LA County constantly fights homelessness. How AI helps prevent it.

Millions struggle with housing affordability in Southern California. But Los Angeles County’s experimental tool combines more than 400 factors to identify the residents most at risk of ending up on the streets – and it’s working.

First Look

Last June, Breanna Sanchez was in trouble: she was behind on the rent, and her landlord in Los Angeles was threatening her family with eviction. But then she was thrown a lifeline, helped by an invisible algorithm.

Ms. Sanchez’s precarious situation had been detected by an artificial intelligence tool being used as part of an innovative experiment by Los Angeles County to identify residents at risk of losing their homes within 12 months.

The tool analyzes more than 400 factors – including data from the county jail and hospitals, safety net programs, and homelessness and foster care systems – to create a list of 100,000 people deemed at risk of ending up on the streets.

Then city officials get to work to deliver counsel and even cash, hoping to stop more from ending up as statistics in a growing nationwide homelessness crisis.

Ms. Sanchez got a letter urging her to contact a county department for housing help. Soon she was talking to an outreach worker, getting signed up for health and other programs and receiving financial assistance that helped her pay her rent, fix her car, and more.

“I probably would have been evicted,” Ms. Sanchez told the Thomson Reuters Foundation. “Me and my husband felt stuck and didn’t know how to move forward.”

The Los Angeles pilot began in 2021 and smaller trials are underway elsewhere as authorities grapple with a homelessness crisis that has led some cities, including Los Angeles, to declare states of emergency.

Of the hundreds of people the new L.A. Homeless Prevention Unit has worked with so far, about 87% retained their housing after they left the program, said Dana Vanderford, associate director of homelessness prevention for the county’s health services department.

“We often find them in a moment of crisis – we’ll call, and they’ll say, ‘I don’t know how you found me. I’m losing my housing next week and not sure what to do’,” Ms. Vanderford said.

“Our ability to appear out of nowhere and intervene and resolve a crisis – we’re really proud of that,” she said.

AI is also being used to help predict homelessness risk in Calgary, Canada, and Geoffrey Messier, who is helping with the development of that project, says the L.A. pilot could be a game-changer.

“I hope it is a bit of a watershed moment,” said Mr. Messier, a professor of electrical and software engineering at the University of Calgary. “This is the first time machine-learning was an integral part in identifying people who needed help.”

“Right way to apply technology”

More than 653,000 people experienced homelessness in the United States in 2023, a 12% increase over the previous year, according to a U.S. Housing and Urban Development report.

Janey Rountree, executive director of the California Policy Lab at the University of California in Los Angeles, said recent research showed a one-time cash payment and other services could help stem that tide.

Last year, the University of Notre Dame published a study that found that at-risk people in California’s Santa Clara County were 81% less likely to become homeless within six months of enrollment in a financial assistance program.

“If you get the household at the right moment, give them $2,000 to $6,000 of assistance to help pay debt or rent, you can put them back on a trajectory toward housing stability,” said Rountree, who helped spearhead the research that led to the creation of the L.A. pilot and is involved in its evaluation.

But first local officials have to be able to find those in need, which is where the new AI pilot comes in.

When an at-risk individual is identified, L.A. Homeless Prevention Unit case managers, like Jocelyn Bataz, go to work.

“I call them out of the blue and introduce myself,” she said, explaining that some people are initially skeptical. She reassures them that the call is not a scam and then tries to find out about their housing situation, utility bills, and other stresses.

Within days those eligible can get financial assistance and access to other services to, for example, sort their debts, find childcare or a job, and get insurance. Ms. Bataz said she also provides a lot of emotional support.

“This is the right way to apply technology – to make sure we’re reaching folks that aren’t otherwise connected to homeless systems but are at risk,” said Alex Visotzky, a California-based senior policy fellow with the National Alliance to End Homelessness.

Eye of AI

AI tools can be more effective than manual approaches to find out who is most chronically in need of housing support, said Stephen Goldsmith, urban policy professor at Harvard University.

“Because preventing [homelessness] is complex, AI tools can help cities be more effective by focusing necessary services on specific individuals and their needs,” he said.

AI is also being used in the Canadian city of London, in Ontario, where the homeless population doubled during the COVID-19 pandemic.

Craig Cooper, director of London’s Housing Stability Services, said an AI tool is being used to analyze data from people already in the shelter system to identify who is most at risk and, so far, the model appears highly accurate.

Similar efforts are underway in Calgary to address the fact that someone new to the shelter system typically has to wait a long time to qualify for housing assistance. But AI can crunch data from the first months in the system to assess longer-term risks.

“If so, that increases their priority for support,” said Mr. Messier, from the University of Calgary.

In Los Angeles, Ms. Sanchez is grateful to have her life back on track just a few months after leaving the program. Her health concerns are being addressed, she will finish her high school degree in June, and she is expecting a baby.

“It really helped a lot,” she said of the pilot. “I feel safer here now.”

This story was reported by the Thomson Reuters Foundation.