The board asks you to hold headcount flat. You announce a freeze on Monday. Four people still start over the next eleven weeks, because they signed in April.
The freeze was real and the payroll grew anyway. You cannot set headcount, you can only change two flows, and both of them have already left your hands.
Why these three models
The decision is when to slow hiring and by how much, and what to do in the months between the decision and its effect. The features that fire are a quantity that accumulates rather than being chosen, a delay between committing and arriving, and a commitment that is far harder to undo in one direction than the other.
Three lenses. Stocks and flows produce a complex answer about why the level resists your instruction. Starting small produces an equilibrium answer about how to build a commitment you can actually halt. Options produce a random answer about the asymmetry that should set the timing. The first is the one everyone nods at and nobody applies to payroll.
1. You cannot set a stock
Headcount is a level. Levels change only through their flows, which here are hiring in and attrition out. Nothing you say about the level touches it directly.1
That sounds pedantic until you count the lags on both flows. Hiring in: weeks to open a role, weeks to shortlist, weeks to decide, then a notice period at the candidate’s current job. From decision to desk is commonly two to four months, and every one of those stages is a commitment you already made. Attrition out: you do not control it at all, you only observe it, and you observe it late.
So a freeze announced today stops nothing that is already in flight. It affects roles that have not been opened. The pipeline you built three months ago lands regardless, which is exactly the pattern in the opening: a real decision, correctly made, arriving into a payroll that keeps growing for a quarter.
Two consequences follow and both are actionable this week.
The first is that your freeze should be specified as a flow instruction, not a level instruction. Hold headcount flat is not an action anyone can take. Open no new roles after Friday, and stop at the offer stage for anything not yet signed, are actions with dates. Write the second kind.
The second is that you need to know your committed pipeline as a number. How many signed offers have not yet started, how many verbal offers are outstanding, how many candidates are at final stage. That is the headcount you have already bought and cannot return. Most founders can produce their current headcount instantly and have no idea of this second number, which is the one that decides what a freeze can achieve.
Then work out how much of that pipeline can still be moved, because a commitment is not uniformly fixed. A signed offer with a start date eight weeks out can often be moved to twelve simply by asking, and most people accept a later start if they are told the reason and told once. A verbal offer not yet in writing can be held at that stage rather than withdrawn. A candidate at final stage can be told the decision is deferred by a month. Those are three levers with three different costs, and they are available for about a fortnight after the freeze and not afterwards.
Assume some of them will refuse, and decide what you will do before the conversation rather than during it. A candidate who has already resigned cannot absorb a twelve-week delay, and asking her to wait reads as an offer being withdrawn slowly. For that person the honest choices are to honour the date or to pay her to release you, and the second has a price you should have worked out in advance.
2. Build the commitment in pieces you can stop
The lag problem has a structural answer that has nothing to do with forecasting better.
Where a relationship has to be built under uncertainty, the reliable approach is to start with a small commitment and expand it as evidence accumulates. Each step is small enough that stopping costs little, and the sequence reaches the same place as one large commitment while leaving an exit at every stage.2
Applied to hiring, this is the difference between three shapes of the same decision. A permanent full-time offer is one commitment made in full at the start. A contract-to-permanent arrangement is the same destination in two steps. A scoped piece of paid work, then a longer engagement, then an offer, is three steps with two exits.
The later shapes cost more per hour and they are not a way of avoiding commitment. They are a way of making the commitment in units you can stop, which is worth precisely as much as your uncertainty about the next two quarters. In a stable year the premium is waste. In a year where you might need to freeze, the premium is the only reason the freeze will work.
Pay for the ability to stop in the months when you can afford to, because you cannot buy it in the month you need it.
There is a real cost here and it should be named rather than smoothed over. Contract-first hiring is harder in markets where good candidates have permanent offers elsewhere, and it filters out exactly the people with the most options. That is a genuine trade and it argues for using the graduated shape on some roles rather than all of them, with the permanent offer reserved for the hires you would not stop under any plausible scenario.
The failure mode of the graduated shape is that it never converts. A contract renewed four times is a permanent hire with worse retention, no benefits and a person who has learned that the conversion date means nothing, and that is where most of these arrangements end up. So write the conversion test at the start, in one line, with a date on it: what has to be true by which month for this to become an offer. If you cannot state that test, you are not staging a commitment, you are deferring a decision you have already made.
And check the classification before you build a hiring practice on it. In most East African jurisdictions a contractor who works your hours, at your premises, under your direction is an employee in substance whatever the paper says, which puts the tax and the termination liability back on you. The staged shape survives only where the early stages are genuinely scoped work with a deliverable and a person free to do it their own way.
3. The two directions are not symmetric
The third lens supplies the timing rule, and it comes from noticing that hiring and freezing are not mirror images.
An accepted offer is close to irreversible. Reversing it costs a person their plans, costs you a reputation in a market where candidates talk, and in many jurisdictions costs money. A freeze is close to free to reverse: you reopen the role and lose a few weeks. Where one direction is expensive to undo and the other is cheap, the value of waiting sits entirely on the expensive side.3
Which produces an unbalanced decision rule, and the imbalance is the point. You should freeze on weaker evidence than you would hire on. Not because caution is a virtue, but because the two errors cost different amounts: freezing wrongly costs you a few weeks of delay, and hiring wrongly costs a year of salary and a conversation nobody wants to have.
