Your support queue is fast because the answers are short. Your onboarding is quick because it skips the configuration step. Your engineering ships weekly because nothing gets written down.
All three are real gains and all three are the same move. You did not remove the cost, you moved it to a ledger you do not read.
Why these three models
The decision is whether a shortcut is a genuine efficiency or a transfer. The features that fire are a cost that falls on someone outside the person deciding, a return trip long enough that the cause is no longer visible when the effect arrives, and a question about frequency that founders answer from memory rather than from a count.
Three lenses. Externalities produce an equilibrium answer about who absorbs the cost and why nothing corrects it. Delay and oscillation produce a complex answer about why the correction, when it comes, arrives late and too hard. Base rates produce a cycle answer about how often this actually returns, which is the number that decides whether the transfer was worth making. The first is the one founders have heard of. The other two are where the money is.
1. The cost did not vanish, it changed ledgers
An externality is a cost or benefit that falls on someone who is not party to the decision. The standard examples are industrial and they make the idea sound like somebody else’s problem. Inside a company it is the most ordinary thing there is.
Sales promises a delivery date. The cost lands on engineering. Engineering ships without documentation. The cost lands on support. Support answers in one line. The cost lands on the customer. The customer absorbs it silently for eleven months and then churns during a budget review, and the churn is recorded as a pricing problem.
The economics here is not moral. Nobody in that chain did anything wrong by their own ledger, which is exactly the point. When the person deciding does not carry the cost, the level of the activity settles higher than it should, and it stays there, because every local incentive is pointing the correct way.
That gives a clean first test, and it is a question about accounting rather than about character. For any process you are proud of being fast, name the party who absorbs the difference. If you cannot name one, either the gain is real or the absorber is too far away to see, and the second is more common than the first.
Name a person, not a department, because departments do not absorb anything. Saying the cost lands on support tells you nothing you can act on. Saying it lands on the two people who answer the evening queue tells you how much capacity you just spent and who to ask whether your estimate is right. The version of this test that fails is the one whose answer is a function name, because any founder can produce that in three seconds and nothing follows from it.
A field example, since the office ones understate the distance. Sales agents on motorbikes collect in cash and record each sale on a phone at the end of the day. That is fast, and the cost of it lands on one accountant reconciling four hundred entries against a bank deposit that does not match, three days later, alone. The sales number is correct in every meeting. The reconciliation is invisible in all of them, and it is why the accountant resigns in March.
Two standard corrections exist and they behave differently. You can charge the decision-maker for the cost, which is a price. Or you can put the two parties in the same room and let them settle it, which works when there are few enough parties and the cost of bargaining is low.1 Inside a company the second is usually cheaper and almost never used, because it requires the person absorbing the cost to be present when the decision is made. Most companies structure their meetings to guarantee they are not.
Where you cannot get the absorbing party into the room, get their number into it. One line in the decision, written by them, saying what this costs on their side. That is a weak version of the bargain and it still changes outcomes, because a stated quantity is harder to walk past than an absent person.
2. The bill arrives late, and by then you correct too hard
If exported costs returned the same week, no founder would export them. The problem is timing.
A system with a delay between action and consequence does not simply respond slowly. It overshoots. The decision-maker keeps pushing because nothing has come back yet, the accumulated effect finally arrives, the response is a large correction, and the correction itself has a delay, so the system swings the other way. This is the standard behaviour of a stock with a lagged feedback loop, and it does not require anyone to be irrational.2
The company version is recognisable. Documentation is skipped for two quarters. Nothing bad happens visibly. Then two engineers leave, three months of knowledge leaves with them, and the response is a documentation mandate that consumes a full sprint and produces material nobody reads. Six months later the mandate has quietly lapsed and the cycle restarts.
What makes this worse than a simple lag is that people reason about the delay badly even when they know it exists. The experimental record on this is unflattering and it holds for educated adults on very simple systems: people systematically misread how a quantity accumulates from its inflows and outflows, and the failure is not explained by unfamiliarity with graphs, by motivation, or by general cognitive ability.3
You will not intuit the lag correctly, so you have to instrument it.
The instrumentation is cheap. When you make a decision that exports a cost, write down the date and the earliest date you would expect the consequence, then diarise the second date. That single act converts an invisible delay into a scheduled check, and it is the whole intervention. Systems with delays are not fixed by acting faster; they are fixed by measuring the state closer to where it changes and by not reacting to noise between the two dates.
