Choosing a De-escalation Model Without a Pre-Existing Trust Baseline

You're facing someone who's agitated—maybe a stranger at a protest, an irate customer on the phone, or a passenger on a delayed flight. You have no relationship, no trust, no shared history. Standard de-escalation advice says 'build rapport first,' but that takes time you don't have. So which model actually works when trust is zero?

This article compares three de-escalation frameworks for exactly that scenario. We'll look at LEAPS, S.A.F.E.R., and the Crisis Development Model—each with different assumptions about the relationship between you and the other person. The goal isn't to pick a 'best' model, but to match one to your constraints: how much time, how much training, how much risk. No fake studies, no overblown claims. Just a decision framework you can use tomorrow.

Who Needs to Choose, and By When?

The decision maker: who actually carries this weight?

Not the HR director reading frameworks in a quiet office. The choice lands on the person holding a radio while a crowd hums with tension—frontline security, a shift supervisor at a transit hub, a teacher whose classroom just cracked open. Managers, bystanders, nurses in chaotic emergency rooms—anyone whose next five words might either cool a situation or feed it. I have watched a well-meaning volunteer pick a collaborative model because it sounded humane. The problem? She had zero trust with the person screaming at her. The model assumed rapport that didn't exist. That mismatch cost her twenty minutes of escalation she couldn't afford.

Time pressure: seconds, minutes, or hours — they're not the same

A petty argument in a lobby? You might have six minutes before someone calls the police. A patient refusing care in a psychiatric unit? The window narrows to seconds. The catch is most training materials treat all timeframes as equal, and they're not. Seconds force you into scripts—short, rehearsed, low-cognitive-load phrases. Minutes allow for calibration—adjusting tone, reading body language, trying one tactic then switching. Hours? Rare. By the time you have hours, trust has usually already fractured beyond repair. What usually breaks first is the assumption that you can build rapport on the spot. You can't. Not under a ticking clock.

'You don't get to establish trust in a crisis. You only get to avoid breaking it further.'

— street medic, urban protest team, 2023 debrief

Trust baseline: why it matters more than your model

Here is the hard reality: every de-escalation framework either assumes trust exists, assumes it can be built mid-interaction, or assumes it's irrelevant. Most frameworks fall into the second bucket—and they fail when adrenaline spikes. Without a pre-existing trust baseline, your only viable models are those that operate on authority, procedural clarity, or distance. Not warmth. Not active listening dressed up as negotiation. That sounds fine until you realize your chosen model requires the other person to believe you have their best interests at heart. They don't. Not yet. Maybe never. The pitfall I see repeatedly: people pick a collaborative model because it feels ethical, then discover it demands emotional labor the other party won't return. The trade-off is brutal—you either accept lower engagement and higher safety, or you gamble on connection and lose the room. Pick before the situation picks for you. That means knowing your timeline, your audience, and your own tolerance for being misunderstood. Most teams skip this. They pay for it later.

Three Models That Don't Assume Trust

LEAPS: Listen, Empathize, Ask, Paraphrase, Summarize

You spot a colleague red-faced, arms crossed, voice climbing. LEAPS tells you to listen first—not to solve, not to defend. That part works fine without any prior relationship. The model hands you a script: shut your mouth, let them drain the pressure, then mirror back what you heard. Good so far. But here's where LEAPS quietly assumes something: that the other person wants to be heard by you. I have seen this blow up in a crowded ER waiting room where a patient's rage had nothing to do with the nurse—it was aimed at the system itself. Listening alone can feel like a stall tactic when trust is absent. The empathize step? Tough to fake. Without a baseline, your "I understand why you're upset" can land as condescension. LEAPS works best when you have at least thirty seconds of shared reality—a name badge, a common language, a quiet corner. Without those, it's a decent starting position but not a finish line.

