Despite countless creative housing developments and advancements in policy to address the homelessness crisis in Los Angeles, California, at least 75,000 people are unhoused in L.A. County.

And for every available place of permanent supportive housing in the county, about four more are needed to meet the demand of rising homelessness.

That’s why the state is turning to predictive artificial intelligence.

Researchers from the California Policy Lab at UCLA have helped lead a major project to figure out how to best allocate housing to people most in need in L.A.

Old triage tools that the city long relied on required self-reported information from people to understand their level of vulnerability — or need — and dispatch case workers to their aid.

The team at UCLA has built a more complex predictive risk model, evaluating old data to identify which questions are most helpful in narrowing down the scope of who to help first — and working to eliminate racial biases from previous methods.

The AI model also uses criminal, hospital, and death records, alongside data from local housing authorities, to gather a more complete picture of the matter at hand.

Their work culminated in a report to help the county’s Homeless Services Authority and Homelessness Prevention Unit better support people facing homelessness. 

“This report offers a clear view of the model’s effectiveness and shows that, when used carefully, predictive analytics can be instrumental in preventing individuals and families from experiencing homelessness,” said the report’s co-author Janey Rountree, the executive director of the California Policy Lab at UCLA. 

“We hope this report will help other state and local governments learn from our efforts and consider similar strategies in their programs to combat homelessness.”

The report breaks down how the predictive model identifies individuals at risk of homelessness or adverse experiences of homelessness, as well as the steps case workers should take to contact these individuals. 

It also includes insights on equity, sharing that the model performs consistently across race, ethnicity, and gender.

“All predictive models make errors — that’s inevitable — but what you want to make sure is that those errors don’t systematically discriminate against certain groups,” co-author Brian Blackwell told Vox. 

To do this, the research team used a simple AI model called “ordinary least squares linear regression,” which is similar to the way a GPS navigates through data to find a best route to a destination: estimating the relationship between variables to predict the most accurate outcome.

In this way, the model is able to identify individuals who are otherwise not accessing other homelessness prevention services and allows case workers to proactively reach out to them.

According to a press release from the policy lab, individuals most vulnerable to homelessness do not always self-identify, may not be aware of programs that can help them, or how to connect to those programs.

“By harnessing predictive models, we’re able to identify those at high risk of homelessness who might otherwise go unseen,” Max Stevens, chief analytics officer at L.A. County’s chief information office, said.

“This approach allows us to be proactive in extending support and makes our homelessness prevention efforts more precise, responsive, and impactful.”

The new model assigns different point values to different questions in surveys among unhoused community members, with more weight given to questions that are most closely associated with negative outcomes. 

Then, it provides a single number summarizing a person’s vulnerability, now with an adjusted scoring system to correct for previous racial biases.

Case workers from the Homelessness Prevention Unit then support clients with financial assistance or a myriad of other social support services.

The preliminary impact of the model showed that out of the 456 that were identified and supported by the Homelessness Prevention Unit, 86% reported living in permanent housing situations upon discharge.

In other words: The model was a success.

“If we can identify people at the highest risk of homelessness and intervene early, we have a real opportunity to prevent them from experiencing the trauma of losing their home,” said Dana Vanderford, associate director of homelessness prevention for the county’s Department of Health Services. 

“By investing in the right proactive interventions, we’re not only preventing homelessness but also addressing challenges like mental health and economic stability — all of which ultimately save costs and lives.”

Although AI remains a hot topic and some housing advocates disagree with a numerical ranking system to prioritize who gets help first, those on the front lines see its potential.

“If it rolls out in the way that we hope, then it will be a huge leap in the right direction,” Debra Jackson, a housing matcher for homeless service nonprofit St. Joseph Center, told Vox.

“The goal is to move someone from unhoused to housed with the least amount of trauma.”

‍

Header image by Russ Allison Loar (CC BY-SA 4.0)