Who Says It’s Done?
The transformation story can be regenerated faster than the transformation can produce evidence.
The answer that keeps everything open
There is a question that changes the atmosphere in a transformation meeting: when are we done? Not when the programme closes, the system goes live or the new structure appears on the organisation chart. When can we say that the transformation has produced what we said it would?
The answers are familiar. Adoption needs longer. The capabilities are still embedding. Benefits lag delivery. Another tranche is already underway. The organisation is still transforming. Often, all of that is true, which is why the question is harder than it sounds.
The cheap accusation is that advisers keep transformations running because they are paid to do so. I do not believe that. It requires conspiracy where incentives are enough, and it lets everyone inside the organisation off the hook. The adviser loses revenue when the transformation ends. The internal transformation function loses some of its reason to exist. The sponsor moves from leading a visible strategic priority to being judged on what it actually produced. The organisation gains something too: while the transformation remains underway, disappointing performance can still be described as transitional. The benefits have not landed yet. The capabilities have not matured yet.
“Not yet” is an extraordinarily useful answer because nobody needs to lie for it to persist. Everyone can behave professionally inside a genuinely complex system where almost every legitimate reason for waiting points in the same direction.
I know that system because I have worked inside it. I have written transformation cases, target operating models, roadmaps and progress narratives. I have taken contradictory organisational reality and made it coherent enough for people to act on. I have also written a future into the present tense. Anyone who has done serious transformation work will recognise the craft: a target operating model describes an organisation that does not yet exist; a roadmap gives sequence to events nobody can guarantee; a progress narrative turns incomplete signals into an intelligible account of movement. Transformation practitioners operate in the distance between reality and intention, and part of the judgement lies in knowing how large that distance can responsibly become.
AI changes that distance.
What the artefact used to carry
The obvious AI argument is about speed. A transformation case that once took weeks can be drafted in an afternoon. Operating models, roadmaps, benefits narratives and executive updates can be produced and revised at a pace no human team could sustain. Useful, impressive and increasingly ordinary.
The more important question is what used to happen while those artefacts were being made. Producing the transformation case forced an argument about what was actually wrong. Evidence had to be assembled, assumptions collided and somebody decided which uncomfortable facts belonged in the story. Producing the operating model forced arguments about authority: who makes this decision, where does this work belong, what becomes central, what remains local, whose control increases and whose diminishes. Building the roadmap forced scarcity into the room because five priorities could not all happen first. Even the progress narrative forced somebody to decide what they were prepared to put their name behind in front of people capable of challenging it.
Some of that effort was bureaucracy. Some was waste. Some was theatre. But some of the friction was doing transformation work.
The old sequence was confrontation → negotiation → artefact → commitment. AI can produce the artefact without proving the preceding transformation work happened.
That does not mean the old process deserves nostalgia. Human effort was never evidence of serious thinking. We have all seen workshops consume weeks while avoiding the only decision that mattered, and AI may improve this by exposing contradictions earlier, generating alternatives and freeing practitioners from drafting. The point is narrower: some confrontation used to arrive bundled with production. Once production becomes cheap, that confrontation has to be designed deliberately rather than assumed to have happened.
A polished proposal can arrive with much of the argument already buried inside its structure and language. The operating model may look coherent before the organisation has confronted the trade-offs that would make it real; the roadmap may be beautifully sequenced before the people controlling its dependencies have committed to honour them. The governance system sees the artefact, not necessarily the confrontation behind it.
The quality of the artefact therefore begins to tell us less about the depth of transformation work behind it. That is where a productivity story becomes a verification story.
“We are always transforming”
Most experienced practitioners will object here, and reasonably so. Organisations do not transform once and return to a stable state. Markets move, technology develops, regulation evolves, competitors act and new capabilities create possibilities that did not exist when the transformation began. A large transformation may itself evolve substantially while it is underway. One tranche may be embedding while another begins. Some benefits arrive quickly; others mature much later. Many organisations therefore maintain permanent transformation functions because transformation capability itself has become permanent.
That premise is sound. The escape it can create is not.
