Almost every change effort gets a launch. New system goes live, new structure is announced, the all-hands happens, the slide deck circulates. Then attention moves on, and within a year or two, an uncomfortable number of those efforts have quietly unwound. Sustaining and embedding change is the unglamorous discipline of stopping that drift: turning a one-off project into the normal way work gets done, so it survives without the people who launched it standing over it.

The quick version

  • Sustaining change means the new way of working is still happening months after the launch energy has faded, not just at go-live.
  • Embedding means it has become "how we do things here": baked into systems, habits, hiring, rewards and stories, so it holds even when leaders stop pushing.
  • The evidence is humbling. McKinsey's surveys repeatedly find fewer than a third of transformations both improve performance and hold the gains over time. The fade is the norm, not the exception.
  • The trap is declaring victory too early. A pilot working is not the same as the change sticking, embedding is a separate phase that has to be planned and resourced, not assumed.

The idea in depth

The classic map of change ends, deliberately, with making it permanent. Kurt Lewin's three-step model, unfreeze, change, refreeze, put the final emphasis on "refreezing": stabilising the new state so the organisation doesn't snap back to the old equilibrium. The picture is a block of ice you melt, reshape, and re-set. Whatever you think of the metaphor, Lewin's instinct was right about the danger: a system under tension tends to return to where it was unless something locks the new shape in place.

The practical move is to treat "after launch" as a phase with its own plan, not as the absence of work. Before you celebrate go-live, write down what would have to be true in twelve months for this to count as embedded, which behaviours, in which roles, visible in which data, and assign someone to own that horizon after the project team disbands. If nobody owns month eighteen, month eighteen owns you.

John Kotter sharpened the same idea into a warning. In his much-cited "Leading Change: Why Transformation Efforts Fail" (Harvard Business Review, 1995), two of his eight common errors are about the end of the journey, not the start: declaring victory too soon, and failing to anchor changes in the corporate culture. Kotter's argument is that change only lasts once it's woven into the shared norms of the place, into "the way we do things around here", and that until new behaviour is rooted in social norms and shared values, it stays fragile and degrades the moment the pressure for change lets up.

Which points to a second move: make the new behaviour the path of least resistance. Don't rely on willpower or memory; change the surrounding machinery so the new way is simply easier than the old. That means retiring the old tool rather than running it in parallel, updating the standard operating procedure and the induction pack, and, crucially, making sure the people who get promoted are the ones who embody the change. Nothing teaches an organisation what really matters faster than watching who rises.

flowchart LR
  A(["Unfreeze
create readiness"]) --> B(["Change
new ways of working"])
  B --> C(["Refreeze
stabilise the new state"])
  C -.->|"no reinforcement"| D(["Drift back
to the old way"])
  C ==>|"anchored in systems
habits & rewards"| E(["Embedded
'how we do things here'"])
Lewin's third step is where most efforts quietly fail, refreezing only holds if the new state is anchored, not just announced. Leaders Loop

Why the gains fade, and what holds them

The numbers are sobering. Across McKinsey's long-running research on transformations, fewer than one in three leaders say their organisation's change effort both lifted performance and sustained that improvement over time, a figure McKinsey notes has barely moved across many years of surveys (see their research on successful transformations, drawing on a global survey of just over a thousand executives in 2021). The widely repeated claim that "70% of change fails" comes from this same family of findings. The headline isn't that change is impossible; it's that holding change is where most of the loss happens.

A pilot working is not the same as the change sticking. Embedding is a separate phase, plan it, resource it, and assign it an owner.

What separates the durable third? McKinsey's own analysis points repeatedly to the same reinforcers: engaging frontline managers and staff (not just the executive sponsor), clear ownership of roles, consistent communication, and, tellingly, committing more rather than less, sustaining the effort past the point where it feels comfortable. The practical reading is that embedding isn't a single mechanism but a stack of mutually reinforcing ones, each individually weak and collectively hard to dislodge.