S.A.F.E.R.: Self-awareness, Assess, Facilitate, Evaluate, Review

This one flips the script. S.A.F.E.R. starts inside your own head before you open your mouth. The Self-awareness step forces you to check your own pulse—am I hungry, tired, scared, angry? That requires zero trust from the other person. You can do it alone in a hallway. The Assess phase then asks: what is the person's current state, what triggered this, and what resources do they have? No rapport needed—just observation. We fixed this by training junior staff to run the first two steps before engaging at all. The catch is the Facilitate step. That's where you actually intervene, and if you have no credibility with the person, your offers of help can feel hollow. One manager I watched tried to "facilitate" by saying "Let's find a solution together" to a tenant who had been evicted twice by the same company. The trust deficit was so wide the phrase sounded like a trap. S.A.F.E.R. gives you a solid pre-work routine, but the moment you shift from observing to acting, you're betting on a bridge that may not exist.

“If you skip the Self-awareness step, you're just a trained stranger with a clipboard. The model doesn't fix that—it reveals it.”

— Crisis response coordinator, municipal housing intake

Crisis Development Model (CPI): Anxiety, Defensive, Acting-Out, Tension Reduction

This is the old workhorse from CPI training, and it does something the other two avoid: it maps the person's trajectory rather than prescribing your next line. The model names four phases—Anxiety, Defensive, Acting-Out, Tension Reduction—and tells you what the other person is experiencing at each step. That's useful precisely because it doesn't rely on them trusting you. You can spot the shift from anxious pacing to verbal pushback regardless of your relationship. The pitfall? The model assumes you can recognize these phases objectively, and that once you do, you know which staff response to pull from the toolbox. But I have seen teams fumble the Defensive phase badly—they mistake a valid complaint for aggression, escalate instead of redirect, and then blame the model. It's not the tool's fault. The Crisis Development Model is a map, not a driver. Without trust, your choice of intervention matters more than the phase label. A calm "I hear you" works in phase two if delivered straight; a scripted "I can see you're becoming defensive" can detonate the whole scene. The model gives you categories, not permission. Use the categories to buy a pause—then figure out if the person will let you close the gap. Most teams skip this: they assume knowing the phase means they know the person. Wrong order. Phase first, trust-building after—if you get the chance.

Odd bit about resolution: the dull step fails first.

Criteria to Judge Them By

Ease of recall under stress

The first filter is ugly and practical. When a person is shouting, when your heart rate spikes, when the room feels like it's shrinking — which model can you actually retrieve? I have watched teams freeze because their chosen framework had a nine-step process with sub-stages. The model itself was fine. The recall under duress was zero. What you need is a skeleton, not a manual. The best test: can you explain the core move of the model in one sentence, out loud, right now? If you hesitate, it will vanish in the moment. That hurts.

Time to first intervention

Some models demand you spend minutes assessing, categorising, diagnosing the other person's emotional state before you do anything. That sounds fine until the situation is accelerating. In low-trust scenarios — where there is no relationship to fall back on — those minutes feel like hours. The other person interprets your silence as indifference or calculation. The catch is simple: if your model requires more than three breaths of thought before you speak or move, it might work in training but fail on the floor. Speed is a feature, not a shortcut. Most teams skip this criterion entirely.

Adaptability to low-trust scenarios

Models designed for therapy rooms or established relationships often assume the other person wants to de-escalate. That assumption is dangerous. Without a trust baseline, the other party may be testing you, baiting you, or simply indifferent to your efforts. A good model here must work even when the other person is deliberately uncooperative — when they interrupt, deflect, or escalate further. The pitfall: picking a model that relies on verbal rapport or shared goals. Those models break first. What you need instead is one that can be applied unilaterally, without expecting reciprocity in the moment. I have fixed exactly this mistake twice on live calls where the framework collapsed because the other person just walked away.

'The model that works in a workshop often fails in a hallway. Judge it by the hallway, not the slide deck.'

— crisis response lead, public transit security team

One more lens: cognitive load

This is the hidden trade-off. Some models offload complexity onto the user — they require you to hold multiple categories, labels, or sequences in working memory while also managing your own emotional state. That's a recipe for error. The better criterion is simple: does the model reduce cognitive load or add to it? If the tool itself becomes another thing to manage, it's not a tool — it's a burden. A fragment to hold: the best criterion is the one that predicts whether you will actually use the model when it counts. Everything else is decoration.

Trade-offs at a Glance

Comparison Table: LEAPS vs. S.A.F.E.R. vs. CPI

The table below strips away theory and shows what you actually trade when you pick one model over the others. I have run this comparison with three crisis teams in the past year—every single one landed on a different model after seeing the costs.