A permanent transformation capability does not require permanently unjudgeable transformation. Nor does a large transformation need one artificial moment at which everything freezes and somebody declares the organisation transformed. Serious benefits-management practice already gives us a better unit of accountability. Benefits can be managed at programme level, within tranches and within discrete components; individual benefits can carry baselines, targets, owners and benefits-realisation milestones after the relevant outputs and capabilities have had enough time to embed for meaningful measurement.
That does not make the transformation discrete. It makes judgement recurrent.
A transformation may continue for years and legitimately evolve while it does so. But particular promises eventually have enough time to encounter reality. At that point the organisation should be able to say whether the benefit was realised, partially realised, not realised, or whether the original proposition was explicitly superseded because circumstances changed sufficiently to justify a different one. Then the transformation can continue.
What cannot happen is for the previous claim simply to dissolve into the next version of the transformation.
Continuous transformation is not the opposite of verification. It is the reason verification must recur.
When the story can outrun the benefit
This matters because the profession already knows how benefits are supposed to work. Good benefits practice establishes baselines and targets, assigns ownership, tracks realisation during and beyond delivery, and creates moments at which evidence should be reviewed. The problem is not an absence of benefits methodology. The problem begins when the evidence arrives.
Suppose a material part of a transformation was expected to produce a measurable benefit after sufficient embedding. The milestone arrives and the benefit is weak. Perhaps the assumption was wrong. Perhaps implementation was partial. Perhaps the environment moved. Perhaps the benefit genuinely needs longer. Any of those explanations may be true, and a serious transformation must be able to evolve in response.
But something should still happen to the previous claim. Instead, transformation can absorb disappointment remarkably well. The benefit horizon moves. The definition broadens. Attribution becomes more complicated. Another tranche begins. A new transformation priority appears. Each adjustment can be defensible on its own; together they can make the original proposition almost impossible to find.
The discipline we need is not rigidity. It is memory. Preserve what the organisation believed before it knew what happened, record why the proposition moved, and judge what became of the benefit that justified the original commitment. A transformation should be free to evolve without being free to rewrite its own history.
AI makes this more urgent because the transformation narrative can now move extraordinarily quickly. Operating models, roadmaps, benefits narratives and transformation cases can all be regenerated around new priorities with remarkable fluency.
Benefits do not mature at machine speed. Capabilities still have to embed, behaviours become real and economic effects emerge before many benefits can be meaningfully observed.
“The transformation story can be regenerated faster than the transformation can produce evidence.”
That asymmetry changes the verification problem. AI may make transformation easier to renew than to verify. If every new narrative is cleaner, more coherent and better aligned to the latest strategy, the organisation can become progressively better at explaining where it is going while becoming progressively worse at remembering what previous versions said would happen.
The latest narrative has the advantage of hindsight. Verification therefore becomes partly a memory problem: can the organisation still recover the original baseline, target, assumptions and benefit logic? Can it distinguish a transformation that adapted intelligently from one that repeatedly escaped judgement?
Something has to be at stake
Before AI, authorship carried information. Not proof and certainly not independence. Practitioners are interested parties too; we are paid by transformations, we want interventions to succeed and we can become captured by the coherence of our own story.
My reputation was never verification, but it was a constraint. If I wrote the progress narrative, there was a limit to how far I could allow it to outrun reality before somebody could challenge me personally on the gap. My professional standing entered the transaction. AI has no equivalent stake.
That is not an argument that machines cannot judge. AI should become extremely useful in transformation verification: interrogating operational evidence, comparing promises with outcomes, reconstructing previous claims, identifying movements in baselines and challenging causal stories against data. It may eventually be better than many humans at deciding whether a body of evidence supports a transformation claim.
The meaningful boundary is not machines produce content; humans produce judgement. That boundary will not survive. The more durable boundary is accountability. When an organisation commits itself to a conclusion — this benefit is real, this part of the transformation worked, this one did not, this proposition has been superseded — the judgement has consequences. Budgets move, priorities alter, careers are affected and future transformations are approved partly because of what the organisation believes it learned.
An AI can contribute to that judgement. It cannot be answerable for it. A judgement without an accountable owner is merely output.
