The message had been in the team channel for years, in one form or another. "Just ask her, she'll know." A new joiner asks. A team lead asks. A client-facing manager, three minutes before a call, asks. And she knows, every time, usually within a minute, usually correctly. She had never been given a title for it. The org chart listed her as a senior associate on a shared-services team and said nothing else.
She went on a planned leave of absence on a Monday. By Wednesday the escalation queue had doubled in age. By Friday a client manager was reading the same ticket aloud to three people on a call, none of whom could say why the exception had been approved the way it had two quarters earlier. The review that followed asked how a single absence had done that much damage. The honest answer was in the channel history, in plain sight for years. Nobody had read it as a risk register.
The org chart was accurate. It was also beside the point.
Every operation has two structures. One is drawn: reporting lines, roles, a skills matrix with green cells. The other is lived: who people actually turn to when the written procedure runs out. The drawn structure is audited, because it is a document. The lived structure is not, because it is a behaviour, and behaviours do not come with a field in the HR system.
The skills matrix said eleven people on that team were qualified to handle the exception category. Eleven green cells. It was a true statement about training completion. It was a false statement about who could do the work. In practice, ten of the eleven routed the hard cases to one person, which meant the real number was closer to one than eleven, and the matrix had been reporting the opposite for as long as it existed.
This is why the loss felt sudden to everyone who had to explain it. The dependency did not appear on the day she left. It had been building quietly across thousands of small requests, each of which was a reasonable decision by the person asking. Nobody chose to concentrate the capability. It concentrated the way water finds a low point: by being the easiest path, repeatedly, with no one measuring the slope.
The only report that exists is the one written after the absence.
Here is how organisations typically learn that a capability was concentrated. A person leaves, or goes on leave, or gets promoted, or is poached. The queue ages, the quality dips, a client notices. A review is convened. Someone builds a slide titled "key-person risk" and a remediation plan appears: document the process, cross-train two people, schedule shadowing. It is a sensible plan. It also arrives at the single moment when it can no longer change the outcome, because the knowledge it is trying to capture is on holiday.
That is the lagging indicator in its purest form. The measurement of dependency is the damage from the dependency. Between the first day the concentration began and the day it failed, the operation had thousands of chances to see it, and the system offered none, because it was not built to look. The cross-training plan is real work, and it is also an admission: this is what we do once the thing we should have forecast has already happened.
A dependency you can only measure by losing it is not a risk. It is a bill.
Eleven people were trained. One person was relied upon.
The distinction that matters here is the one the matrix cannot hold. Training is a count of who has been exposed to a procedure. Reliance is a pattern of who the organisation actually leans on when exposure runs out. They are different quantities, and they diverge precisely in the cases that are hard, which are the cases that cost money.
Consider how the escalation looks from inside the work. A straightforward request is handled by whoever picks it up. An unusual one gets a flicker of hesitation, then a message to the person who is known to resolve them. Over months, the unusual requests become her requests. Her own handling time is excellent. The team's median looks fine. The aggregate dashboard shows a healthy operation right up to the week it does not, because the aggregate is an average of the easy cases and the one hard-case specialist, and averages are exactly where a concentration hides.
There is also a human cost the dashboard never sees. The person everyone relies on usually receives no credit for it, no adjustment in grade, and no relief from the volume of ordinary work that still lands on her desk. She absorbs the interruptions, then gets a performance review about her own throughput. Some of the people who leave a team are the ones who carried it, and the organisation learns what they carried only from the silence they leave.
What the same Monday looks like when it is reported forward.
Rewrite the review. Not "why did one absence do this," but a statement made weeks before the leave began: on this team, in the exception category, capability is concentrated in one person. If she is unavailable for three weeks, the queue for that category ages past the client's threshold by the second week. Two others are on a path to cover it by the quarter's end. Two are not.
That is a forecast, and a forecast can be wrong, which is the point of it. Three things make it possible without a new system or a new store of data.
First, read trajectory, not the snapshot. The matrix is a snapshot: who was qualified on the day the cell turned green. Routing is a trajectory: whose hands the hard cases are moving toward, and how fast, against the rest of the cohort. That information already exists in the ticketing and escalation data the operation produces every day. Nobody has to type anything in, which is also why nobody can flatter it. A person cannot give themselves a better routing pattern.
Second, spend analysis by exception. Most of a team's work is routine, and cheap signals will say so. The expensive look is reserved for where the signals disagree: a category where the matrix shows breadth and the routing shows a single point. You keep the indicator, you discard the raw material, and you only open the case file where the two views contradict each other. That is where the concentration is found, in week two of building rather than in week three of an absence.
Third, price the cover. A category with proven second and third cover is a different product from one resting on a single person, even if both are staffed to the same headcount on the rate card. Readiness that is verified can be offered as commitment, such as a response window the delivery owner is willing to sign. Readiness that is assumed remains a risk someone else eventually absorbs. When cover has a price, the conversation moves from the training team, who must justify the effort, to the revenue owner, who can use it.
Where this breaks.
The forecast has limits, and an honest version names them. Routing patterns are a proxy for reliance, and a proxy can mislead: a team that has simply formed a habit of asking one person is not the same as a team that cannot do without her, and the only way to tell them apart is a deliberate test of cover. Some operations restrict the escalation data a forecast would read, and you discover which only once you are inside. Cohorts that are small or newly formed give noisy slopes. And the person identified as the concentration point may not welcome being named as one, which is a conversation that needs care, not a dashboard. A method that cannot describe where it fails has not been used on anything real.
The method, not the link.
This is the work we do at SSUNDAR under the name Capability Forward. It begins with the March Bench Diagnosis: one account, one question, whether the bench you will depend on in March is the one you are building now, including who it quietly depends on. The forecast and the benchmark stay with us, and part of the fee rides on the result the forecast shows, so some of the risk of being wrong is ours. That is the proof. A claim the author will not be paid against is only a claim.
The channel history is still there, years of it, with the same four words repeated by different people. The record of the dependency was never missing. Nobody had decided it was data.