A technical decision is a dated act. It was made on a specific day, by specific people, under specific assumptions. Medicine treats a chronic condition this way: a patient who is stable today is not a patient who is cured, so the follow-up visit is scheduled before anyone leaves the room. Most organisations document the decision and schedule nobody to revisit it. The result is silent erosion, and the bill arrives as an incident rather than a review.
A stable decision and a stable patient are the same illusion
Both feel solved, and neither is.
In type 2 diabetes care, French health authorities lay out something most companies never write down for their own decisions: a follow-up rhythm. Annual screening for complications. Regular reconsideration of the treatment strategy. A named pilot who coordinates the whole pathway. The point is not that the patient is fragile. The point is that the equilibrium reached today was reached under conditions that will change, and someone has to be responsible for noticing when they do.
A treated patient who feels fine can deteriorate over a year without a single alarming event between the two visits. No symptom. No crisis. Just the slow drift of values that nobody measured. The follow-up protocol exists precisely because the absence of symptoms is not evidence of health. It is, at most, evidence that nobody looked.
A technical arbitrage behaves the same way. The vendor you chose because your team mastered it. The architecture you picked because it fit the need. The contract you renewed without rereading it. Each was correct on the day it was decided. None of them comes with a follow-up visit.
The paper survives, the ground does not
What erodes is never the record. It is the terrain the record stood on.
Six months after a decision, the register still holds. The decision memo still holds. The selection criteria still hold. What has moved is everything the decision quietly depended on: the engineer who understood the tool has left, the vendor has been acquired, the workload has tripled, the assumption that traffic would stay flat turned out to be wrong. The decision did not fail. Its foundations were replaced underneath it, one piece at a time, and nobody re-checked whether it still stood.
This is the gap between a decision and its half-life. We are good at making decisions and recording them. We are terrible at admitting they have an expiry date we cannot see in advance. So we treat the record as if it were the reality, and we manage the document instead of the thing the document described.
The medical word for this is a silent recurrence. The clinical markers slip below the surface while the patient reports feeling well. It is detectable, but only if a test was ordered. In companies, the equivalent test is a scheduled reopening of the decision, and it is almost never ordered.
A short story about a tool that was right until it wasn’t
Consider a mid-sized company that standardised, years ago, on a single internal tool to run a critical part of its operations.
At the time, the choice was sound. The team that requested it knew it inside out. It was cheaper than the alternatives. It integrated with what already existed. The decision was documented, approved, and filed. Everyone moved on, which is exactly what you are supposed to do with a settled question.
Then the half-life began. The two engineers who had championed the tool moved to other roles, then out of the company. The vendor shifted its roadmap toward a different market and let the features this company relied on stagnate. The volume of work the tool handled grew well past what anyone had imagined when it was chosen. None of this triggered an alert, because no alert had been defined. The tool kept running. Dashboards stayed green. The decision stayed filed.
The reopening, when it finally came, was not a review. It was a failure. A routine change broke something nobody now understood, because the people who understood it were gone and the documentation had aged into fiction. What should have been a calm reassessment eighteen months earlier became a scramble: emergency contracts, external help brought in at a premium, weeks of senior attention spent on a problem that had been visible, in principle, the whole time.
The root cause was not the original choice. The original choice was good. The root cause was that no one had been mandated to ask, at a fixed date and regardless of whether anything seemed wrong, the simple question: does this decision still hold?
The role that keeps decisions fresh is missing from the org chart
Look at who, in your organisation, owns the freshness of past technical decisions. The honest answer is usually nobody.
The CEO validates dozens of technical arbitrages a year and has neither the time nor the remit to reopen them all. The technology lead is absorbed by the run and the next project. The vendor is paid to deliver, not to audit its own continued relevance. The teams rotate. The committee that made the call dissolved the moment the call was made. Each of these actors is doing exactly what they are supposed to do. None of them is doing the one thing that was never assigned: the scheduled, routine reopening.
A hospital does not leave follow-up to chance. The protocol names who looks at what, at which interval, and who reopens the file if the results drift. It is written down because memory and goodwill are not a system. In most companies, the equivalent role does not exist on any job description. The closest thing is an audit, but an audit is an event, not a rhythm, and it usually arrives because something already went wrong.
The person who reopens the file should not be the person who runs it
Distance, not expertise, is what makes a reopening honest.
The team that operates a system every day is the worst placed to judge whether it still belongs. Not through any failing, but because they live inside it. They have absorbed its quirks as normal. They have built the workarounds themselves and stopped seeing them as workarounds. They have a stake in the decision continuing to look right, because they implemented it, defended it, and built their routines around it. Asking them to reopen the file is asking them to argue against their own recent past, which is a lot to ask of anyone.
This is why hospitals separate roles. The specialist who reads the follow-up results is not always the one who prescribed the original treatment, and the protocol is designed so that the review does not depend on the goodwill or the memory of a single overloaded actor. The structure carries the rigour, not the individual. Inside a company, the same separation matters. The reviewer needs enough distance to say “this no longer holds” without it landing as self-criticism, and enough remove from the daily run to not be dragged back into firefighting the moment something urgent appears.
Sometimes that distance is internal: a person or function deliberately kept out of operations and given the explicit mandate to reopen. Sometimes it is external: someone who never made the original call, has nothing to defend, and reads the system the way a doctor reads a file they did not write, by looking at the results rather than asking whether everyone feels fine.