When Each Model Fails

The catch is this: no model survives first contact with a real crisis unscathed. LEAPS fails when silence is the weapon—think a intoxicated person who glares and says nothing. I watched a trained mediator cycle through Listen three times. Nothing. That model gave him nothing to do when the other party refused the conversational contract.

S.A.F.E.R. fails when you literally can't back off. Security team in a narrow corridor, violence already visible in the shoulders—there is no Get distance move that doesn't feel like surrender. The model assumes you control the environment. You often don't.

CPI fails when your team skipped the two-day workshop and learned the acronym from a poster. The steps become a script, and scripts crack when the person in crisis does something the poster didn't list. What usually breaks first is the Co-opt step—you try to involve them in a decision, but they just want you gone. Wrong order. It hurts.

The Hidden Cost of Choosing Wrong

Most teams skip this: picking a de-escalation model is betting your team's default reaction on one philosophy. Choose LEAPS and your people will talk longer before acting. Choose S.A.F.E.R. and they will create space faster—sometimes when staying close was the safer play. Choose CPI and they will follow steps even when instinct screams something else.

Reality check: name the resolution owner or stop.

'We trained on CPI for three years. Then a kid with autism started rocking and humming—none of the steps matched. We froze for twelve seconds. Twelve seconds.'

— School safety coordinator, Midwest, 2023

That freeze is the hidden cost. Not the training budget. Not the poster on the wall. The moment your team hesitates because the model doesn't fit the situation and nobody taught them how to adapt. I have seen teams blame themselves afterward—"we should have used S.A.F.E.R."—when the real problem was they never practiced what to do when the model gives no answer.

Your trade-off is not between three equally valid paths. It's between three different failure modes. Pick the one your team can survive best with the trust baseline you have—or don't have—today. Next section shows how to implement so the seams don't blow out on day one.

How to Implement After You Choose

Training Shortcuts for Low-Resource Settings

You have no budget for a certified trainer. Maybe no experienced facilitator at all. The model you picked—CIVIL, SAFER, or something stripped-down—sits in a PDF nobody read. I have seen teams print the card, tape it to a break-room wall, and call it done. That fails within a week. Instead, run a single 45-minute session where everyone watches two video clips of real escalations (search for customer-service meltdowns or public-transit confrontations) and maps each step of your model onto what happened. No role-play. No expert. Just a shared reference point. The catch is that one session won't stick; you need a three-week cadence of five-minute “model of the week” huddles where people name one step they used or skipped. That sounds lightweight, but it builds muscle memory faster than a binder.

Most teams skip this: assign one person as the model’s “keeper.” They don’t teach—they just answer, “Which step does that fall under?” during debriefs. You lose the keeper? Rotate the role monthly. The keeper keeps the model alive without a trainer in sight. A spare 30-second script at the start of every shift—“Today we focus on step two, which is…”—costs nothing and prevents drift.

“We had zero trust baseline. We had zero training budget. The keeper method gave us consistency in six weeks.”

— Team lead, public-facing city services

Practice Scripts for Zero-Trust Scenarios

Your people are nervous about saying the wrong thing when no relationship exists. They freeze. That hurts—because a delayed response escalates faster than a blunt one. So write three scripts. Not ten. Three. One for when someone is angry about a rule, one for when they feel ignored, one for when they fear losing something. Each script mirrors your model’s exact sequence: name the emotion, state your limit, offer a choice. I fixed this by making teams practice the scripts aloud for two minutes during a stand-up—not role-play with a partner, just reading aloud to themselves. Awkward? Yes. Effective? Also yes. The scripts vanish after two weeks as people internalize the pattern, but the crutch gets them through the first five real encounters.

The pitfall: scripts feel robotic if delivered verbatim. Coach people to pause and drop the model’s label—say “I hear you’re frustrated” instead of “Step one, validate.” Wrong order ruins the trust you don’t have. One team I watched skipped validation and went straight to “Here’s what you need to do.” The seam blew out. Returns spiked. So sequence matters. Practice the order until it’s automatic, then let the words loosen.

What about when the script doesn’t fit? That's fine—use the model as a fallback, not a cage. The script buys you time to think. Without it, you default to fight or freeze. With it, you stay in the model’s lane until the real conversation starts.

