Change frameworks

Choosing a change framework

A caution on this whole section. Change models are the weakest-evidenced part of the field. They are useful as checklists and shared vocabulary. They are not predictions, and a program that has "completed step 4" has completed nothing.

Lewin change model

Unfreeze → Change → Refreeze. The foundation; almost every later model elaborates it.

Unfreeze creates readiness: disconfirm the current state while providing enough psychological safety that people can face the disconfirmation without defending. Schein's refinement matters: disconfirmation alone produces denial; disconfirmation plus safety produces learning. The complementary error is relentless bad news that makes people protect themselves.

Change is the transition itself. Competence is temporarily low; performance dips. This is the stage most programs under-resource because it looks like failure while it is working.

Refreeze stabilizes. Update policies, metrics, incentives, onboarding, and stories. Without it the system reverts within about two quarters, and the reversion is remembered as proof that change is impossible here.

On the modern criticism that organizations never stabilize, so refreeze is obsolete: read "refreeze" as institutionalize. The requirement survives — a new behavior must become cheaper than the old one, or it decays.

Health care: unfreezing a ward works when the team sees its own audit data, not national averages. Refreeze happens when the behavior enters the ward round checklist and the induction pack for rotating junior staff. Software: unfreeze with the team's own delivery telemetry; refreeze by encoding the change in the pipeline, templates, and definition of done, so the new way is the path of least resistance. Executive: executives unfreeze themselves months before the organization and then mistake their own readiness for shared readiness. The organization is at stage 0 when the executive is at stage 3.

See topics/lewin-change-model.

Kotter 8-step change model

John Kotter, Leading Change (1996), derived from studying failures.

1. Create a sense of urgency — evidenced, not manufactured. Kotter's claim: over half of transformations fail here.
2. Build a guiding coalition — position power, expertise, credibility. Not a steering committee of delegates; if members send substitutes, you have no coalition.
3. Form a strategic vision and initiatives — explicable in five minutes.
4. Enlist a volunteer army — communicate relentlessly; enable opt-in.
5. Enable action by removing barriers — structures, skills, systems, and managers that block the vision. This is where most programs quietly stop, because the barriers are usually owned by senior people.
6. Generate short-term wins — visible, unambiguous, clearly related, within 6–18 months.
7. Sustain acceleration — do not declare victory early.
8. Institute change — anchor in culture, hiring, promotion, succession.

Kotter's later Accelerate adds the dual operating system: the hierarchy is good at running things and bad at changing them, so a parallel network of volunteers runs alongside, with people in both.

Evidence: weak. Practitioner-derived, rarely tested as a whole. Strongly top-down; assumes leadership knows the right answer, which is false in complex domains.

The steps are outcomes, not activities. Urgency is a state of the organization, not a launch event. "We did step 4" is the model's characteristic failure mode.

Health care: step 5 is the crux. Clinicians usually agree with the vision and are blocked by rotas, estate, IT access, and job planning. Removing those barriers is unglamorous and is the whole job. Software: step 6 maps to shipping something real — pick a win engineers would brag about, not a milestone that appears only in a program report. Executive: steps 1 and 8 are non-delegable. Urgency cannot be delegated to communications, and anchoring cannot be delegated to HR — it happens through who gets promoted.

See topics/kotter-8-step-change-model.

ADKAR change management model

Jeff Hiatt's individual-level model: organizational change happens only when individuals change.

1. Awareness of the need — do I know why?
2. Desire to participate — do I want to?
3. Knowledge of how to change — do I know what to do?
4. Ability to implement — can I do it in real conditions?
5. Reinforcement to sustain — does anything keep me doing it?

The barrier point. The sequence matters: find the first missing element and address it. Interventions aimed past the barrier point are wasted.

The two universal mistakes: training people who lack Desire — they attend, they pass, nothing changes; and assuming Knowledge equals Ability — knowing the technique is not being able to do it under pressure at 3 a.m. with an interruption.

