When the Data Says Stop and the Politics Say Go — and Why More Data Is Never the Answer
A number that cannot survive contact with the organisation's commitments was not a finished piece of analysis; it was half of one.
The Slide Nobody Wanted to Build
The programme is eleven months old, and the slide on the screen is the one nobody wanted to build. The benefits case that justified a forty-million-pound back-office consolidation has quietly come apart. The payback sold to the board at eighteen months is now, on any honest reading of the run-rate, closer to five years. The headcount reduction the whole case rested on assumed a degree of process standardisation that the operating divisions have spent nine months resisting. The analyst has done careful work: the assumptions are laid bare, the sensitivities shaded, the single version of the truth finally assembled from four incompatible spreadsheets. Around the table there are grave nods. Someone calls the analysis “very helpful.” And then the committee decides, without ever quite deciding, that the programme will go on.
Everyone who has worked inside transformation has been in that room. Afterwards the story we tell is always the same, and it always flatters the teller: the data said stop, and the politics said go. The numbers were sound; the organisation lacked the nerve to act on them. If only the sponsor had been braver, the analysis sharper, the truth spoken more plainly to power.
That story is comfortable, widely believed, and mostly wrong. And believing it is precisely why the scene repeats.
Data Rarely Says Stop
Begin with the phrase, because it hides the whole problem. Data almost never says stop. It says the case is worse than we thought. It says the payback has moved from eighteen months to somewhere between three and six years, depending on assumptions that are themselves contested. It offers a range, a confidence, a set of conditions — and a range is not an instruction. Into the space between “worse than we thought” and “therefore abandon,” the organisation pours everything the numbers cannot settle: judgement, hope, reputation, and the plain fact that the alternative to continuing is not neutral but its own kind of loss.
We flatter analysis when we imagine it delivering verdicts. Even at its most rigorous it delivers estimates under assumptions, and the assumptions are the argument. The moment a sponsor says “but standardisation will land once the new divisional leadership is in,” the debate has left the territory the data governs and entered the territory it never could. This is not politics corrupting a clean number. It is the ordinary condition of every consequential decision: the number was never clean, and pretending it was is the first mistake.
The analyst who believes the data speaks for itself has mistaken a well-built estimate for a verdict. The organisation has not ignored the truth. It has done what organisations always do with a range — filled it with everything the range left out.
The Question the Numbers Cannot See
There is a deeper reason the numbers lose, and it has nothing to do with courage. By the time the data says stop, the organisation has almost always made commitments the data cannot see.
The consolidation was announced to the board six months ago as the centrepiece of the year’s efficiency plan. Two divisional structures have already been merged in anticipation of it. A multi-year contract with the systems integrator has been signed. Three careers are now visibly staked on the outcome. None of this appears on the analyst’s slide, because none of it is an investment question — and yet all of it is what the room is actually weighing.
So two entirely different questions are being asked at once, and the meeting conflates them:
| The analyst’s question | The room’s question |
|---|---|
| Is this a good investment from here? | What does it cost to reverse a commitment we have already made? |
| Do the future benefits justify the future costs? | Who has to stand up and say the plan they announced was wrong? |
| What do the numbers say? | What does stopping do to our ability to ask for the next thing? |
The analyst is answering the left-hand column with real rigour. The decision lives almost entirely in the right-hand one. When we say politics beat the data, what we usually mean is that the right-hand column beat the left — which is not a scandal so much as a category we failed to plan for.
The Honest Version of the Objection
Now the objection, in its strongest form, because it deserves one. Sometimes politics-over-data is exactly the failure it appears to be. Sometimes the sponsor continues not because reversal is genuinely costly but because admitting error is personally unbearable, and the organisation pours good money after bad to protect a reputation. The literature has had a name for this for thirty years — the escalation of commitment, the sunk cost that ought to be irrelevant and never is — and anyone who has watched a doomed programme limp to its second anniversary knows the pathology is real.
I am not arguing that every “go” is wise. Many are cowardice dressed as continuity. The point is narrower, and more useful: the remedy that follows from the data-said-stop diagnosis — get better data, hire a braver analyst, speak truth more loudly — almost never cures the disease, because it treats a symptom. You cannot out-evidence a decision that was never about evidence. The programme continued not for want of a clearer slide but because, by month eleven, stopping had been made unaffordable — and it was made unaffordable long before the analyst walked into the room.
“You cannot out-evidence a decision that was never about evidence.”
What Experience Actually Taught
If the disease is that stopping becomes unaffordable before the data arrives, the cure is not louder truth-telling. It is design. Four disciplines, learned the slow way, have done more to close the gap between analysis and decision than any dashboard ever built:
- Get the data in before the commitment hardens. The window in which stop is cheap closes early — usually before the first public announcement. An analysis delivered at month eleven is too late by construction, however good it is. The benefits case has to carry its own conditions from the outset, agreed when no one’s reputation is yet staked on the answer.
- Decide the stopping conditions in advance, and put names to them. At the point of approval, while everyone is still relaxed and honest, write down the specific evidence that would tell you to halt: the standardisation milestone that must be met by a date, the take-up that must reach a threshold. Agreed early, a later stop is not a betrayal of the plan — it is the plan, executing as designed. Reached for the first time at month eleven, the same decision is a humiliation. The difference is entirely in when it was framed.
- Translate the number into the currency of the decision. “Payback has moved to five years” is an analyst’s sentence, and it dies on contact with the room. “Continuing spends the credibility we will need to fund next year’s core-systems replacement” is a decision-maker’s sentence, and it lands. The practitioner’s real skill is not producing the number but rendering it in the terms the decision is actually denominated in — reputation, optionality, the next investment.
- Separate the two questions out loud. Force the room to answer the forward-looking case and the cost-of-reversal case separately, rather than letting them blur into one reluctant nod. What is the case from here, ignoring every pound already spent? And, quite apart from that, what does reversal actually cost — and is that cost real, or merely someone’s discomfort wearing a suit? Named apart, sunk cost loses much of its grip.
Notice that none of these is an act of individual bravery in the meeting. They are all things done before the meeting, while the politics are still soft. The heroic image of the analyst speaking truth to power is not only rare; it is a sign that the real work was skipped months earlier.
The Uncomfortable Conclusion
The phrase the data said stop but the politics said go survives because it casts us — the analysts, the planners, the people who built the honest slide — as the wronged party. We told the truth; they would not listen. It is a story with a hero, and the hero is us.
The less flattering account is the more useful one. Data and politics are not opponents in these rooms, and treating them as opponents guarantees the data loses. A number that cannot survive contact with the organisation’s commitments was not a finished piece of analysis; it was half of one. The other half — getting the evidence in early, agreeing the exits while everyone is calm, translating the finding into the language of the decision — is not a betrayal of rigour. It is what rigour looks like when it intends to change something.
We are, most of us, fluent in method and far less fluent in the timing and translation that let method matter. The gap between what the analysis knew and what the organisation did is real. But it is not, as we prefer to believe, a gap in courage. It is a gap in craft — and craft, unlike courage, can be taught.