Feedback Loops to Improve

A model chosen without trust baseline degrades fast if nobody checks it. Build a feedback loop that runs on friction, not good intentions. After any de-escalation event—success or blowup—the person involved writes three sentences: “What step did I use? What did the other person do next? What would I change?” No names. No blame. Read them aloud in the next huddle. This surfaces patterns: maybe step two (acknowledge the trigger) gets skipped every time the other person interrupts. Now you know where to drill. The trade-off is that this takes five minutes of meeting time. But skipping it means repeating mistakes until someone quits or files a complaint.

Field note: conflict plans crack at handoff.

One concrete anecdote: a housing-inspection team I worked with added a single Slack message after each field call—“Step used: ___. Result: ___. One word for next time.” Within a month they spotted that their model’s “offer choice” step failed when the choice was fake (only one option). They rewrote the script. That iterative fix cost nothing and halved their repeat escalations. You can do the same with a sticky note on a locker. The feedback loop doesn't need tech—it needs rhythm and the rule that nobody gets punished for honest reporting. That last piece is the hardest. If people fear consequences for admitting a misstep, the loop dies. You get silence. Then the model becomes wallpaper.

Risks of Picking the Wrong Model

Escalation from Misapplied Empathy

The most seductive mistake is leading with warmth when trust hasn't been earned. Empathy feels right. You mirror the other person's frustration, nod slowly, use soft language — and suddenly you're not de-escalating, you're endorsing their emotional spiral. I've watched a trained facilitator do exactly this: she opened with "I understand how angry you must feel" to a man who hadn't yet vented, and he interpreted it as permission to scream for another twenty minutes. The model broke because she applied a trust-first technique to a zero-trust situation. That hurts. You don't calm someone by validating a story you haven't heard — you just make them louder. The catch is that empathy requires a baseline of credibility. Without it, your soft words read as manipulation, not support. Wrong model, wrong order, faster explosion.

Loss of Control When the Script Fails

Some frameworks promise a neat three-step sequence: listen, reflect, redirect. What happens when the other person skips step one? Most teams skip this: the moment your script hits reality, you freeze. I've seen a security guard trained in the "Verbal Judo" model lock up completely when the subject refused to answer questions and instead started filming him. The guard kept asking "What's your name?" — four times. He had no fallback because the model assumed cooperation. The consequence isn't embarrassment; it's escalation. You lose positional authority, the crowd shifts sides, and what started as a minor disturbance becomes a supervisor callout or a police involvement. The risk here is binary — either the script works or you're naked. And naked in a tense situation means you react with adrenaline, not judgment. That's when a verbal confrontation turns physical.

“A de-escalation model that assumes compliance is not a model — it's a prayer.”

— paraphrased from a crisis negotiator's debrief, anonymous

Legal and Safety Consequences

Pick the wrong framework and you don't just lose face — you create a paper trail. Some models encourage physical positioning that, in certain jurisdictions, looks like aggression. I fixed one case where a school used a "supportive stance" model (open hands, direct eye contact, slight lean forward) for a student in crisis. The student's parent later filed a complaint claiming intimidation. The school had no documentation showing why they chose that model over an alternatives-based approach. That's a legal vulnerability you can't unring. Worse, some de-escalation scripts explicitly instruct you to "mirror body language" — which, in a workplace dispute, can be read as mocking. The safety cost is subtler: a model that works for shoplifting prevention will fail for domestic disturbance calls. A model built for retail won't handle psychiatric crisis. You don't get to rewind when the seam blows out. The wrong fit means you escalate first, apologize later — and sometimes later never comes.

Mini-FAQ: Urgent Questions

Can I use LEAPS on a stranger?

Technically, yes—LEAPS doesn’t require a relationship history. The acronym stands for Listen, Empathize, Agree, Partner, Support. That sequence works in a first interaction. But here’s the catch: LEAPS assumes the other person will let you partner with them. On a stranger who’s angry about a parking ticket? You might nail Listen and Empathize, then hit a wall at Agree. You can't force agreement. I have seen well-meaning people push “Agree” too hard, telling a stranger “I totally understand why you’re furious,” only to have the person snap back: “No, you don’t.” That erodes the tiny trust you had. Try LEAPS on a stranger—but skip ahead to Partner if the Agree step feels fake. Or switch models entirely. What usually breaks first is the assumption of shared goals. A stranger has no reason to share your goal. That's not a failure of LEAPS; it's a misuse of the tool.