Health care: a new documentation standard. Nurses have Awareness and Knowledge; they lack Ability, because the workstation is two corridors away and the shift has no slack. Training moves nothing; moving the workstation does. Software: adopting a new testing practice. Engineers have Knowledge and lack Desire, because the previous three mandated practices were abandoned. The barrier is credibility, not competence. Executive: executives are strong on Awareness and Desire — they chose the change — and weak on Reinforcement, because reinforcement is boring and happens after attention has moved on.

Evidence: weak, and the model is silent on structural causes. A person can have all five elements and still fail inside a system that prevents the behavior.

See topics/adkar-change-management-model and https://github.com/joelparkerhenderson/adkar-change-management-model.

Beckhard-Harris change equation

D × V × F > R

Dissatisfaction × Vision × First steps must exceed Resistance.

Multiplication, not addition: if any one is zero, the product is zero.

F = 0 is the most common pattern in mature organizations, because leaders invest heavily in D (burning platform) and V (strategy deck) and almost nothing in F, which is unglamorous, local, and specific.

Practical use: score each term 1–5 with the leadership team, then separately with a sample of frontline staff. The gap between the two scorings is more informative than either score — leaders reliably rate D and V far higher, because they have been discussing both for months.

Decompose R rather than treating it as a lump: loss (status, competence, autonomy, relationships, meaning); cost (effort on top of existing workload); risk (what happens to me if this fails); history (previous failed changes and what happened to their champions); and legitimate objection — sometimes the change is a bad idea, and this should be tested before it is treated as resistance.

Health care: D is often extremely high — everyone knows the pathway is broken — while V is contested between professions and F is absent because nobody has authority over the first step. High D with zero F produces burnout, not change. Software: engineers frequently have high D and clear F ("delete this service, automate this gate") but no V, so you get local optimization without direction. Executive: executives own V and can manufacture D. The highest-leverage executive act is supplying F — naming the first concrete step, resourcing it, and removing the specific obstacle.

See topics/beckhard-harris-change-equation.

Bridges transition model

William Bridges distinguishes change (situational, external, happens on a date) from transition (psychological, internal, has no date).

Transition starts with an ending, not a beginning.

1. Ending, losing, letting go. People must let go of the old way and the identity attached to it. Name precisely what is ending and for whom. Identify who loses what: status, competence, relationships, territory, routine, meaning, future plans. Do not denigrate the past — people built it, and attacking it attacks them. Mark the ending explicitly.

2. The neutral zone. Old gone, new not yet operational. Productivity drops, anxiety rises, old weaknesses resurface, and people are simultaneously exhausted and creative. Say out loud that this phase exists and that it feels bad — normalizing it is the single highest-value intervention available. Set short-range goals. Protect people from additional change. Use the phase for experimentation; the system is most plastic here.

3. The new beginning. Cannot be forced or scheduled. Communicate the four Ps: Purpose, Picture, Plan, Part to play.

Phases overlap, and different people are in different phases at once, including within one team.

Health care: a service reconfiguration ends a team that worked together for fifteen years. The rota is the change; the loss of a functioning team and a professional identity is the transition. Clinicians who "resist" a clinically sound reconfiguration are usually in the ending phase and being told they are obstructive. Software: migrating off a long-lived internal system ends the expertise of the people who knew it best. Their reluctance is loss of hard-won mastery, not technical conservatism — name it, and give them a visible role in the new system. Executive: an executive who has been through the ending privately over months announces the beginning on day one; the organization hears a beginning before it has had its ending, and the mismatch reads to the executive as resistance.

Caution: the model can be used to pathologize legitimate disagreement. "They're just in the neutral zone" is a way of not listening. Some resistance is correct.

See topics/bridges-transition-model.

Satir change model

Virginia Satir's model, adapted for organizations by Gerald Weinberg, is the best available account of the performance dip.

