LA County Homelessness Prevention Unit

LA County's Homelessness Prevention Unit uses a risk model built by the California Policy Lab at UCLA. It is reported to score residents on about 580 factors from county records with names removed, estimating who may become homeless within about twelve months. It ranks about 90,000 people for outreach and misses a majority of those who later become homeless.

How it is used

Case managers in the unit work from a list of roughly the top 10,000 names. They make three to four contact attempts by phone, letter, and email. Taking part is entirely voluntary, and about nine in ten of the people they reach accept help. The model never denies anyone help or cuts off a benefit.

Enrollees received an average of $6,469 in flexible financial assistance, plus about six months of case management with small caseloads. Enrollment, not scoring, is the documented hard step. The enrollment rate rose from 21 percent to 35 percent after the unit added a dedicated outreach team and standardized discharge. The sources do not describe that discharge step, a set way of ending a person's time in the program. They do not say what the enrollment rate is a share of, so it cannot be set against the nine in ten who accept once reached.

Where it came from

The approach began as a 2019 study by the California Policy Lab and the University of Chicago Poverty Lab. That study predicted first-time homelessness among single adults, using linked records from seven county agencies. The county Board of Supervisors approved and funded the program in 2020, with roughly $26 million in federal COVID-relief money. The unit launched in 2021.

The unit is part of the county Department of Health Services' Housing for Health division. The county Chief Information Office links the de-identified data. Reporting on the deployed model names three departments whose records it uses: public social services, mental health, and health services. Its inputs include emergency-room visits, behavioral-health care, arrests and probation, benefit enrollment and applications, and prior homeless-services contact.

Whom the model misses

A false negative is a person the model ranks low who later becomes homeless. The California Policy Lab's November 2024 equity audit tested the model on 47,582 people eligible for a score. Of those who later became homeless, it missed about 56 percent of Black individuals and roughly 63 to 65 percent of other groups.

So the model misses a majority of the people who later become homeless. It does so roughly evenly across race, ethnicity, and gender, and identifies Black individuals slightly better. The audit found no evidence of systematic exclusion. The 2019 study noted risk factors its county data did not include, such as staying doubled up with others, financial shocks, and domestic violence.

The model still points outreach toward higher risk. CalMatters reported that about 7 percent of all roughly 90,000 eligible people became homeless within 18 months, against about 24 percent of the top 10,000.

What the early results show

In a pilot from May 2022 to February 2023, 335 people enrolled. The California Policy Lab reported they were 71 percent less likely to enter a homeless shelter or have street-outreach contact within 18 months. The comparison group was 1,285 eligible people who did not enroll, adjusted statistically for differences between the groups.

The California Policy Lab states this is an association, not yet proof that the program caused it. A randomized controlled trial is meant to settle that. It decides by chance which eligible people are referred to the unit, then compares the two groups. Its results are expected in 2027.

As of July 2025 the program had served 1,498 people. In 2024 alone it served 712 households. About 86 percent of those who completed it kept or entered permanent housing.

Open questions

The program's federal COVID-relief funding was scheduled to end in 2026. Observers questioned the value of identifying more people in need without more housing to match.

Who checks the model

The California Policy Lab built the model and is also the program's designated independent evaluator. It runs both the equity audit and the trial. Both look back over months, not at each case as it happens. Besides them, the sources describe no check of whom the list leaves off, and they name no outside regulator.

What this network is drawn from

This network follows the pattern the case file describes. It does not reconstruct the actual model or program. It shows the model, the ranked list, the case managers, the linked county records, and the California Policy Lab's evaluation.

Several of its links are ones the sources do not describe. They include case managers comparing notes, a separate review of people left off the list, and enrollment records as data for later models. The people the model ranks are outside the network.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. A tool closes a pathway when mistakes stop passing along it. The work along it may go on. On this network a mistake is mostly a missed person: someone at high risk whom the model ranks low or outreach never reaches.

This case has a budget of 8 units and starts with no pressure applied. It opens at tipping point, where mistakes are on the edge of building on one another. Explore (No Targets) sets no targets. The other three levels all ask for the network's mistakes to be contained, meaning corrected rather than building on each other.

Under Service Targets Only, the targets can be met in hundreds of ways. The cheapest cost 2 units: any one of Escalate checks, Keep skills sharp, Mark AI-written records, Review on schedule, or Peer sharing rules. Lingering effects are effects that stay after their cause is gone, and the Lab opens with them on. With them turned off in the Dynamics menu, Vet connections also meets the targets alone, for 3 units.

Service and Safety Targets also asks you to close every failure pathway and keep up with the work, among other targets. Three pathways are open before any tool is used: Ranked list given to case managers, County records used as model inputs, and Outreach results recorded. Escalate checks closes the first. Mark AI-written records closes the second. Store less data closes the third.

Those three tools together cost 7 units and meet the targets. Two other ways fit the budget, each for 8 units. One uses Mark AI-written records at its stronger setting. The other uses Vet connections in place of Mark AI-written records, which works here because lingering effects are off at this level. That makes two tool sets, or three ways once stronger settings are counted.

All Governance Targets asks for more again. Only one tool set meets it: Escalate checks, Mark AI-written records, and Store less data. It costs 7 units, or 8 with Mark AI-written records at its stronger setting. Vet connections does not work in its place here. Lingering effects are on at this level, so it works at reduced strength and leaves County records used as model inputs open. Adding Understand the system restores it, but that set costs 11 units, over the budget.

More is not better here. Using every tool at its strongest setting costs 32 units, well over the budget of 8. It contains the mistakes and closes every failure pathway. But the benefit the model brings to the work falls below what the targets ask, so it meets them at none of the three levels.