Can AI help government prevent homelessness?
A person stands next to an encampment in Skid Row in downtown Los Angeles on July 26, 2024. APU GOMES/AFP via Getty Images
A predictive model is helping Los Angeles authorities link people at risk of homelessness to assistive services to keep them housed.
In Los Angeles County, more than 75,000 people were unhoused on any given night this year. Statewide, homelessness rates in California show little signs of slowing down, but preliminary results from a predictive model suggest that artificial intelligence could play an imperative role in mitigating the situation.
“A critical thing to understand is that most people are not becoming homeless immediately from a leaseholder situation or other situation where they could have been evicted,” said Janey Rountree, executive director of the research institute California Policy Lab at UCLA.
Many people become unhoused because they are leaving a “couch surfing” situation, a drug treatment facility or carceral setting, she explained during a webinar Monday hosted by the Joint Center for Housing Studies at Harvard University. Because of that, she explained, “a homelessness prevention program that’s going to reach people at highest risk of homelessness can’t be limited only to eviction prevention.”
That’s where a predictive model developed by the California Policy Lab comes in. The model uses administrative data from health, mental health, social service benefits, arrests, probation and homelessness service records to identify people in Los Angeles County who are at high-risk of losing their housing.
The model’s results help the county’s homeless prevention unit conduct proactive outreach for people who might not leverage prevention services otherwise, said Dana Vanderford, associate director of homelessness prevention at the LA County Department of Health Services. The unit serves between 400 to 600 people annually, offering participants cash assistance, case management, legal services and job-training services, among others.
The unit reaches clients by cold-calling them — which is why the predictive model uses administrative data that already includes people’s contact information — or by sending them letters, she said.
The California Policy Lab started building the predictive model in 2021, using data from the Department of Human Services and the Department of Mental Health to identify people who engaged with those agencies between 2018 and 2019 and experienced homelessness within the next 18 months. Researchers were able to verify the model’s predictions by comparing its results against people’s housing outcomes that were already known.
On a test sample of more than 47,000 people and based on 2019 data, for instance, the model flagged that 10,714 of them were at a high risk of homelessness, according to a CPL report released yesterday.
“The people on the high-risk list have much higher physical and mental health needs and are very vulnerable, and so intervention [services] needed to meet these clients where they are,” Rountree said during the webinar.
Between May 2022 and October 2023, the homeless prevention unit then contacted more than 2,000 high-risk individuals to offer homeless prevention services, with 472 eventually enrolling in the program. A little more than half of enrollees were single adults and 48% of enrollees were families. More than 60% were women and 43% of participants were Black. The large majority of enrollees were ages 25 to 54.
According to the report, preliminary results show that more than 90% of participants completed the program, and 86% of people reported having a permanent housing option — such as renting, living with family or friends or owning their own home — upon being discharged from the county’s homeless prevention program.
“We think it’s remarkable that such a large proportion of our clients stick with the program and receive the full menu of services and support available to them,” Vanderford said.
The model and the homeless prevention program are still undergoing evaluation by the California Policy Lab, which Rountree says is slated to conclude in 2027 once more enrollees have participated in the program.
NEXT STORY: Local fire department testing new wildfire risk reduction app