Helping people with serious mental illness hinges on giving better tools and guidance to those who take emergency calls.
On a sunny day in early June 2025, practitioners, policymakers and researchers from across the U.S. gathered at NYU’s Kimmel Center to share insights in the growing movement to reassign low-acuity 911 calls from police to unarmed crisis response teams. In his keynote address, New York City Public Advocate Jumaane Williams didn’t mince words: New York City, he said, needs to expand its B-HEARD program — which dispatches trained behavioral health teams instead of police officers to 911 calls involving people in crisis — if it wants to better serve the 1 in 5 New Yorkers with psychiatric illness.
His words were met with applause, reflecting the growing enthusiasm for alternative response programs across the country. Some cities have even created new Departments of Community Safety to coordinate these teams and other nonpolice responders under one roof. But that enthusiasm — and the enthusiasm for alternative response expressed by Mayor-elect Zohran Mamdani, who wants to create such a department in New York City and send social workers to respond in cases of certain behavioral health crises — stands alongside mounting real-world evidence that cities are struggling to make these programs work.
Those who believe in the promise of alternative response to reduce the scope of policing and offer a more suitable response to a range of social problems must grapple seriously with the practical challenges it introduces to underlying systems, most specifically 911 dispatch. Otherwise, good intentions will collide with messy realities, producing less-than-ideal results.
A recent audit by the New York City comptroller found that even in the limited pilot phase, B-HEARD teams were unable to respond to more than a third of eligible calls. In Louisville, the U.S. Department of Justice found that most 911 calls that could go to behavioral health teams still end up with police. And in Phoenix, DOJ investigators reported that the majority of behavioral health calls continue to be sent to patrol officers — even when clinical teams are available. In short, cities are building the programs — but struggling to deliver the response.
Policymakers often point to funding gaps, staffing shortages or limited hours of operation as reasons why alternative response programs are underutilized. Those are real constraints, but they overlook something more fundamental: These programs only work if the right emergency calls reach the right responders. That decision largely falls to 911 dispatchers, who are the frontline workers judging which situations are appropriate for unarmed responders and which are not. And when there’s any doubt, they often default to sending the police.
Williams understands this challenge well. In a public statement, he acknowledged, “It can be difficult to discern over the phone what is happening at the scene, and whether there is a risk of harm to the caller or to the responders … It is imperative to ensure that 911 dispatchers are properly trained in how to effectively determine which calls can be sent to B-HEARD.”
As more jurisdictions roll out alternative response programs, they’re grappling with the same concerns Williams raised. Many have turned to structured protocols — scripts, decision trees and checklists — to guide 911 dispatchers through call diversion decisions, much like physicians now go through checklists when assessing many medical conditions. Protocols can be useful for structuring and standardizing decision-making, but only when they’re designed with care, tested against real-world practice and supported with additional forms of guidance and training — elements that largely have been missing from alternative response program implementation so far but urgently are needed.
A structured, rule-bound approach to call-taking isn’t a new idea. For decades, public officials and police reformers have treated dispatch as a clerical task that can be mastered through rigid adherence to procedure. In his 1950s textbook, “Police Administration,” influential police chief O.W. Wilson put it plainly: The dispatcher “makes no actual decisions regarding an operation but merely classifies it and applies the routine procedures established by department regulations for this classification in the deployment and assignment of men.” That conception is far removed from the practical realities of the job — yet its legacy lives on in today’s call-taking protocols that treat complex judgment calls as if they can be reduced to clerical tasks.
To better understand how these protocols function in practice, we conducted research in two cities at the forefront of alternative response: Denver and San Francisco. Drawing on interviews and fieldwork into Denver’s Support Team Assisted Response (STAR) program and San Francisco’s Street Crisis Response Team (SCRT) program, we found that their protocols can work reasonably well under ideal conditions: when callers provide clear, accurate information, are invested in the outcome of the call and share a mutual understanding of subjective terms like “violence” or “threat.”
Most crisis calls, however, don’t unfold that way. They’re messy, ambiguous and full of subjectivity. In those moments, today’s protocols often fall short — and can even become obstacles to sending out alternative responders altogether.
Take third-party calls. These are calls that come from passersby, business owners, neighbors — people who often lack the detailed knowledge that protocols demand. As a dispatcher in San Francisco explained: “You’re asking a third-party caller, ‘Is that guy who’s ranting and raving and throwing stuff out of the gutter suicidal?’ and they’re like, ‘I don’t know.’” She noted that in her city, the protocol treated “I don’t know” not as a neutral answer but as a risk — automatically escalating the response and ruling out a street crisis team, even when she believed it was appropriate.
