Single Points of Failure Have Names and Notice Periods

The dashboard that is green for the wrong reason

There is a particular kind of quiet that follows a resignation letter. Not the social awkwardness. The operational kind. A senior person gives notice, and within a week the organization discovers that a process it thought it understood ran through that one person's head, and that nobody wrote any of it down because to the person doing it the work never felt like knowledge. It felt like Tuesday. The competency matrix said the role was covered. Two people were listed against it. The dashboard was green. It was green about a thing that was not true.

The usual explanation, that people fail to document, is both true and useless. The deeper problem is structural. Every instrument an organization uses to see itself, the org chart, the skills matrix, the CMDB, the succession plan, records positions and credentials. None of them records dependency. A role can be staffed twice over on paper and still have exactly one person who can do the part that matters, and there is no field on the form for that. The map shows two roads where there is one road and one painted line. You find out which is which when the traffic stops.

How much of the job lives in one head

The size of the thing should stop a reader cold. Panopto's Workplace Knowledge and Productivity survey put it at roughly 42 percent^[1]: about that share of the knowledge employees need to do their jobs is unique to them, held by no one else and recorded nowhere. Read that against how most firms reason about staffing. A team of five feels like five-fold redundancy. If two-fifths of each person's working knowledge is theirs alone, the team is not five deep on its hardest tasks. It is one deep, five times over, on five different things, and the redundancy you are counting on covers only the part that was always easy to replace. The same survey priced the friction at about 2.4 million dollars a year for a thousand-person firm, and IDC put the Fortune 500's knowledge-sharing failures at 31.5 billion dollars a year^[2]. Those are the visible drag numbers, not the one that matters. The number that matters has no dollar sign, because nobody can price a capability until the morning it is gone.

The demographics make this a wave rather than a series of incidents. The GAO has reported that 31.6 percent of federal employees on board were eligible to retire within five years^[3], and entire technical workforces sit on the same cliff: the median COBOL maintainer is past 55, and that code still clears most daily transaction volume in banking. The single points of failure are not randomly distributed across the age curve. They cluster at the senior, tenured, about-to-leave end, precisely the population an organization is least able to see clearly, because seniority is read as safety and tenure is read as coverage.

The matrix finds the people, never the knowledge

The nuclear industry, which cannot afford to learn this the expensive way, built the cleanest tool for it. The Tennessee Valley Authority and later the IAEA codified a knowledge-loss risk process^[4] around a Total Attrition Factor: for every incumbent, score a retirement factor from the projected departure date, multiply it by a position risk factor from how critical, unique, and hard-to-refill the knowledge is, and let the system roll the product up to the department level. It is the most disciplined version of what every succession plan gropes toward. And the most important sentence in the IAEA write-up is the admission that the matrix only finds the people. The actual inventory happens later, in structured interviews with the at-risk expert and their supervisor, because the score tells you a cell is red and nothing about what is in it.

Steve Trautman's Knowledge Silo Matrix makes the failure mode visible in a way a spreadsheet usually hides. Trautman colors each cell of a people-by-skills grid^[5]: purple means this person can do the work and teach it, green means works independently, yellow means still learning, white means nothing. The single point of failure is the silo with one purple and no greens, and his blunt observation from large engineering and IT groups is that managers are routinely wrong about which cells are which, because people who claim green fail the test questions when you probe whether they can do the task alone. That is the gap a competency dashboard launders into a reassuring color. The dashboard records that a course was completed. It cannot record that the person who completed it still hands the hard cases back.

Why the exit interview captures the wrong thing

Dorothy Leonard and Walter Swap named the trap better than anyone. Their work on deep smarts sorts an expert's knowledge into the explicit, which documents handle fine, the implicit, which the expert can say if you ask the right way, and the tacit, which the expert cannot articulate at all and which has to be observed. Leonard's rare-case rule follows directly^[6]: a general question on the way out produces general answers, the textbook material the organization already owns. The valuable knowledge surfaces only when you ask about a specific hard call, a specific near miss, the one time the expert did something a competent generalist would not have. The standard exit interview, a form and a final paycheck, is engineered to harvest the layer least worth keeping and to miss the one that is irreplaceable. It is not a weak version of the right instrument. It is the wrong instrument pointed at the right person on the last day they will ever answer.