Cheap to reverse is not free to reverse, so price the reversal before you lean on the asymmetry. Reopening a role does not resume where it stopped. The shortlist has taken other jobs, the agency wants its fee a second time, and the employee who referred a friend into a process that went quiet will not refer another. Budget six to eight weeks and one shortlist rebuilt. That is still an order of magnitude below the cost of unwinding an offer, which is why the rule holds, and it is the reason a freeze should be announced as a pause with a review date rather than left open ended.
Keep the pipeline freeze and the budget freeze separate while you are at it. Stopping new roles holds the flow and can be undone by a sentence. Cutting the budget line removes the ability to restart quickly, because the approval has to be won again from people who have since made other plans for the money. Freeze the first and leave the second alone for as long as the cash allows.
There is a second reason to write the trigger early, and it is structural rather than a matter of resolve. A balancing loop with a long delay overshoots whenever the correction is sized as though the delay were zero, so a freeze decided at the moment the gap becomes visible is already late by the length of your pipeline and will be sized against a number that has since moved.4 By the time the shortfall is obvious, the correct freeze is larger than the one that feels proportionate.
The trigger should be written before the pressure arrives. Something observable and dated: if cash at end of month is below a stated figure, or if the pipeline coverage ratio falls below a stated multiple, then no new roles open and no verbal offers are made until it recovers. A trigger set in advance survives the meeting where the strongest candidate you have ever seen becomes available. A judgement made in that meeting does not.
What the three say together
- Write the committed pipeline as a number: signed but not started, verbal, and at final stage. That is headcount you have already bought.
- Specify freezes as flow instructions with dates, never as a level to hold.
- Sort roles into stop-able and not. Use the graduated shape on the first group and accept the premium.
- Set the freeze trigger now, as an observable, while nothing is at stake. Freeze on weaker evidence than you hire on, because the errors are not the same size.
Where they disagree
Starting small and the option argument pull in opposite directions on your best candidates.
Starting small says build every commitment in increments, because the increments are what let you stop. The option argument says the value of flexibility is proportional to your uncertainty, and for a candidate you are certain about there is no uncertainty to hedge, so paying the flexibility premium is pure cost and the graduated offer may lose you the person.
The resolution is that they are answering different questions. Starting small is about your uncertainty regarding the person. The option argument is about your uncertainty regarding the business. You can be entirely certain about a candidate and entirely uncertain whether you can pay them in nine months, and in that case the flexibility you need is not a trial period. It is a smaller total commitment: a shorter guaranteed term, a lower base with a review, a part-time start. Confusing the two uncertainties is why founders run trial periods on people they were never worried about while carrying fixed costs they cannot support.
What none of them contain
None of the three prices what a freeze does to the people already inside. A hiring freeze is read by the current team as a signal about the company’s health, and it arrives without the context the founder has. The models treat headcount as a quantity and it is also a message, and the message travels faster than the flow does.
None of them handles the role that is load-bearing rather than incremental. Most hiring is additive and can be paused. Some hires are the difference between a function existing and not existing, and freezing those does not slow growth proportionally, it stops a capability. Sorting your open roles into additive and load-bearing is a five-minute job that none of these models will do for you.
And one property the ensemble will not produce: attrition responds to the freeze. All three treat the outflow as independent of your decision. In practice a freeze raises the workload on the people who stayed, which raises attrition, which widens the gap the freeze was meant to close. That interaction is the thing most likely to surprise you, and no single-flow analysis will show it.
The one action that survives the ignorance: today, write down how many people have accepted an offer and not yet started, plus how many verbal offers are outstanding. Add that to current headcount. That number, not the one on your dashboard, is the payroll a freeze announced this month would leave you with.
Who has to move
The founder writes the trigger, and it has to be written before the quarter in which it is needed, because in that quarter every function will have a reason why its own role is the exception. The instinct when cash tightens is to announce a freeze and consider it done, which changes nothing already in flight and buys a quarter of false comfort. The cheapest first test is the committed-pipeline count, which takes one message to whoever runs recruitment and typically returns a number two to four people higher than the founder expected.
Sources and notes
- John D. Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill, and Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing. Stocks change only through their flows, a stock cannot be set directly, and delays in the inflow mean a decision to change the rate does not change the level for as long as the delay lasts. Used in section 1. The related experimental finding, that people systematically misjudge how a stock accumulates from its flows and that this failure survives education, incentives and experience, is Sterman’s; it is why this is presented as a structural fix rather than as advice to plan more carefully.
- George J. Mailath and Larry Samuelson, Repeated Games and Reputations: Long-Run Relationships, Oxford University Press. The result that a relationship under uncertainty can be built by starting with small stakes and expanding them as the relationship proves out, with each stage small enough that defection is not worth the loss of the future, is developed in the treatment of reputation and of gradualism. Used in section 2. Applied here to a commitment you may need to halt rather than to a counterparty you may need to trust.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press, and the real-options literature it references. The claim used in section 3 is structural: the value of deferring a decision rises with the irreversibility of the commitment, so where one direction is expensive to reverse and the other is cheap, the evidence thresholds for the two directions should differ. No option value is computed, since the inputs are not available for a hiring decision.
- Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing. The treatment of delays in a balancing loop, and of why a correction sized as though the delay were zero produces overshoot, supports the section 3 recommendation to set the trigger in advance rather than to respond when the gap is visible. Also the source for the observation in the closing section that a system’s structure generates behaviour the people inside it did not intend.
A note on a number this article does not give. There is no correct lag from freeze to effect. It depends on your notice-period norms, your stage mix and how many roles are already open. What transfers is that the lag is not zero and that you can measure your own from the last four hires: decision date to start date, averaged.
Joshua Agonya Pi’Rwot, Founder.