What Are the Risks?

Mobile crisis response has an unusually strong evidence base for a policy this young. It also has real operational limits and political vulnerabilities that a leader needs to understand before launch.

---

The Scaling Gap

The open question isn’t whether these programs work, it’s whether cities can build them large enough to meet actual demand. Denver’s mobile crisis team handles only 44% of eligible calls.¹ Austin needs three times its current staffing to meet demand.² A dispatch data analysis of 15 U.S. police departments, reported in American Police Beat Magazine, found that some cities may see up to 20%³ of police dispatch time spent on behavioral health calls, suggesting “the true unmet need for alternative crisis response is far greater than previously thought.” The question isn’t whether programs work at current scale, it’s whether cities can build them large enough to handle the full volume of need. That’s a workforce and funding challenge, not an effectiveness challenge. But it’s the challenge that will determine whether mobile crisis response transforms public safety systems or remains a valuable but partial solution.

44%

of eligible calls handled by Denver STAR, operating 8 vans citywide

The size differential underscores the gap. Austin’s EMCOT has less than 100 employees, roughly a quarter of whom are embedded in 911. The Austin Police Department has approximately 2,484 personnel, including roughly 1,809 officers. Scaling a program to balance against a department that size takes years and sustained investment, which is a different challenge than whether the model works.

The Workforce Shortage

Professor Amy Watson of the University of Illinois reported that “just about half the population of the U.S. lives in behavioral health workforce shortage areas.”⁴ Cities struggle to find qualified mental health clinicians willing to do field response work, which is harder, less predictable, and more physically demanding than office-based practice. Not all clinicians who excel in clinical settings can handle street response: entering unfamiliar situations, working with people in acute crisis, and making rapid assessments in uncontrolled environments.

We don’t have good data on the workforce pipeline at scale. Cities are building creative pipelines: community college programs, peer specialist certifications, and competitive compensation. But whether those pipelines can produce enough clinicians to staff 24/7 citywide programs in mid-size and large cities is an open question. Harris County discovered this after launching.⁵ Contractor turnover was undermining quality, which drove the county to transition to in-house employees. This transition allowed for better retention, but came with alonger ramp-up and a higher per-employee cost.

Dispatcher Resistance

Dispatchers are the gatekeepers. If they don’t route calls to the crisis team, the program exists on paper but not in practice. NYU researchers studying Denver’s STAR program found two distinct forms of resistance.

If STAR can’t protect themselves, I’m not sending them by themselves. It’s a liability that falls on us as the dispatcher. “Why did you send them there? Now they’ve gotten hurt. Now it’s your fault.”

Denver 911 dispatcher, NYU Policing Project Report⁶

Some dispatchers insist on sending police backup to all mental health calls. Others simply forget, after decades of routing every call to police, the mobile crisis option breaks muscle memory. One dispatcher told researchers it’s “more of a, ‘Oops, I forgot,’ because it’s just newer.”⁶ Programs report the resistance improves over time, but the transition period lasts months or years.

The ARPA Funding Cliff

Many programs launched on American Rescue Plan Act funds, one-time federal pandemic relief dollars that are now expiring. Programs that planned for the ARPA expiration — that used the startup window to build outcomes data, cultivate council support, and layer in Medicaid and state funding — are surviving the transition. Programs that treated ARPA as operational funding without a sustainability plan are now in budget crises.

The most prominent casualty is CAHOOTS itself. In April 2025, the pioneering Eugene, Oregon program that had operated for 36 years, and inspired mobile crisis programs nationwide — shut down entirely in its home city. White Bird Clinic’s Interim Executive Director Amée Markwardt told reporters⁷ that the city’s contract covered only about 40% of CAHOOTS’ operating costs; the rest depended on federal grants that had evaporated under a new administration. The city and clinic severed their contract, approximately 20 workers were laid off, and as of mid-2025, CAHOOTS continued operating only in Springfield on reduced hours. A grassroots “Friends of CAHOOTS” coalition organized to bring it back, and by late 2025 the City of Eugene issued an RFP for a new crisis response program — but the damage was done: the nation’s most recognized mobile crisis program collapsed from inadequate public funding. In Montana, two mobile crisis programs in Great Falls and Billings also recently shut down, leaving just six units statewide. A February 2026 NPR investigation documented that a 2024 survey had found at least 1,800 mobile crisis teams nationwide, but “financial support for them is often inadequate and inconsistent.”⁸

The pattern from cities that lost programs is consistent: budgets built on expiring grants without a transition plan to sustainable public funding proved structurally fragile, regardless of how well the program performed.