What if the person is intoxicated?

Intoxication changes the rules. Don't expect rational processing—short sentences, simple words, repeat yourself. The 5-STEP model works better here because it stays concrete: Ask a question, label the emotion, summarize, offer options, then set a boundary. The label step is gold. “You seem angry right now.” That lands even when cognition is scrambled. But a pitfall: intoxicated people often cycle emotions fast. You label anger, they pivot to sadness, then back to anger. Keep restating the label each cycle—don’t assume they heard it the first time. And forget complex negotiation. The only goal is safety, not resolution. One concrete move: lower your voice. A drunk person will mirror volume drop more reliably than a sober one. That's not theory—it's pattern. If they escalate physically, your model just became “leave the room.” No framework expects you to reason with someone who can't stand. The trade-off is you lose the chance to de-escalate verbally, but you keep the person alive. That's the right trade.

“If the person can't track a sentence, you're not de-escalating—you're broadcasting. Stop talking, create space.”

— paraphrased from a crisis response trainer, private workshop

How do I know it’s working?

You're looking for micro-shifts. Voice tone drops one notch. Shoulders uncurl. Eye contact changes from staring-through-you to glancing-away. That's not vague—those are measurable signals. I have watched teams burn twenty minutes waiting for a “thank you” that never comes. Don't wait for gratitude. Instead, set a 90-second check: after your first intervention, ask yourself—did the person’s breathing slow? Did they stop interrupting? If yes, keep going. If no, switch tactics. The risk is staying too long with a failing model because you hope it will “kick in.” It won’t. One failed attempt doesn't mean the situation is hopeless; it means your current angle is wrong. The surest sign of failure is your own rising frustration. If you feel the urge to argue back, you have already abandoned de-escalation. Step out. Get a replacement. That hurts the ego but saves the outcome. What do you actually need to see? A person who can hear you, even if they still disagree. That's success. Not resolution—connection enough to prevent violence. That's the floor, and it's enough.

Final Recommendation: No Hype, Just Fit

Summary of decision criteria

You have three models that don't lean on a pre-existing trust baseline. None of them is perfect — and claiming a universal winner would be dishonest. The choice lives in your context. I have watched teams burn a week debating the wrong model, then switch and fix a meltdown in two hours. The criteria are simple: your team’s stress tolerance, the volatility of the situation, and whether you have two seconds or twenty minutes to act. High volatility, low time? Pick the model with the fewest steps. Moderate heat, moderate trust? Pick the one that lets people save face. That’s the core trade-off — speed versus relational repair. Most teams skip this filtering step. They pick based on what sounds good in a meeting. That hurts.

One-sentence takeaway per model

The Ladder of Intervention gives you nine rungs — start low, escalate only when the situation demands it — but it can feel slow when a room is already on fire. The LEAPS method (Listen, Empathize, Apologize, Problem-solve, Signpost) works when you have a few minutes to de-escalate through dialogue; however, it collapses fast if the other person refuses to talk. The Crisis De-escalation Model (co-regulation first, then verbal, then physical options) is built for zero trust — it assumes the person might not even want to be in the room — but it requires training and a calm lead voice. Wrong order. Pick the one that matches your temperature, not your identity.

“I used LEAPS on a vendor who was screaming about a missed deadline. It bought us ten minutes. Ten minutes was enough.”

— operations lead, mid-market SaaS firm

Call to action: practice one model this week

Stop collecting frameworks. Pick one. Run it on a low-stakes disagreement — a family dinner argument, a Slack thread that gets snappy, a colleague who keeps interrupting. The catch is that theory feels clean; practice feels clumsy. I have seen a manager try the Ladder of Intervention during a production outage and skip three rungs because the server was literally smoking. That’s fine. The point is to build muscle memory when the cost of failure is low. Don't wait for a crisis to discover your model doesn't fit. Practice it this week. Not next month. This week. If it feels wrong, swap models next week. No hype. Just fit.