouu via Getty Images
By automating parts of the housing permitting process, officials in the city of Corona hope to reduce administrative burdens for staff and encourage development among consumers.
In California, local governments have been tasked with creating 2.5 million new housing units by 2030, one million of which are supposed to be affordable. For many jurisdictions, however, it’s a steep hill to climb, as development projects continue to be bogged down by complicated, time-consuming permitting processes.
The city of Corona is no exception to permitting delays that often drive up building costs and further fuel housing shortages. Residents and developers are constantly submitting building plans that have to undergo evaluation by reviewers who have to check those documents against hundreds of policies and requirements, according to Chris McMasters, chief information officer for Corona, California.
That means obtaining a permit in Corona could take several weeks to several months, sometimes even stretching out to more than a year.
“We see that as a constant pain point,” McMasters said, and with staff and budget constraints, current resources to sift through the incoming and backlogged permit processes are “never enough, and it’s never fast enough.”
Enter artificial intelligence. By automating parts of the permit application and compliance process, “we want to reduce the menial tasks that government does, and make [the process] more accurate and make it faster to help our employees,” he said.
Leveraging AI for permitting was an ideal use case for introducing the technology into city operations because “you’re not telling it to create art, you’re just telling it to compare this against that,” McMasters said.
In partnership with government technology provider Blitz Permits, Corona officials have developed an internal large language model that leverages visual AI to check electronic permitting documents for errors or missing data, comparing them to the city’s housing and development policies that were used to build and train the model, McMasters explained.
The tool can, for instance, distinguish if information on an application does not match city requirements, like the number of parking spots.
The first few iterations of the model were “not great,” McMasters said, though still significantly better than human-only reviews. For instance, early versions of the AI tool could take hours to generate an output with about 60% accuracy.
But now, the tech can output results within five minutes with an accuracy rate of 90%.
“It continues to learn against your documents,” McMasters said, adding that the continual growth helps build the AI’s accuracy and reliability for future uses with “the more data we feed it.”
The city is also developing a customer-facing component of the tool, which he said aims to streamline the application process for residents and developers. Sometimes, for instance, people submit incomplete permit applications because they want their plans to get in the queue sooner.
They may think that speeds up the approval process, but “for cities, it’s a headache,” McMasters said. Plan reviewers then have to flag any errors or empty fields and return it to the applicant before all parties can move forward with the process, further delaying progress.
Making the permitting and development process easier for applicants will, officials hope, encourage more people to launch housing projects to help address citywide shortages.
While AI can be a game changer for improving government operations and service delivery, humans are still necessary to check over the AI’s results, he said, “I look at AI not as a replacer … [but] more of a companion to what you do.”
The tech is “like your best friend that has a photographic memory,” McMasters said. “It’s to help you maintain accuracy, maintain focus and maintain efficiency throughout.”