1. Late status quo — familiar and predictable, not necessarily good.
2. Foreign element — something the existing model cannot accommodate. First reaction is to reject or minimize it. This denial defends a working model; it is not stupidity.
3. Chaos — the old model no longer works and no new model exists. Performance drops, sometimes below the original baseline. Normal and necessary.
4. Transforming idea — the insight that makes the new situation make sense. Usually arrives from inside the group, and cannot be scheduled.
5. Practice and integration — performance climbs, typically above the old baseline, but only after deliberate practice.
6. New status quo — until the next foreign element.

Three predictable errors in chaos:

• Abandonment — leadership sees the dip, concludes failure, reverts, one week before the transforming idea. The organization learns that changes get reversed.
• Escalation — pressure, targets, and monitoring added, consuming the capacity needed to find the transforming idea.
• Stacking — a second foreign element introduced during chaos because the calendar said so. Two chaoses do not resolve faster than one.

Accelerating the transforming idea: time and permission to experiment at low cost of failure; pairing with someone who has already made the transition; frequent surfacing of partial insights, because the transforming idea usually appears first in one person's workaround; and protected slack, because insight does not happen at 100% utilization.

Health care: an electronic health record go-live produces a documented and substantial productivity drop for weeks to months. Organizations that plan reduced clinic lists during chaos recover faster than those that maintain full throughput and burn out their clinicians. Software: adopting a new language or architecture makes an experienced team temporarily junior; the transforming idea is usually a specific worked example in their own codebase, not documentation. Executive: changing the operating model while simultaneously raising targets stacks a second foreign element onto chaos. Expect a longer, deeper dip — and expect to be told everything is fine.

See topics/satir-change-model.

Kübler-Ross change curve

Five stages of grief — denial, anger, bargaining, depression, acceptance — adapted by consultants into a curve of morale over time during change.

State the caveats whenever you use it. Kübler-Ross's stages were not empirically validated even for grief, and she later said they were never meant as a linear sequence. Bereavement research has largely rejected stage models. The transfer from terminal illness to a software migration is an analogy, not a finding.

The real harm: the model is easily used to dismiss legitimate objections. Labelling a colleague "in denial" converts a substantive argument into a psychological symptom.

What it is good for: building patience and empathy, and normalizing the expectation that morale will dip. That is all.

Better alternatives for the same job: Bridges for empathy with actionable guidance; Satir for the performance dip; immunity to change for treating resistance as a rational competing commitment.

Health care: staff facing a service closure are experiencing actual loss, including of colleagues and professional identity — the closest the analogy gets to legitimate, and the response should still be Bridges' ending work. Software: a team told to migrate to a platform they consider worse are not "in denial"; they may be right. Test the objection before psychologizing it. Executive: the useful takeaway is simply that a dip is expected, has a duration, and is not evidence that the decision was wrong — nor that it was right.

See topics/kubler-ross-change-curve.

Rogers diffusion of innovations

Everett Rogers, 1962. Among the better-evidenced models here, drawn from hundreds of empirical diffusion studies.

Adopter categories: innovators (2.5%), early adopters (13.5%), early majority (34%), late majority (34%), laggards (~16%).

Early adopters are the critical group — respected opinion leaders whose endorsement legitimizes. Innovators are useful for early access and useless as proof, because the mainstream does not identify with them.

The label "laggard" is pejorative and frequently unfair. Late adopters disproportionately include people with the most operational risk exposure and the longest institutional memory. Interview them; they know the failure modes.

The five attributes that predict adoption — the strongest practical part of the model, and a design checklist:

1. Relative advantage — better than what it replaces, in the adopter's own terms, not the sponsor's.
2. Compatibility — fits existing values, experience, and needs.
3. Complexity (inverse) — easy to understand and use.
4. Trialability — can be tried in a limited, reversible way.
5. Observability — results are visible to others.

Before any rollout, score your change on all five and fix the weakest.