What Happens After A Serious Incident

No documented program failures exist in the literature, which itself should be noted. We have not found case studies of programs that launched and failed outright, or that suffered a serious safety incident with lasting consequences. This could mean the model is genuinely robust. It could also mean that failed programs don’t generate press coverage or academic studies. A field this young almost certainly has failures we haven’t documented. The absence of documented failure is encouraging but shouldn’t be treated as proof that failure doesn’t happen.

The safety record across tens of thousands of calls, zero or near-zero serious responder injuries in every documented program⁹, is significant. The political vulnerability is that a single serious incident could shift the public conversation, regardless of the aggregate safety record.

The Evidence Gaps

Longitudinal Data

Longitudinal data is still thin. We know short-term outcomes, calls resolved, officer hours saved, costs reduced, arrests avoided, crises de-escalated. We don’t yet have rigorous longitudinal data on whether people connected to services by mobile crisis teams stay connected over years. Do they have fewer crises a year later? Two years? Five? The early indicators from repeat-call reduction (Guilford County’s 344 to 4, St. Petersburg’s multi-year sustained engagement) are encouraging, but the long-term picture is still developing.

Concentration Of Rigorous Research

The rigorous evidence base is growing but still concentrated. In 2023, the evidence was mostly program-reported metrics. Now we have the Stanford study (Denver)²¹, the NBER study (Durham), the Michigan peer-reviewed study, the BMC multi-study review, and the Portland trust study, independent research with comparison groups and rigorous methodology. But these studies still cover a handful of cities. The pattern is consistent and the direction is clear, but a skeptic can fairly note that the controlled research base, while no longer thin, hasn’t yet reached the scale of evidence behind interventions that have been studied for decades.

A related limitation: the same programs appear repeatedly across different analyses, Durham, Harris County, Denver, St. Petersburg, Austin, Minneapolis, Oklahoma City, Guilford County. That’s partly because these are the programs with the most robust documentation and the most transparent data. It’s also because the field is young enough that only a few dozen programs have operated long enough and at sufficient scale to generate the kind of evidence a rigorous product demands. The breadth of the field, 130+ programs in cities of every size and political context, is real. But the depth of documented evidence is concentrated in roughly a dozen cities. The distinction between “many cities are doing this” and “many cities have published rigorous outcomes data” matters: the first statement is well-supported, the second is aspirational.

Fiscal Evidence

The most rigorous cost-benefit analysis is from one city: Durham. The NBER finding that the program generates $902 in net savings per call¹⁰ is a landmark result, but it’s one study of one program. Whether that ratio holds in cities with different cost structures, different Medicaid populations, or different program designs is an assumption, not a finding. The SAMHSA data on 23% lower per-case costs and 79% hospitalization cost reduction¹¹ is broader but less methodologically rigorous. A skeptical budget director would be right to ask whether the Durham numbers are generalizable, and the honest answer is that we expect so, but can’t prove it yet.

The Medicaid sustainability question is unresolved. The 85% federal match is for three years.¹² After that, states drop to their standard match rate , which is substantial but significantly lower. No state has yet made the full transition from enhanced to standard rate, so we don’t know how program funding holds up on the other side. States that built strong outcomes data during the three-year window are better positioned, but the fiscal stress test hasn’t happened yet.

And the liability offset, cities paying billions in police encounter settlements — is a real fiscal argument, but it’s hard to attach a specific dollar figure to any individual city’s expected savings. The argument is directional and logical, not actuarial. A city attorney can make the case; a budget director may still want harder numbers.

Program Design

We don’t have reliable implementation timelines. How long from council authorization to the first call answered? From pilot to citywide? From launch to 24/7? We have “months to years” in various places but no city-specific timeline data published in a form that would allow a new city build a realistic project plan. No city-specific timeline data has been published in a form that would allow a new jurisdiction to build a realistic project plan.

We don’t have head-to-head comparisons of program design choices. Does housing the program in public health produce better outcomes than housing it in the fire department? Does a civilian-only model outperform the co-responder model in cities that started with the co-responder model and transitioned? The Wayne State study found that mobile-only teams outperformed co-response¹³, but the research base comparing design variations is still thin. Most cities chose their model based on political feasibility and institutional relationships, not on evidence that one design consistently outperforms another.