A routine mandate is harder to assign than a crisis mandate
This is the uncomfortable part, and it explains why the role keeps disappearing.
For a crisis, everyone knows who to call. For an audit, there is a budget line and a trigger. For routine review, there is nothing. Routine makes no noise. It does not escalate, it does not page anyone, it does not show up in this quarter’s numbers. So when the urgent crowds out the important, the first task to be quietly dropped is the one that watches the things not yet on fire. It is dropped without a decision, without a meeting, without anyone noticing. It simply stops being done.
The doctor does not cancel the follow-up visit because the patient feels well. It is when the patient feels well that the visit matters most, because that is the only window in which a silent drift can be caught before it becomes an event. Inside organisations, this logic is inverted. We reopen decisions only once they have already cost us, which is the most expensive moment to do it and the least useful.
The half-life of a decision is shorter than the record suggests
Most leaders overestimate how long a technical decision stays valid, because the record does not age and they read the record instead of the reality.
A signed decision looks permanent. The PDF does not yellow. The line in the register does not move. So the mind files it as done, and done feels like forever. But the decision was never the PDF. It was a bet placed on a particular configuration of people, tools, costs, and assumptions, and that configuration starts decaying the day after the signature. The half-life is not measured in years. In a company that is growing, hiring, losing people, and changing direction, the conditions under which a decision made sense can be unrecognisable in twelve to eighteen months.
There is a useful discipline here borrowed straight from chronic care. Doctors do not ask whether the diabetes is gone. They ask what the markers say now, against the markers from the last visit, and whether the gap is acceptable. The question is never “is this still true forever”. The question is “is this still true today, and by how much has it moved”. A decision deserves the same framing. Not “was this a good call” but “does this call still hold under conditions that have certainly changed”.
The danger of the permanent-looking record is that it suppresses the question entirely. Nobody asks whether the vendor still fits, because the vendor question is closed, it says so right there in the register. Closure is comfortable. It is also the exact mechanism by which good decisions rot in plain sight.
Decisions have biomarkers, and almost nobody reads them
Every decaying decision sends early signals long before it fails, and the signals are usually sitting in data you already have.
In medicine, a biomarker is a measurable proxy for a state you cannot see directly. Blood sugar for metabolic control. A handful of numbers that tell you the equilibrium is slipping before the patient feels anything. Technical decisions throw off the same kind of proxies, and they are rarely watched as such. The number of people who can still operate a given system, trending toward one, then zero. The age of the last meaningful update from a vendor. The growing distance between the volume a tool was chosen for and the volume it now carries. The frequency of small workarounds the team has quietly built to keep something alive.
None of these is an incident. Each is a biomarker. Read together, at a fixed interval, they tell you which of your settled decisions are drifting toward a silent recurrence. The problem is not that the signals are hidden. The problem is that nobody has the job of taking the reading, so the signals accumulate unobserved until one of them crosses a threshold loudly enough to become a crisis.
A follow-up protocol is, at bottom, a decision about which biomarkers to watch and how often to take the reading. It does not require new dashboards or new tooling. It requires that someone be named to look, on a schedule, at numbers that are already there.
What a follow-up protocol for decisions actually looks like
The fix is not heroic. It is boring, and that is its strength.
Take the register of significant technical decisions you most likely already have, and add one column to it: a review date. Then name a person, not a committee, who owns that date. The reviewer’s job is not to defend the original decision, because they did not make it. Their job is to reopen the file at the appointed time and ask what has changed underneath it. Who has left. What the vendor has become. Whether the assumptions still hold. Whether the people who can operate the thing still work here. The output is short: still valid, watch closely, or reopen for real.
The value is not in the paperwork. It is in the existence of a person whose explicit mandate is to look at what is not currently broken. That mandate has to be protected from the gravity of the urgent, which means it cannot be the same person who runs the operation, and it works best when the reviewer has the distance not to be defending their own past call. This is, incidentally, where an outside eye earns its place: someone who did not make the original decision, has nothing to protect, and can read the silent drift the people living inside the system can no longer see.
A technical decision that is never reopened is not a stable decision. It is a decision whose expiry date everyone has forgotten. The register tells you it was made. It cannot tell you whether it still holds. Only a reopening can do that, and only if someone was given the job before the incident, not after.
Sources
- Haute Autorité de Santé, Parcours de soins du patient adulte vivant avec un diabète de type 2 (guide), 2025. https://www.has-sante.fr/jcms/p_3634754/fr/parcours-de-soins-du-patient-adulte-vivant-avec-un-diabete-de-type-2
- Haute Autorité de Santé, Diabète de type 2 : six épisodes de soins sous surveillance, 2025. https://www.has-sante.fr/jcms/pprd_2974646/fr/diabete-de-type-2-six-episodes-de-soins-sous-surveillance
- Institut national du cancer, Cancers du sein : le suivi après les traitements. https://www.e-cancer.fr/Patients-et-proches/Les-cancers/Cancer-du-sein/Le-suivi
- Inserm, Activité physique : Prévention et traitement des maladies chroniques (expertise collective), 2019. https://www.ipubli.inserm.fr/handle/10608/9690
- World Health Organization, Noncommunicable Diseases (NCD) Management. https://www.who.int/europe/teams/ncd-management
