Your head of operations resigns on a Tuesday. Within an hour you are writing a job description that describes them: same background, same seven years, same unlikely combination of skills.
You will not find that person. You are not trying to replace a person, you are trying to get back to a state, and those are different searches with different success rates.
Why these three models
The decision is what to do in the ninety days after a key person leaves, and specifically whether to search for a like-for-like replacement. The features that fire are a target state that can be reached by more than one route, an output that mixes the person’s contribution with everything around them, and a set of dependencies that all ran through one node.
Three lenses. Equifinality produces an equilibrium answer about the routes available to the state you want. Luck and skill produce a random answer about how much of their output was actually them. Exposure topology produces a complex answer about what else was routed through them that nobody wrote down. The third is the one that decides how urgent this is, and it is the one nobody runs in the first week.
1. You want the state, not the person
Start by separating two things that the job description immediately conflates: the condition your company was in while they were there, and the particular individual who produced it.
In open systems the same end state can be reached from different starting points and by different routes. The state is the target; the path is one of several.1 That is not a consolation, it is an instruction about where to search.
Write the state rather than the person. Not seven years in logistics and a network of clearing agents, but: customs clearance completed in under four days, a named backup for each of the three corridors, and a weekly exception report that someone actually reads. Those are conditions, and conditions can be reached by a senior hire, by two junior hires plus a documented process, by an agency, or by a software change that removes the need entirely.
Most companies discover after nine months of searching that the state was reachable by a route they never listed, because the search was framed as a person from the first hour. The framing happens in the hour after the resignation, which is exactly when nobody is thinking clearly, and it then governs the next three quarters.
The route list is cheap to build and most founders never build it. Take the clearance example. Route one is a senior hire who arrives in four months and costs a full salary. Route two is promoting the coordinator who already files the paperwork, paying for a clearing agent’s licence course, and retaining the departing person for six hours a month as a paid adviser. Route three moves two of the three corridors to a bonded warehouse operator who employs their own clearing staff, which converts a fixed salary into a per-consignment fee that falls when volumes fall. Route four cuts the third corridor entirely, if it carries eight per cent of volume and most of the complexity. Each route reaches a different version of the state at a different price and a different speed, and none of them can be compared until they are written next to each other on one page.
One caution the model itself carries. Equifinality does not say every route reaches the state. Some paths genuinely cannot get there, and a person who holds relationships built over a decade may be one of them. The instruction is to enumerate the routes and rule them out deliberately, not to assume substitutability.
2. How much of the output was them
The second lens attacks a number nobody computes: how much of what you are losing is actually the person.
Any individual’s observed results mix their contribution with the conditions they worked in, and where the chance component is large the mixture is hard to separate from a short record. Someone who ran operations through a period when your volumes were stable and your main supplier was reliable has a record that includes a market, not just a person.2
The practical version is a comparison you can make in an afternoon. Look at their results across the periods when conditions changed. Did performance hold when volumes doubled, when the currency moved, when a supplier failed? Consistency across changing conditions is evidence of the person. Excellent results in a stable period followed by no test is a record you cannot read.
A record with no bad quarter in it is not proof of skill, it is proof of no bad quarter.
The test has a failure mode worth naming, because it is easy to run it and reach a confident wrong answer. Conditions do not change on a schedule, and a founder scanning for a stress period will usually find one the person themselves chose. A head of operations who avoided a corridor during a fuel shortage produced a good quarter by not being tested, and the decision to avoid it may itself have been the skill you are trying to measure. So ask what they declined to do, not only what they did. The second failure mode is sample size. Two hard quarters out of eleven cannot separate a person from a market, and treating them as though they could produces a hire priced on noise. When the record is too short to read, say so and price the role from the route list in section one rather than from a judgement you cannot support.
This cuts both ways, which is why it is worth doing rather than comforting. Sometimes the exercise shows the person was carrying more than you thought, and you should widen the search and pay more. Sometimes it shows the function ran well because the environment was easy, in which case the urgent problem is not the vacancy, it is that you have never seen this function under stress and you are about to.
3. What else went through them
The first two lenses are about the role. The third is about everything else the person was quietly holding, and it is the one that generates the surprises in week three.
Look at the person as a node rather than as a job title. Which relationships terminated at them and nowhere else. Which approvals only they could give. Which piece of undocumented knowledge exists in their head. Which supplier deals with your company because of them personally.3
Each of those is a dependency that fails when the node is removed, and none of them appears in the job description. The pattern is familiar from any system with concentrated structure: it tolerates the removal of ordinary nodes easily and fails hard when a hub goes, and from inside the organisation the hubs are not marked.
The exercise is a list made in the first week, with the person still present, and it is the highest-return thing you can do in that week. Ask them directly: what will break in a month that nobody has thought about. People who are leaving on good terms answer this question honestly and in detail, and almost nobody is asked it, because the first week is spent on the search rather than on the handover.
Be prepared for the list to contain something worse than a dependency. Occasionally the answer is that a broken process has been working only because this person absorbed it personally, in which case their departure will look like a collapse they caused and will actually be a structure that was failing the whole time.4 Replacing that person rebuilds the absorption and hides the problem for another year.