Geographic Gaps

The 130-program count includes programs of widely varying maturity and scale. A 44-team county program operating 24/7 and a two-person pilot covering one zip code during business hours are both counted. The majority are small, early-stage, and still building capacity.

The documented programs are disproportionately in Western and upper-Midwestern cities. The South is underrepresented in the published data, with the notable exceptions of Harris County, St. Petersburg, Baton Rouge, Durham, and Nashville. A mayor in the Deep South or Appalachia will find fewer direct comparables than a mayor in the Pacific Northwest. This is a gap in the documentation, not necessarily in the programs themselves, but it means some leaders will have to build the case with less local precedent than others.

We now have documented cases of programs that stalled. CAHOOTS’ 2025 collapse in Eugene is the highest-profile example, but the pattern is broader: New York City’s B-HEARD program, which the NYC Comptroller’s 2024 audit found still only covered 40%¹⁴ of the city’s police precincts — halted expansion plans after budget cuts. Portland Street Response saw its hiring frozen by a new city commissioner who campaigned against it. Montana lost two programs entirely. These aren’t hypothetical risks — they’re documented failures of the fiscal and political infrastructure needed to sustain programs even when the model works. The pattern suggests that building a sustainable program is harder than launching a pilot.

The rural frontier remains exactly that. The models are emerging, the results are early, and the operational challenges (distance, workforce, call volume) are real. Rural jurisdictions that have launched programs describe innovating rather than replicating what worked in Denver or Durham.

Public Opinion After Failure

We don’t have polling from jurisdictions where a mobile crisis program had a serious negative incident. Would support hold after a responder is injured? After someone routed to a crisis team commits a violent act? The safety record makes these scenarios rare, but the question is untested.

We don’t have strong trend data. Is support growing, stable, or plateauing? The Harris County exposure effect suggests familiarity increases support. But we don’t have multi-year tracking polls showing whether national support has moved since 2020-2021, when these programs entered the mainstream political conversation in the wake of the George Floyd protests. The political conversation has shifted from “defund the police” to “right response, right call” — and the polling we have is from the current frame, not the earlier one.

We don’t have polling specifically testing the strongest opposition messages. We know support holds after voters hear arguments against mobile crisis teams — but we don’t know which opposition messages perform best, or whether there’s a messaging attack that could meaningfully erode support. Any political consultant would want to see an opposition message test before advising a candidate to make this a centerpiece. What we have instead is a policy that polls at 80% and a forced-choice margin of 24 points — numbers strong enough that most consultants would call this safe terrain.

The Political Criticism Landscape

Political criticism comes from multiple directions — and from named voices. Tim Davis, president of the Sacramento Police Officers Association, argued that¹⁵ “our 911 dispatchers do an amazing job and are the perfect people to handle those in crisis” and that “it is imperative that 911 remain under the direction of the police department.” Tom Saggau, a spokesman for police unions in Los Angeles and San Francisco, criticized¹⁶ restructuring proposals as outgrowths of the “defund the police” movement. In Portland, then-City Commissioner Rene Gonzalez ran on a law-and-order platform in 2022¹⁷, defeated the council member who created Portland Street Response, froze hiring for the program in 2023, and described care-centered responses to homelessness as “enabling.”

From the research community, Georgetown Law’s Professor Christy Lopez¹⁸, who directs the Innovative Policing Program, has offered a measured but important caution: police crisis teams “have proved useful and important, but only to a point.” She notes that even well-designed crisis programs are “no replacement for an adequate mental health care system in a community.” The Congressional Research Service’s 2023 review found¹⁹ that “it remains less clear whether these changes translate into actual improved outcomes for people with mental health needs, such as fewer arrests and reduced use of force against them.”

Progressive critics argue from the opposite direction — that these programs don’t reduce police budgets and collaborate too closely with law enforcement. In New York City, the Correct Crisis Intervention Today coalition argued²⁰ that the city’s B-HEARD program was too limited from the start, noting that the city “proposed adding only five mobile crisis teams” — far too few for a population of nine million.

The cross-cutting criticism is itself informative: a program attacked simultaneously for being anti-police and too close to police occupies a different political space than one attacked from a single direction. The bipartisan support data — 80% national, 72% Republican, 89% Democrat — explains why targeted political campaigns against mobile crisis programs have not materialized in the way that “defund the police” attacks did.

The Bottom Line

These gaps define the honest boundaries of what’s known. The evidence supports action but not claims of certainty. Every city that launched successfully launched into some version of these same unknowns and built the answers through operational experience.