This is also why the standard rebuttal, that you can write it down, falls apart on contact. David DeLong's Lost Knowledge^[7], the founding book of this field, splits the problem into four kinds of knowledge that each die differently: human, the skill and the heuristics; social, the relationships and trust networks; cultural, how decisions actually get made here; and structured, the documents and systems, which DeLong points out decay the moment the people who understood their context leave. That last one is the cruel joke. The documentation is not a hedge against the departure. It is one more thing that quietly stops working when the only person who knew which parts to ignore walks out. DeLong's first move is the one most organizations skip: start from the strategic capabilities at risk and trace them to people, rather than starting from headcount and assuming the bodies imply the capability.

The successor who is rated ready and is not

Here is the failure that costs the most, because it wears the mask of a solution. An organization sees the risk, names a successor, marks the box, and relaxes. The successor sits beside the expert, is rated ready, is reported up the chain as coverage. Then the expert leaves, and within weeks the successor is on the phone to the retiree's personal cell, or quietly routing the hard cases to someone else, or letting the rare-event work pile up against the day the expert might pick up. The org believed coverage existed. What existed was an apprentice who could handle the common case and had never once been left alone with the case that mattered. The plan recorded the overlap of two chairs. It did not record that the knowledge never crossed the gap between them.

The people who run this professionally treat the named-successor box with open suspicion. Gary Klein and colleagues built the Critical Decision Method^[8] precisely because expert performance turns on cues and anticipations the expert mentions only in passing, the I-knew-where-this-was-heading that no procedure captures and no apprentice absorbs by watching. The nuclear operators encode the remedy as a rule of thumb: INPO practice is to seat the successor two to three years ahead of the expected retirement^[9], not two weeks, because real judgment transfers on the calendar of apprenticeship, not the calendar of HR. Read the petroleum-engineering estimate of three to eight years to grow an autonomous professional against the typical two-week exit checklist, and the mismatch is not a rounding error. It is two different theories of what knowledge is, and only one is correct.

The same anatomy outside the org chart

None of this is an engineering quirk, and the place it gets priced in real money is the sale of a business. John Warrillow's Value Builder work calls owner dependence Hub and Spoke^[10], and his diagnostic questions are deliberately behavioral rather than self-reported, because owners systematically overestimate how little runs through them: who do customers ask for, what happens when the owner takes a real vacation, who can quote and hire and sign. The buyer is not fooled by the org chart either. Key-person discounts of 5 to 25 percent on valuation are routine in mid-market deals, and earnouts of one to three years exist to handcuff the one head the price depends on. The market has simply decided to price the thing internal dashboards refuse to measure. A discount is just a single point of failure with a number finally attached to it.

And the bus factor compounds where the work is least visible. Noam Wasserman's founder research^[11] is a reminder that even the person at the center is usually the last to admit how much depends on them, which is why the useful signal in a transition comes from what the team and board privately predict the founder will do, not from what the founder says. The pattern holds from the control room to the corner office. The information that would let an organization see its own single points of failure exists, in the heads of the people doing the work and the people next to them, and it is never what the formal systems are built to collect.

Coverage you can audit, not coverage you can hope for

Pull the threads together and the indictment is specific. Organizations are blind to their own single points of failure not because the knowledge is unknowable but because every instrument they own measures the wrong layer. The org chart counts positions, not dependency. The skills matrix records exposure, not the ability to work alone. The exit interview harvests the common case on the one day it could have reached the rare one. The succession plan logs the overlap of two chairs and never tests whether anything crossed the gap. Each instrument is locally reasonable. Their sum is a leadership team that learns the true shape of its risk from a resignation letter, on the worst possible day and too late to act.

The grim comedy is that the cure is not exotic. Every expert above, under the framework, is describing the same recovery move: get a neutral party to sit with a specific person and ask about a specific recent moment, the hard call, the near miss, the case they quietly routed elsewhere, in a setting where admitting the gap does not cost them their standing. Leonard's rare-case probing. Klein's incident walkthrough. The IAEA's structured elicitation behind the red cell. Trautman's test questions that separate real green from claimed green. None of it is a survey and none of it is a form, because surveys and forms are how the knowledge stayed hidden in the first place.

Which points, finally, at the kind of instrument this calls for. Not another skills matrix, which finds the people and stops at the cell. Not another exit interview, which collects the part the organization already has. Something closer to what the canon has described for two decades and few have run at scale: a neutral, confidential conversation anchored to a real recent moment rather than a checklist, read back as an inventory of what only one head holds and an honest map of where the named coverage is paper. That is the instrument we are building at Latent Variables. The competency dashboard was always going to be green. The open problem was only ever to find out what it was green about before the notice period ran out.