Assume for a moment that the person does not cooperate. They are leaving for a competitor, or the conversation went badly, and the hour with a blank page is not available. The list still has to be made, and most of it can be reconstructed from records rather than from goodwill. Read their sent mail for the last six months and count who they wrote to outside the company. Pull the approvals log, or where there is no log, every payment above your authorisation threshold and see whose name is on it. Ask the finance officer which suppliers telephone asking for a person rather than for a department. This is slower and less complete than the conversation and it recovers most of the same dependencies.
Then do the transfer in front of the counterparty rather than by announcement afterwards. A distributor who has dealt with one person for four years will not accept an email naming a new contact, and will keep calling the old number until it stops working. Go in person, with both people, before the last day, and let the departing person introduce the successor or the interim holder. Where that is impossible, hold the relationship yourself for a quarter rather than assigning it to someone the counterparty has never met. A relationship that lapses during a vacancy is rebuilt at the price of winning a new customer, not at the price of a handover.
The list also reprices the urgency. If three critical relationships terminate at this person, the immediate problem is transferring relationships during the notice period, not filling the role. Those are different activities and only one of them has a deadline you cannot move.
What the three say together
- Write the state you want as a set of conditions, then list every route to it, including the ones that are not a hire.
- Check their record across periods when conditions changed. Consistency is evidence; an untested run is not.
- In the first week, list what terminated at them: relationships, approvals, undocumented knowledge. Ask them to help you write it.
- Sort the list into what must be transferred before they leave and what can wait for a successor. Only the first has a real deadline.
Where they disagree
The route argument and the node argument give opposite advice about speed.
Equifinality says slow down. Do not hire in the first month, enumerate the routes, and you will often find a better and cheaper path than the like-for-like replacement. The exposure argument says move now: while the node is still present you have a window to transfer relationships and extract knowledge, and that window closes on their last day regardless of what you have decided about the role.
Both are right about different work, and reading them as one decision is what produces the common failure. The transfer work is urgent and should start the day the resignation lands. The hiring decision is not urgent and is usually made too fast because it feels like the response to the event. Founders routinely invert this, opening a search in week one and starting the handover in week five, by which point the person has mentally left and the relationships have started to lapse.
What none of them contain
None of the three sees what the departure does to the people who stayed. A resignation by someone senior is read as information about the company, and the second resignation often follows within a quarter for reasons that have nothing to do with the role you are filling.
None of them tells you whether the person was the right person. All three take the state they were holding as the thing to restore, and a role that has not been examined since it was created is often held to a standard nobody has revisited either. The resignation is one of the few moments when asking whether this function should exist in its current form costs nothing extra, and it is the moment everyone is most anxious to skip past.
None of them sees the successor’s problem. Someone will take this role and be measured against a predecessor whose record you have just established was partly circumstance. If the exercise in section two showed that the environment was easy, the honest thing is to say so to the person who arrives, because otherwise you have built a comparison they cannot win and will not know they are losing.
And one property the ensemble will not produce: the state you are trying to return to may not be the state you want. The company has changed since the role was designed. All three lenses take the previous state as the target, and the resignation is one of the few moments when redesigning the function costs nothing extra.
The one action that survives the ignorance: in the first week, sit down with the person leaving and write the list of what terminates at them. Relationships, approvals, undocumented knowledge, informal arrangements. Do this before you write a job description, because the list frequently shows that the role you were about to advertise is not the problem you have.
Who has to move
The founder, in the first week, and the work is a conversation rather than a search. The instinct is to open the role immediately, which feels like decisive action and spends the only window in which the departing person is still available to explain what they were carrying. The cheapest first test is one hour with them and a blank page, and it typically produces three or four dependencies nobody in the leadership team knew existed.
Sources and notes
- Ludwig von Bertalanffy, General System Theory: Foundations, Development, Applications, George Braziller. Equifinality, the property of open systems whereby the same final state can be reached from different initial conditions and by different paths, is central to the treatment of open systems. Used in section 1. Note the scope condition, which the source is explicit about and which business writing usually drops: equifinality is a property of open systems and does not assert that every route reaches the state, only that the state does not determine the route.
- Michael J. Mauboussin, The Success Equation: Untangling Skill and Luck in Business, Sports, and Investing, Harvard Business Review Press. The luck-skill continuum, the difficulty of separating contribution from circumstance in short records, and the use of performance across varying conditions as the discriminating test underpin section 2. The claim used here is directional; no specific number of periods is asserted, because the required record length depends on how variable the function’s environment is.
- Albert-László Barabási, Network Science, Cambridge University Press. The robustness and fragility of networks with concentrated structure, in which removing ordinary nodes has little effect and removing a hub has a large one, is developed in the chapter on network robustness. Used in section 3 as the framing for a person who is a hub in the company’s dependency structure. No claim is made here that any organisation’s dependency graph follows a particular distribution.
- Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing. The observation that a person can hold a failing structure together, so that their removal reveals rather than causes the failure, follows from the treatment of system structure and of the difference between events and behaviour. Used in the blind-spot section.
A note on a number this article does not give. There is no reliable figure for how long a key-person replacement takes or costs. It depends on how much of the state was in the person, which is exactly what the exercises here are for. Get your own estimate from the route list, not from a benchmark for the job title.
Joshua Agonya Pi’Rwot, Founder.