The machinery exists. The verdict does not.
None of this is entirely new. The UK National Audit Office warned in 2018 that transformation programmes can have vague intended outcomes, evolve over time and make it difficult to measure real impact or know when a programme has succeeded or should close. Gate Review 5 addresses operations and benefit realisation; benefits-management guidance establishes owners, baselines, targets and realisation milestones; the 2026 Magenta Book pushes evaluation towards establishing the basis of evaluation before results are known and making later revisions visible.
The profession therefore has tools. What it does not have is an equally accepted discipline for converting that evidence into a verdict on the transformation claim. A Benefits Realisation Plan can tell us what to measure. A dashboard can tell us what happened. Evaluation can improve our confidence about attribution. A benefits owner can be accountable for realisation. But someone still has to connect those things and say what they mean for the transformation, with failed remaining an available answer.
Credible verification does not require certainty, but it should leave certain things difficult to hide:
- the benefit that was expected, and which part of the transformation was expected to produce it
- the baseline, target and causal assumptions that existed before the result was known
- the point at which sufficient embedding was expected for meaningful measurement
- the evidence that actually appeared, including evidence outside the transformation reporting machinery
- any material revision to the benefit, target, timing or causal argument, with the previous version still visible
- the judgement reached: realised, partial, not realised or explicitly superseded
- the person or institution prepared to own that judgement
None of this freezes transformation. It gives transformation memory, and memory is what makes learning possible.
Three claims that should be able to fail
- As AI takes on more production of transformation artefacts, artefact quality will become a weaker indicator of the organisational confrontation and commitment behind them.
- Where benefit-realisation milestones do not produce explicit judgements on previous transformation claims, narratives and targets will be revised more readily than adverse transformation outcomes are recorded.
- As transformation narratives become cheaper and faster to regenerate while benefits still require time to embed, the distance between the current story and the original claims will widen unless organisations deliberately preserve provenance, baselines and prior commitments.
AI could prove all three wrong. It could make transformation memory stronger, not weaker: reconstruct every previous claim, continuously compare it with evidence and make quiet revision almost impossible. Organisations could use that capability to make benefit realisation more rigorous and accountability more visible than it has ever been.
Good. Then the evidence should overturn the argument.
But the test cannot be whether the transformation documents improve. That is exactly the trap.
The question underneath the question
Return to the transformation meeting. Someone asks when we are done.
Perhaps that was never quite the right question. A major transformation may continue. Its ambition may evolve. New tranches may begin. A permanent transformation function may remain because the organisation expects transformation capability to be permanent.
The more useful question is: which promises are ready to face evidence now?
Which benefit has had enough time to embed? What did we originally say would happen? What actually happened? What are we carrying forward, what are we explicitly abandoning, and who is prepared to own that judgement?
A transformation does not need one mythical final moment of truth. It needs repeated moments when promise becomes evidence and the evidence is allowed to mark the story. AI did not create the weakness. It makes the story easier to regenerate than ever before, which is why the discipline of judgement now matters more, not less.
So when the next transformation narrative arrives — fluent, coherent, data-rich and perfectly aligned to the latest priorities — do not begin by asking how good it is.
Ask what happened to the promises in the previous one.
Then ask the question transformation should always be capable of answering.
How would you prove your transformation happened?
— The Fringe makes claims. The Lab tests them. Verdict to follow.
Sources
Infrastructure and Projects Authority and Cabinet Office — Guide for effective benefits management in major projects — 20 October 2017 — https://www.gov.uk/government/publications/guide-for-effective-benefits-management-in-major-projects
National Audit Office — Transformation guidance for audit committees — 24 May 2018 — https://www.nao.org.uk/insights/transformation-guidance-for-audit-committees/
Infrastructure and Projects Authority — Gate Review 5: Operations Review and Benefit Realisation — 15 July 2021 — https://www.gov.uk/government/publications/ogc-gateway-review-5-operations-review-guidance-and-templates
HM Treasury and Evaluation Task Force — Magenta Book: Central Government guidance on evaluation — updated 15 May 2026 — https://www.gov.uk/government/publications/the-magenta-book