Organisational change has a durability problem, not a design problem. Programmes launch competently, land visibly, and then — over the following twelve to eighteen months — quietly unwind, as behaviour reverts to the pattern the change was meant to replace. The standard explanations (poor communication, weak sponsorship, insufficient urgency) describe the surface. This paper argues that change fails to hold because change efforts operate on structures — org charts, processes, systems, messaging — while behaviour is held in place by something structures do not touch: the standing network of commitments, assumptions, and privately held futures that people are already living toward. Drawing on the competing-commitments research of Kegan and Lahey, on the language-action tradition, and on two decades of transformation practice, we set out a model of where change efforts aim versus where behaviour is anchored; reinterpret resistance as fidelity to undisclosed commitments rather than opposition to the new; and describe the discipline — surfacing competing commitments, working the conversational layer, and changing how leaders themselves operate — that separates change that holds from change that merely lands.
Every treatment of organisational change begins with the same statistic: seventy per cent of change initiatives fail. The figure is quoted in board papers, keynotes, and consulting proposals as settled fact. It is not. When Hughes traced the claim to its sources, he found no original empirical study behind it — the number circulates from citation to citation, each borrowing authority from the last, with the primary evidence never materialising.1 The most honest reading of the actual research base is less quotable but more useful: rigorous success-rate data is thin; what exists suggests that transformations judged fully successful by their own leaders are a minority — McKinsey’s long-running global surveys have consistently put sustained success around a third2 — and, critically, that the dominant failure mode is not collapse at launch but erosion after landing.
That last finding deserves more attention than it gets, because it changes the problem. If change mostly failed at launch, the remedy would be better launches: clearer cases, stronger sponsorship, more compelling communication — the standard toolkit. But launches have become genuinely competent. The modern transformation lands: the structure changes, the system goes live, the training completes, the town hall applauds. Then, over four to six quarters, the organisation metabolises the change — and excretes it. Old routing reasserts itself around the new structure. Workarounds bloom around the new system. The new behaviours, performed under observation, decay when observation moves on. Nobody decides to abandon the change. It is simply, gradually, no longer what happens.
A failure pattern with that shape cannot be explained by launch quality. Something is actively holding the old pattern in place — something the change effort never touched.
Look at where transformation effort actually goes: operating model design, process re-engineering, system implementation, governance, communications, training. Every one of these operates on structures — the formal, visible, documentable layer of the organisation. The implicit theory is behavioural: change the structures and behaviour will follow, because behaviour is assumed to be downstream of structure.
The assumption is roughly half true, and the wrong half is expensive. Structures constrain behaviour; they do not generate it. What generates day-to-day behaviour is the layer beneath: the standing network of commitments people already hold — to colleagues, to professional identities, to ways of working that have kept them safe and successful — and the mostly unexamined assumptions about what is really rewarded here, what is really punished, and what this organisation is really like beneath its announcements. Maturana’s observation about living systems applies uncomfortably well to organisations: a system responds to a perturbation according to its own structure, not according to the intent of the perturber.3 The transformation is the perturbation. The commitment network is the structure that determines the response.
Kegan and Lahey gave the commitment layer its sharpest empirical treatment. Their research on what they called immunity to change found that people who genuinely endorse a change and visibly fail to enact it are not weak-willed or hypocritical; they hold competing commitments — typically unarticulated, often self-protective, and entirely rational from the inside — that the new behaviour would violate.4 The manager who endorses delegation and cannot delegate is often committed, beneath articulation, to being the person who cannot be blamed. The team that endorses candour and stays silent is committed to the relationships silence protects. The new behaviour loses, not because the case for it was weak, but because it is competing against a commitment that was never allowed into the ring.
This reframes the most maligned actor in every transformation: the resister. In the standard account, resistance is friction — irrationality, fear of change, dead weight to be managed with a stakeholder matrix. In the commitment account, resistance is fidelity: people are not fighting the future you propose; they are staying loyal to a future they already hold — a promise made to a team, a professional identity built over decades, a hard-won understanding of how to survive this organisation — which your change threatens and which nobody has put on the table.
