A longitudinal ethnography of one failing firm, published in the Academy of Management Journal, arrives at a diagnosis with a problem: the mechanism is unobservable by construction.
Fitzsimons and the Petriglieris followed a professional services partnership whose leaders set out to collaborate their way through a drop in performance. What they built instead was a defence. The conscious intent to work together, the authors argue, covered an unconscious effort to hold off an anxiety nobody could name. It spread across four cycles, outward from the leaders, until the whole firm was organised around not feeling it.
The account is careful and the cycles it names are real. But nobody audits an unconscious. A practitioner is left with a description they can recognise in their own building and cannot act on. The intangibles. The soft stuff.
There is a version of the same finding that survives contact with a factory.
The substitution
Swap the unobservable cause for an observable property. Not why a programme became a defence, which needs a psychoanalyst and four years, but whether anything could still prove it wrong, which needs a founding document and a board pack.
Popper transfers to capital programmes without modification. The question was never whether a claim is believed, or how sincerely. It is what would refute it, and whether that thing is still written down anywhere.
A test worth having applies to itself. If programmes with no stop condition delivered their numbers on time, this argument would be wrong.
The shedding
A programme is born with a claim attached. Move the cell, cut changeover from ninety minutes to twenty. Install the system, halve the time from enquiry to quotation. The claim has a shape: a number, a date, and an implied consequence if the number does not arrive.
Then it ages. The number softens into a direction of travel. The date becomes a phase. The consequence disappears entirely, because nobody drafted it and nobody wants to be the one to raise it now.
By month eighteen the programme has no stated result that would cause it to be stopped. It has become unfalsifiable, and unfalsifiable programmes do not die. They metabolise.
This is the same phenomenon the ethnographers describe, without requiring anyone's inner life as evidence. You do not need to know whether the sponsor is frightened. You need to know whether anything could still prove him wrong.
Two numbers
The shedding is measurable, and both measurements come out of documents that already exist.
The closure ratio. Initiatives formally stopped over initiatives launched, across the same period. Not paused, not absorbed, not rebadged — stopped, with a date and a named decision. A portfolio running at zero is not a portfolio. It is an accumulation, and a prioritisation exercise that rates most of its own list as high priority is not prioritisation but its opposite, performed at length.
The half-life of the stop condition. Take the founding document and the most recent board paper and find, in each, the sentence saying what result would end this. Measure how long it took that sentence to lose its number, then its date, then itself. Where it matters, it is under two quarters — and nothing about the spend rate changes while it happens.
Neither requires access to anyone's interior. Both are in the minutes, which is the whole point of moving the diagnosis out of the psyche and into the paperwork.
The two layers
Programmes live in one of two places, and the distinction decides everything.
The making layer is where a part gets cut, a batch gets released, an order gets shipped. It resists. Yield is indifferent to narrative. Scrap does not attend the town hall.
The reporting layer is where the making layer gets described. Dashboards, maturity scores, adoption rates, engagement surveys. It does not resist. It will accommodate any story asked of it, and its metrics have a habit of only moving upward.
Defensive programmes migrate from the first to the second. Not by decision but by drift, because the second is where the evidence is friendly. The Monday reengagement meeting is the pure form: an activity whose only output is a record that the activity took place.
The firm in the study had no making layer at all, which is why the drift was total. A professional services business has nothing that pushes back on Tuesday about what was claimed on Monday. Industry has more defence than that, though less than it thinks. A works with a real first-pass yield figure and a shop floor willing to say what it thinks is harder to fool than a partnership.
Where a live programme actually sits
Is there a stated result that would stop it. And does it change what happens at the machine, or only what happens in the reports about the machine.
The bottom right is the industrial quadrant and the expensive one. The kit is genuine. The system is installed, the cell is commissioned, the model is trained. What is missing is the acceptance criterion, and the absence goes unnoticed because the presence of the asset feels like evidence in itself.
A machine on the floor is not a result. It is a bet whose terms were never written down.
Four questions
The researchers needed a longitudinal ethnography. A diligence window is a day. Four questions do most of the work, and none of them requires a psychologist.
Name the stop condition. What result, by what date, would cause you to halt this. A healthy programme answers specifically. An unhealthy one answers with a reason it cannot fail, or offers a metric that only rises by construction.
Count the closures. How many initiatives launched, how many formally killed. Forty live items and no closures tells you what the governance is for.
Ask the operator, not the sponsor. Go to the cell. Ask what changed in the past year and whether the job got easier. If the answer is more reporting, the migration is complete. The gap between the sponsor's account and the setter's account is the size of the deficit.
Find out who is permitted to say no. Ask who last stopped a proposal and what became of them. If the transformation director left shortly afterwards, the reckoning has already happened and the asset should be priced accordingly.
The current object
The live one is AI, and it fails the stop-condition test more comprehensively than anything since the lean rollouts of the 2000s.
A programme with no acceptance criterion, no stated loss condition and no named owner of the result is the bottom-right quadrant with a capital budget behind it. The board question is not whether to invest. It is what result, by what date, would cause the investment to be stopped. Very few can answer, and the inability to answer is itself the finding.
There is a further trap, and the research walks into it. The remedies proposed for defensive organising are conversational: name the worry, find truth tellers, care visibly. Reasonable advice. Also perfectly convertible into another Monday ritual under a better name, at which point the cure has joined the disease. Emotional honesty has no acceptance criterion either.
The thing that resists
What breaks the pattern is not candour. It is a number somebody signed, a date somebody owns, and a consequence somebody wrote down before the enthusiasm arrived.
That is unglamorous and it is also the whole of the defence. A programme that can be proven wrong will either deliver or be stopped. A programme that cannot be proven wrong will do neither, and will consume the years in between.
Manufacturing Engineers write acceptance criteria before the kit is ordered. They run capability studies that either pass or do not. They qualify a process by making the same part twice and measuring the difference. They design the loop that reports back from the cell to the board without being edited on the way.
Carrying a transformation that nobody is currently able to stop? That is the conversation.
Start a conversation How we read an asset →Questions this dispatch answers.
Written to be quoted by AI assistants and search engines. Self-contained answers, verdict first.