The chasm (Geoffrey Moore): early adopters accept incomplete solutions for advantage; the early majority require completeness, low risk, and support. A rollout that succeeds with enthusiasts and stalls at scale has hit the chasm — the answer is completeness and support, not more evangelism.

Health care: clinical practice change spreads through respected clinicians and specialty networks, not management communication. Observability is often low in clinical work, so make results visible with local data. Software: internal platform adoption follows this precisely; trialability and observability matter more than mandates, and a mandated platform with poor relative advantage produces shadow systems. Executive: resist mandating at 15% adoption. Mandating before the early majority has peer evidence converts adoption into compliance, and compliance does not produce the benefits the business case assumed.

See topics/rogers-diffusion-of-innovations.

Switch framework

Chip and Dan Heath's organizing metaphor: a Rider (rational, analytical, easily exhausted), an Elephant (emotional, powerful, seeks comfort), and a Path (the situation).

Direct the Rider — what looks like resistance is often lack of clarity.

1. Follow the bright spots. Find where it already works here and clone it.
2. Script the critical moves. Not "be more patient-centered" but "sit down when speaking to a patient."
3. Point to the destination. A vivid, concrete picture and why it is worth it.

Motivate the Elephant — what looks like laziness is often exhaustion.

4. Find the feeling. Knowing is not enough.
5. Shrink the change. Progress creates motivation more reliably than motivation creates progress.
6. Grow your people. Identity-based change outlasts incentive-based change.

Shape the Path — what looks like a people problem is often a situation problem.

7. Tweak the environment. The highest-leverage and most-neglected tactic.
8. Build habits. Action triggers: "when X happens, I will do Y." Checklists and defaults are habits made external.
9. Rally the herd. Publicize what peers are already doing.

Health care: bright spots — study the ward with the lowest infection rate rather than benchmarking against the national mean. Path — default order sets and pre-filled checklists change prescribing more than education campaigns. Software: scripting critical moves means a concrete definition of done, not "improve quality"; shaping the path means paved roads, templates, and pre-commit hooks. If the right way is harder than the wrong way, exhortation will not close the gap. Executive: shrink the change — a 90-day pilot on one business unit gets agreement that a three-year transformation cannot. Find the feeling — take the executive team to observe the actual work, in person, once.

Caution: "find the feeling" is a persuasion technique and deserves ethical scrutiny.

See topics/switch-framework.

McKinsey influence model

Four conditions, all of which must be present for behavior to change:

1. Understanding and conviction — "I know what is being asked, and it makes sense to me."
2. Reinforcing formal mechanisms — structures, processes, incentives, promotion criteria, budgets, tooling.
3. Talent and skills — practice under realistic conditions, with feedback.
4. Role modeling — leaders and peers visibly behaving the new way, especially under pressure.

The completeness audit. Most failing programs score high on 1 and 3 — communication and training are the easy, purchasable interventions — and low on 2 and 4, which require leaders to change what they reward and how they behave.

If your program's investment is concentrated in the left column, expect compliance without change.

Health care: a patient safety program with excellent training and strong messaging fails when incident reporting still leads to blame and senior clinicians visibly bypass the protocol when busy. Staff read the behavior, not the poster. Software: a quality initiative fails when the promotion ladder rewards shipping features and staff engineers skip tests under deadline. Engineers are excellent at detecting the difference between stated and enacted priorities. Executive: the executive's calendar and escalation behavior is condition 4. If the CEO says "sustainable pace" and emails at midnight, the organization has received an unambiguous instruction.

See topics/mckinsey-influence-model.

Theory of change

An explicit, testable causal chain from what you do to the outcome you want, with the assumptions named.

Inputs → Activities → Outputs → Outcomes → Impact

Build it backwards, and write the assumption between each pair of boxes.