The practical difference between the two readings is total. Friction is overcome with pressure, and pressure produces exactly what long-running transformation experience shows: compliance under observation, reversion after it. Fidelity, by contrast, can only be worked with by surfacing — getting the competing commitment said out loud, honoured as legitimate, and renegotiated. An unspoken commitment cannot be renegotiated; it can only win quietly. Which is precisely what the erosion pattern looks like from the outside.
The two layers produce two different adoption curves, and the difference is only visible after the programme team has moved on — which is why it so rarely enters the transformation’s own success measurement.
The compliance curve is seductive because it front-loads the good news. Behaviour changes fast when scrutiny is high — dashboards green, adoption metrics climbing — and the transformation declares victory on exactly that evidence. The commitment curve looks worse for the first year: slower, messier, full of surfaced disagreement, because the competing commitments are being worked rather than overridden. The curves cross somewhere after the programme office disbands. Which curve an organisation is on is determined almost entirely by whether the work ever crossed the line in Figure 1.
What does crossing the line look like in practice? Three disciplines, in our experience, carry most of the weight.
First: surface the competing commitments — deliberately, safely, and early. This is structured conversational work, not a survey. Kegan and Lahey’s diagnostic sequence — from the improvement goal, to the behaviours that work against it, to the worry box (“what do I fear would happen if I did the opposite?”), to the competing commitment the fear reveals — is the most reliable instrument we know for it.4 The condition for it working is psychological safety: people will not disclose the commitment that protects them into a room that punishes disclosure — which is why change-readiness and voice are not separate topics but the same topic (see White Paper 06). A change effort that has never heard a competing commitment spoken aloud has not yet begun; it is still negotiating with an empty chair.
Second: work the conversational layer, not just the structural one. A transformation is, concretely, a re-weaving of the organisation’s network of commitments: hundreds of standing promises must be ended honourably, and hundreds of new requests, promises, and declarations must be made — explicitly, with the freedom to decline that makes acceptance mean something.5 Most transformations skip this entirely: the new operating model is announced, and the announcement is mistaken for the thousands of individual recommitments it actually requires. Zaffron and Logan’s formulation is exact: performance correlates with how situations occur to people, and occurrence lives in language — the future the organisation speaks and the futures its people privately inhabit.6 Until the private futures have been spoken to, the announced future is one voice among many, and not the strongest.
Third: the leaders change first, visibly, or nothing else does. In every erosion case we have examined, the reverting organisation was accurately reading a signal: the leaders who announced the change had not changed their own operating — the same meetings, the same escalation patterns, the same behaviours the transformation supposedly retired. People do not attend to the announced change; they attend to whether the people announcing it are subject to it. One senior team that alters its own observable routines — and owns, in public, the difficulty of doing so — outweighs the entire communications budget, because it is the only evidence that the commitment layer is genuinely in play.
A composite case makes the sequence concrete. A financial-services operating-model transformation — second attempt; the first had landed and unwound in the classic pattern — began its second run not with a new design (the design was largely fine) but with three months of competing-commitment work across the top sixty leaders. What surfaced was a single dominant undisclosed commitment: under the old model, seniority had meant owning a vertical, and the new model’s horizontal accountabilities read, privately, as demotion — a loss of the identity most of the sixty had spent careers building. No communication plan could have touched this, because it was never going to be raised in a town hall. Once spoken, it could be worked: the identity stakes were named, honoured, and renegotiated into the new structure’s terms. The second landing held. The design was the same. The layer the work operated on was not.
The second discipline above — working the conversational layer — deserves unpacking, because it is where transformations most reliably substitute a cheap action for the expensive one it stands in for. The cheap action is the announcement. The expensive one is the recommitment, and the two differ in kind, not degree.