Or take calls about “violence.” The term sounds clear, but its interpretation often depends on context. As another dispatcher in San Francisco explained, “If someone has a weapon and seems violent, we’re not going to send Street Crisis (San Francisco’s unarmed behavioral health response team). But that doesn’t mean yelling and screaming. That doesn’t mean having a stick. As long as they’re not brandishing it. So, it can be really subtle.” The problem is that today’s protocols rarely accommodate that kind of nuance. Most rely on rigid yes/no prompts, and dispatch leaders have largely abdicated their responsibility for helping dispatchers make sense of the wide range of responses they actually hear from callers.
In response to these protocol challenges, we observed a range of adaptations by dispatch staff. Some of the most sophisticated dispatchers developed creative workarounds — reframing questions, stretching definitions or entering responses after they’d decided who to send — to dispatch an alternative response despite the system’s constraints. Others, facing similar pressures, defaulted to sending police, either by avoiding the protocol altogether or deviating from it. These varied practices point to a broader issue: The protocols, as currently designed, are not offering consistent or adequate support.
So what is the better way? Our research, and that of others, points the way.
As the police rulemaking literature reminds us, well-crafted rules can play a critical role in guiding officials’ discretion and promoting both appropriate and equitable treatment for those who call upon the government. For example, many policing agencies now use vehicle pursuit policies that structure discretion by instructing officers to consider environmental conditions and public safety risks before initiating a pursuit. Dispatch centers adopting alternative response can take a similar approach by redesigning their protocols to account for the messy realities of 911 calls and the nuanced judgments call-takers must make in real time.
First, agencies should refine the questions dispatchers ask, drawing directly from the people doing the work. How do experienced dispatchers phrase their questions to surface need rather than threat? What conversational techniques help shift the trajectory of a call toward alternative response? One dispatcher we spoke with described rephrasing her protocol questions to focus on what the subject of the call needs — not just what they’re doing. This call-taker believed a small change like this avoided unnecessary police dispatches.
To capture and share this kind of insight, agencies need real feedback systems that help dispatchers learn from everyday calls. That means setting aside time for supervisors and peers to review calls together — not just when something goes wrong, but also when a call goes well — to talk through what worked, what didn’t and why.
One promising approach to achieve this, adapted from mediation and conflict resolution settings, is the Conversation Analytic Role-Play Method (CARM). This method provides agencies with a framework to review recordings of real conversations in slow motion and build an evidence base for what conversational practices are most effective. A version of this technique could help 911 centers learn from their own calls and improve the wording of their protocols accordingly. Agencies could start by analyzing their call-for-service data to flag dispatchers who are especially skilled at diverting calls to alternative responders, then listen to how those calls unfold. These kinds of structured feedback loops could help improve protocols and strengthen alternative response programs nationwide.
Second, cities should supplement protocols with more flexible forms of guidance. Rigid scripts alone won’t cut it in complex situations. Dispatchers need a principled compass to follow in moments of ambiguity. Take Washington, D.C., where the police department’s use-of-force policy opens with a clear moral anchor: “to value and preserve the sanctity of human life.” Dispatch centers similarly should adopt value statements — ones that pair a commitment to safety with a presumption in favor of nonpolice response. The balance between these values is delicate, but making the organization’s preferences more explicit would help guide call-takers when the rules run out. Perhaps with such a value statement, the San Francisco dispatcher who struggled over how to handle the call about the man “ranting and raving and throwing stuff in the gutter” could have leaned on a departmental preference for a nonpolice response in that moment.
This kind of framing matters. Instead of positioning diversion as a neutral sorting task (“send it to the right responder”), protocols could make clear that if a situation can be handled without police, that’s the preferred outcome.
Definitions matter too. Vague terms like “violence,” “threat” or “weapon” leave too much room for guesswork. Agencies should clarify what these concepts mean in practice — and ensure they’re informed by the real-world experiences of alternative responders. Clearer definitions act as value statements too, helping dispatchers interpret gray areas with confidence and care.
And when a script truly falls short? Dispatchers shouldn’t be left to figure it out alone. Agencies should invest in real-time support, like access to supervisors or embedded behavioral health specialists who can help dispatchers think through tough calls in the moment. There may even be a role for carefully designed AI tools to assist dispatchers here. Denver, for example, has begun testing software from Corti AI to analyze past 911 calls and better identify ones that could be eligible for diversion in the future.
Integrating alternative response into the 911 system marks a major shift in how cities handle crisis calls — one that’s still in its early stages. It’s no surprise that initial forms of call diversion guidance have fallen short of what’s needed to ensure alternative response reaches its full potential. For these programs to expand, they will require a more sophisticated approach to 911 rule design: one that is grounded in reflection, shaped by frontline insight and open to revision.