Worked example — a psychological safety program:

Assumptions, stated explicitly with confidence:

• Leaders can change behavior with coaching alone. (Low — test early.)
• Reported safety translates to actual speaking-up behavior. (Medium.)
• Higher reporting reflects better reporting, not more incidents. (Critical — this metric moves the "wrong" way if misread. Pre-agree the interpretation with leadership before the data arrives.)
• Earlier surfacing reduces harm rather than relocating it. (Medium.)

That third assumption is the kind that sinks programs. A rising reporting rate is success, and someone will present it as failure.

Health care: mandatory in most improvement funding, and routinely produced as compliance paperwork. Its real value is forcing agreement on whether a rising metric is good or bad before it rises. Software: a developer experience program whose theory stops at outputs ("we shipped a new CI system") cannot answer whether it helped; chain it to lead time and retention. Executive: ask for the theory of change and specifically the assumptions list before approving any major program. A program that cannot name its assumptions has not been thought through.

See topics/theory-of-change and https://github.com/joelparkerhenderson/outputs-vs-outcomes.

Plan-Do-Study-Act

Shewhart and Deming's improvement cycle, wrapped by the Model for Improvement's three questions:

1. What are we trying to accomplish? A specific, measurable, time-bound aim.
2. How will we know that a change is an improvement? Measures agreed in advance.
3. What changes can we make that will result in improvement?

The cycle:

• Plan — state the change, who does what, where, when, and predict the result. The prediction is mandatory; without it, Study degenerates into rationalization.
• Do — run it at small scale; record what actually happened, including departures from plan.
• Study — compare result to prediction. The learning is in the difference.
• Act — adopt, adapt, or abandon. Then the next cycle.

Scale of testing. The most common error is testing too big. Ramp deliberately: one clinician, one patient, one shift → one clinician, five patients → three clinicians, one week → one ward, one month → spread. "Test with one patient tomorrow" gets agreement that "pilot across the directorate" never will.

Three kinds of measure: outcome (did the thing we care about improve?), process (is the change actually happening? — usually the first thing to check when the outcome does not move), and balancing (what got worse elsewhere? — non-negotiable).

Plot data over time, not before-and-after. Two bars cannot distinguish improvement from normal variation, and that is how most improvement claims are inflated.

Health care: the lingua franca of clinical improvement, and most often degraded into ritual PDSA — paperwork completed, no prediction recorded, one cycle run, called a project. Software: the same logic as a hypothesis-driven experiment; attach the prediction to the ticket. Executive: ask any improvement program for its predictions, recorded before the change. Programs that only report post-hoc successes are marketing, not learning.

See topics/plan-do-study-act.

Improvement kata

Mike Rother's Toyota Kata: improvement capability is a habit acquired by deliberate practice, not a project run by specialists.

The improvement kata — four steps:

1. Understand the direction / challenge — a longer-term aspiration set by leadership. "Zero avoidable harm." "Deploy on demand."
2. Grasp the current condition — factual, measured, at process level, by going to see. Most teams discover they do not know.
3. Establish the next target condition — how the process should operate, by a specific near date, typically one to four weeks out. A pattern of operation, not a result. The hardest step and the one that carries the method.
4. Experiment toward it — rapid cycles, one obstacle at a time. The path is unknown by design.

The coaching kata — five questions, used daily in five to twenty minutes:

1. What is the target condition?
2. What is the actual condition now?
3. What obstacles are preventing you from reaching the target condition? Which one are you addressing now?
4. What is your next step, and what do you expect to happen?
5. When can we go and see what we have learned from taking that step?

The purpose is to build the learner's thinking pattern. A coach who answers the questions has destroyed the exercise.

Health care: a ward's challenge is "no patient waits more than 30 minutes for analgesia"; the target condition in three weeks describes how the process will operate, then daily experiments against obstacles. Software: challenge "deploy on demand"; current condition measured; target condition "pipeline under 15 minutes for the top three services" in four weeks. Executive: the executive contribution is setting the challenge, then practicing the coaching kata with direct reports instead of asking for status. This is the hardest behavioral change in the method and decides whether it survives.

See topics/improvement-kata.