Consider what an operating-model change actually changes, at the level of a single senior manager. Her standing promises to three internal customers are now void, though nobody has said so to any of them. Her team’s implicit contract — what they can expect of her, what she will protect them from — has been rewritten by a slide she did not draft. The peers she has traded favours with for a decade now sit in a different reporting geometry, and nobody knows which of the old obligations survive. Multiply by every manager in scope, and the transformation’s real task becomes visible: it is a mass renegotiation of the organisation’s promise network, of which the new structure is merely the occasion.
Handled implicitly — which is to say, not handled — this renegotiation happens anyway, but privately, unevenly, and conservatively: in the absence of explicit new commitments, people default to honouring the old ones, because those are the ones attached to faces they know and reputations they own. The old pattern’s persistence, so often read as attitude, is frequently just this: the old promises were never formally ended, so people of integrity keep them.
The explicit version is unglamorous, conversation-shaped work, and it has a known protocol. Standing commitments in scope are named and honourably closed — thanked, completed, or explicitly released, counterparty by counterparty. New mutual expectations are established as actual requests and promises — specific, dated, declinable — rather than as role descriptions, which commit no one to anyone. And the residue is surfaced: the obligations that fit nowhere in the new model, which otherwise persist as an invisible shadow structure taxing its official successor. Transformations that run this protocol through their top three leadership layers report a distinctive experience: the middle of the organisation, usually the graveyard of change, moves — because for the first time the change has arrived there as commitments people personally made, rather than as an announcement that happened to them.
One further discipline separates practitioners from evangelists: treating reversion, when it appears, as data before treating it as failure. Behaviour that slides back is pointing at something — usually at one of three things, each with a different response. It may point at an unsurfaced competing commitment, in which case the surfacing work was incomplete and resumes. It may point at a load-bearing function the old pattern quietly performed and the new design does not — the workaround as diagnosis — in which case the design, not the workforce, needs amendment. Or it may point at leaders whose own operating never changed, in which case the organisation is not reverting at all; it is accurately tracking its actual instructions. Transformation offices that hold quarterly reversion reviews on exactly these three questions — for at least four quarters past landing — convert their erosion risk into their richest source of design feedback. The ones that declare victory at go-live have simply chosen not to receive the information.
Three qualifications. First, the two-layer account explains erosion — change that lands and unwinds. Changes that fail at launch (genuinely poor design, absent sponsorship, broken economics) fail for the conventional reasons, and the conventional toolkit applies; this paper adds nothing to a transformation whose structural layer is wrong. Second, not all reversion is pathology. Organisations sometimes revert because the old pattern was, in fact, better adapted than the design acknowledged — reversion is data, and a team should read it before overriding it. Third, the surfacing disciplines assume a minimum of voice safety; in genuinely low-trust environments they must be preceded by the trust and safety work described in White Papers 04 and 06, or the surfacing will produce performance rather than disclosure — the compliance curve wearing the costume of the commitment curve.
For leaders, the model replaces the question “how do we drive adoption?” with three sharper ones. What are the competing commitments this change is up against, and have we ever actually heard one spoken? Which of the organisation’s standing promises does this change break, and have they been ended honourably or just overridden? And what, observably, have we changed about how we operate — what would a sceptical employee point to?
For transformation offices, the implication is a redesign of the effort’s centre of gravity: less weight on the announcement layer, a formal workstream for the commitment layer, and success measures that extend at least four quarters past landing — because that is where the actual result lives.
And for the field, the conclusion runs parallel to this desk’s other papers: the decisive layer of organisational life is conversational. Change does not hold because the case was compelling. It holds when the futures people are already committed to have been surfaced, honoured, and rewoven — one conversation at a time, starting with the leaders’ own.
IMPACT THINKING RESEARCH · BY BEN BOTES · WHITE PAPER 05 · 2026
The Change-Readiness Diagnostic reads where change holds and where it slips — in about three minutes.
Take the Change-Readiness Diagnostic → Back to the research desk