The move here is to build redundancy into how the change is held. Don't pin survival on one charismatic sponsor or one clever incentive; layer several reinforcements, a metric on a dashboard a senior leader actually reviews, the new process embedded in the system so the old one isn't an option, a short ritual (a weekly stand-up, a monthly review) that keeps the behaviour visible, and the story retold often enough that newcomers absorb it. Any one of these can lapse without the whole thing collapsing.

flowchart TD
  A(["The change you
want to last"]) --> B(["Systems & process
old way retired"])
  A --> C(["Metrics
a leader reviews"])
  A --> D(["Habits & rituals
keep it visible"])
  A --> E(["Rewards & promotion
who rises here"])
  A --> F(["Stories & onboarding
newcomers absorb it"])
  B & C & D & E & F --> G(["Embedded: holds even
when leaders stop pushing"])
Embedding is a stack of reinforcers, not a single lever, any one can lapse without the change collapsing. Leaders Loop

An honest limitation. Treat the famous "70% of change fails" figure with care. It is genuinely contested in the academic literature, researchers including Mark Hughes have argued in Leadership (2016) that the number has been repeated far more often than it has been rigorously evidenced, and that its origins are murky. Lewin's "refreeze" metaphor draws fire too, for implying organisations ever sit still in a world of continuous change. Use these models as lenses that direct your attention to the after-launch phase, not as precise measurements or laws of nature.

A worked example

Take a 400-person services firm rolling out a new CRM to replace a tangle of spreadsheets. (Illustrative scenario; the figures are for teaching, not a real company.) The launch goes well: training sessions run, adoption hits a healthy 85% in the first month, leadership declares the project a success and stands the steering group down. The project lead rolls onto the next initiative.

Six months later, login data tells a quieter story. Active use has slipped toward 50%. Several senior salespeople have drifted back to private spreadsheets because "it's faster for me," and because the spreadsheets were never actually taken away. New joiners weren't trained on the CRM at all, onboarding still pointed at the old shared drive. Nobody was reviewing usage, so nobody noticed the slide until a quarterly forecast came in visibly wrong.

The fix isn't another big launch; it's embedding the change the first push skipped. Retire the old shared drive so the spreadsheet shortcut closes. Put a single adoption metric on the sales leader's weekly dashboard, so drift is seen in weeks, not quarters. Rewrite onboarding so every new hire starts in the CRM by default. And tell the story straight in the next sales meeting, including the forecast that went wrong, so the cost of the old way is concrete, not abstract. None of these is dramatic. Together they turn a fading rollout into the firm's actual way of selling.

Frequently asked questions

When is a change actually "embedded"?

A useful test: would it survive its sponsor leaving? If the new way only holds because one senior person keeps pushing, it's sustained-by-effort, not embedded. Embedded means it's carried by systems, habits, onboarding and incentives, so a newcomer who never met the original champion still does it the new way because that's simply how the place works.

How long does embedding take?

Longer than the launch, and longer than most plans allow. There's no universal number, and anyone who gives you a precise one is guessing, but the practical guidance from the research is to sustain attention well past the point it feels finished. A reasonable planning assumption is that the embedding phase outlasts the implementation phase, and that the first renewal of staff (new joiners, internal moves) is the real test of whether it took.

What's the single biggest cause of change unwinding?

Declaring victory too soon, Kotter's seventh error. The launch high feels like the finish line, leaders redeploy their attention and the project team, and the reinforcement that would have held the change simply never gets built. The old equilibrium does the rest. Guard against it by deliberately separating "we launched" from "it stuck," and keeping resource on the second.

How do you sustain change without exhausting people?

By shifting from effort to infrastructure. Sustaining through sheer willpower, constant reminders, nagging, heroic monitoring, burns people out and fails the moment you look away. The point of embedding is to make the new way the default, so it costs less energy than the old one, not more. If keeping a change alive feels relentless, that's a signal it isn't embedded yet, it's still being propped up.

Can you embed change without changing the culture?

Up to a point. Process, systems and incentives can carry a lot, and changing those is often more tractable than "changing the culture" head-on. But Kotter's point stands: if the new behaviour openly contradicts what the organisation rewards and celebrates, the culture eventually wins. The durable approach is to align the visible signals, who gets promoted, what gets praised, which stories get retold, with the change, so culture reinforces it rather than fighting it.

Related in the Toolkit

Embedding is the last and most-skipped phase of the frameworks covered in change models, and it depends on the groundwork laid earlier, the resistance you defused and the adoption you built (managing resistance & driving adoption) is exactly what determines whether a change holds once the pressure lifts.

Where to go next