Prevention Targeting Works. Most CoCs Aren't Built to Do It.

By Common Ladder · May 29, 2026 · 8 min read

TL;DR

• Targeted homelessness prevention — directing time-limited assistance to households statistically most likely to enter shelter within 90 days — reduces shelter entry by 76-81% compared to first-come-first-served or intake-first approaches.^1, 2
• The Marginal Value of Public Funds for well-targeted prevention is approximately 2.47, meaning each dollar produces more than two dollars in measurable social value — among the strongest cost-effectiveness profiles in the homelessness response field.³
• The reason most CoCs aren't producing those results is not motivation or scarcity of evidence. It is the absence of three pieces of operational plumbing: a defined eligibility pool, a validated risk score, and an intake flow that can act on the score in days rather than weeks.
• Universal prevention — assistance offered to anyone who calls and qualifies on income — burns roughly four out of every five dollars on households that would not have entered shelter anyway. The math is brutal and well documented.
• Building prevention targeting takes 18-24 months of sustained investment in data infrastructure before service delivery improves. Foundations that fund only the assistance dollars and not the build are subsidizing the inefficiency they say they want to fix.

A case manager in a mid-sized CoC took a call last fall from a woman three weeks behind on rent, with an eviction filing scheduled for the following Tuesday. The CoC had prevention dollars. The woman qualified on income. The case manager wrote up the application and put it in the queue. Six weeks later, when the funding came through, the woman was already in a shelter on the far side of town. She had been one of 340 applicants that quarter. The CoC had funded 92 of them. Of the 92, no one knew how many had been on a path to shelter and how many had not.

This is the operational reality of prevention in most American CoCs. There is money. There is need. There is a queue. What there is not — and this is the thing the field has been quietly avoiding for a decade — is a system that can tell the difference between a household three weeks from shelter and a household three months from a hard month that would have resolved itself.

The argument

What the evidence on prevention targeting actually says is this: when a CoC can identify the households most likely to enter shelter within 90 days and direct time-limited prevention assistance to them specifically, shelter entry rates for the targeted population drop by 76-81% compared to comparison groups.^1, 2 The Marginal Value of Public Funds for that kind of targeting comes in around 2.47 — among the strongest cost-effectiveness numbers in the entire field.³ The work of Beth Shinn and Andrew Greer (2013), William Evans and colleagues (2016), and more recent replication work by Phillips and Sullivan (2025, still PROVISIONAL) all converge on the same direction.

What the evidence does not say is that prevention dollars produce those results in any configuration. Universal access prevention — assistance offered to anyone who calls, qualifies on income, and is willing to fill out the paperwork — produces a much smaller effect, because most of the households served would not have entered shelter regardless. This is not a moral failing of universal access. It is the predictable result of allocating scarce resources without a triage tool. The strongest version of the targeted-prevention research is roughly four times as cost-effective per shelter entry averted as the universal version.

The takeaway is uncomfortable but precise: prevention works, and most prevention spending isn't getting the prevention effect, and the reason isn't lack of evidence. It is lack of infrastructure.

What targeting actually requires

A CoC running real prevention targeting has three things built that most CoCs do not.

A defined eligibility pool larger than the funded slots. Targeting is impossible if every household that walks in gets served. The pool has to be large enough — typically three to five times the available slots — that prioritization is meaningful. In practice this means the CoC has built relationships with referral sources upstream of homelessness: courts handling eviction filings, utility companies notifying of shutoffs, school districts flagging address instability, emergency departments and behavioral health crisis services. The infrastructure is the relationships, not just the data.

A validated risk score, not a vibes score. The score should be built from outcomes data — which households in your historical HMIS, given which observable characteristics at intake, ended up in shelter within 90 days. The variables that matter consistently across the literature include recent eviction filing, prior shelter contact, doubled-up status with a host who has signaled imminent end, household composition (single adults with a recent shelter exit are higher-risk; families presenting for the first time without a prior CoC touch are often lower-risk than intuition suggests), and behavioral health crisis contact in the prior 60 days. The model does not need to be sophisticated. It needs to be empirical, locally calibrated, and re-validated annually.

An intake flow that can act in days, not weeks. A targeting model is useless if the average time from application to assistance is longer than the average time from application to shelter entry. The build has to push intake-to-decision into the five-to-seven-day range for the targeted high-risk tier, with funds disbursing in the same window. This usually requires pre-authorized vendor relationships with landlords, utility companies, and storage facilities; standing approvals from the funder; and an intake team trained to triage upward rather than queue downward. It is operationally demanding. It is also what makes the rest of the apparatus matter.

Why this isn't built in most places

The diagnosis is rarely a leadership failure. It is a funding and design problem that has been hiding in plain sight.

Most homelessness prevention dollars arrive in CoCs from federal or state pass-throughs that fund assistance, not the data plumbing required to target it well. ESG Prevention funds the rent check; they do not fund the eviction-court data integration that would tell you whom the rent check should go to. CoCs that have built targeting did so by braiding philanthropic capacity-building dollars, internal operating reserves, or one-time pandemic-era flexibility with the federal assistance dollars. Foundations that fund only direct services and not the build are, in effect, ratifying the inefficiency they then critique in their next strategy memo.

There is also a cultural barrier worth naming. Prioritizing a household that is not yet homeless over a household that visibly is feels wrong on the first pass. Intake staff trained in trauma-informed practice often resist explicit triage that downgrades anyone. The honest response is that prevention targeting and shelter response are different operations with different goals, and conflating them produces worse outcomes for both. A CoC that uses prevention dollars to respond to households already in shelter is not doing prevention — it is doing rapid rehousing with a different funding label, and the math comes out worse on both ends.

What this means for the next budget cycle

For CoC leadership, the question to put on the table is plain: do we have the three pieces — defined pool, validated score, days-not-weeks intake — and if not, what is the 18-month build plan to get them, and what does it cost? Treat the answer as a capital project, not an operational tweak.

For foundation program officers, the corresponding question is whether your prevention RFP rewards CoCs that have built targeting infrastructure or only those that can spend assistance dollars quickly. If your reporting template asks how many households were served but not how many of those households were in the predicted high-risk tier, you are funding throughput, not prevention. Adjust the reporting template and the kinds of grantees that win prevention dollars will shift within two cycles.

For state housing agencies, the leverage point is technical assistance. A statewide investment in a shared risk-modeling toolkit, an HMIS-to-court-records data integration template, and a pre-authorized vendor framework would do more to accelerate prevention performance across CoCs than any equivalent increase in per-capita assistance funding. The evidence has been clear for more than a decade. The buildable answer has been clear for at least five years. What is missing is a decision, at each layer of the funding stack, to fund the build that makes the evidence usable.

The work is the work. The evidence is on the shelf. The plumbing is what has to get done next.

Have a correction or an update? Contact us — we keep this information current.