That diarised date has a failure mode which makes it worse than no check at all. You look on the day, see nothing unusual, and record an all-clear that was never earned, because the return trip was longer than you guessed or because you never decided what you were looking at. So write the measurement down with the date. Not whether skipping documentation hurt us, but a named number: time to first commit for the next engineer who joins, or tickets referencing a feature nobody wrote up. A check that could not have come back negative is not a check.
And know what happens when the correction is taken out of your hands. An investor, a board, or a departing employee’s handover sometimes forces the fix at the exact moment the bill lands, which is the worst moment to size it. If you hold the dates and the measurement, you arrive at that conversation with a schedule instead of a mandate, and the difference is the sprint you do not lose to a policy that lapses six months later.
3. How often does it actually come back
The first two models tell you the cost exists and returns late. Neither tells you whether it returns at all, and some exported costs genuinely never do.
This is a base rate question and it is the one founders answer worst, because the available evidence is vivid and unrepresentative. You remember the customer who churned over an undocumented feature, and you never counted the four hundred who did not notice. The correct question is not whether the failure is possible; it is what proportion of cases like this one produce the failure.
Under a weak specific signal, the sensible estimate sits close to the underlying frequency and moves away from it only in proportion to how diagnostic the new information is. Ignoring that frequency in favour of a vivid case is the standard error, and it is stable enough to be treated as a default rather than as an occasional slip.4
Run it as arithmetic. Of every ten times you skipped configuration during onboarding, how many produced a support ticket within ninety days. Not how bad the worst one was. How many. If the answer is one in ten and the ticket costs you twenty minutes, the shortcut is correct and you should keep it. If the answer is six in ten, the shortcut is a subsidy you are paying at full price and calling growth.
If the volume is too low to give you ten past instances, run the count forward rather than abandoning it. Tag the next ten as they happen, one column for the date and one for whether anything came back. The cost is a column in a sheet you already keep, and in most businesses ten instances arrive inside a quarter.
The frequency also expires, and that is the distinction the count hides. One return in ten at forty onboardings a month is four returns. The same rate at four hundred is forty, and twenty minutes each has become two working weeks. Rates that were correct to ignore become the largest item on a support roster without anybody changing a decision. So record the volume next to the rate, and recount when the volume doubles rather than when something goes wrong.
Most founders have never run this count for any exported cost, which means the entire portfolio of shortcuts is being carried on impressions. That is the actual finding. The individual answers matter less than the fact that none of them are known.
What the three say together
- Name the absorber. Which specific party carries the cost of this shortcut.
- Date the return. When is the earliest the consequence could reach you, and is that date in your calendar.
- Count the frequency. In the last ten instances, how many actually came back.
- Then price it. Frequency times cost of one return, against the value of the speed. If the shortcut still wins, it was never an externality; it was a real efficiency and you can defend it.
Where they disagree
The externality model says internalise the cost, which means slow down until the decision-maker carries it. The base rate model says that most exported costs never return, so internalising all of them is a large and certain price paid against a small and uncertain one.
Both cannot govern. The resolution is the count, and the count also decides how much process you need. A company that internalises every exported cost is a company that has replaced judgement with procedure, and its speed advantage disappears while the underlying risk was never large. A company that internalises none of them is running an unpriced liability. Neither is a philosophy; both are consequences of a frequency nobody measured.
There is a second conflict worth naming. Delay and oscillation says the correction should be gentle and early, because a big late correction causes the swing. Base rates says that if the frequency is high the correction should be structural and immediate. When the delay is long and the frequency is high, and this combination is common in anything involving customer trust, the two models give opposite advice about how hard to pull the lever, and the honest answer is to pull it early and gently and accept that you will not know for two quarters whether it worked.
What none of them contain
None of the three prices the reputational component of an exported cost, which does not behave like the operational one. An operational cost returns as a ticket. A reputational one returns as an absence: the customer who does not refer you, the engineer who does not apply. Absences are not recorded anywhere and no frequency count will find them, so the estimate you produce from this method is a floor rather than an estimate.
None of them handles the case where the absorber is a person with no exit, which is most junior employees. Externality logic assumes the absorbing party eventually resists or bargains. Someone who cannot leave does neither, and the cost accumulates silently for far longer than the model expects.
And none of them tells you which shortcuts are legitimate. Deliberately exporting a cost to a party who agreed to carry it, with the trade named, is normal commerce. Exporting it to someone who does not know they are carrying it is the thing this article is about, and the models are silent on which of the two you are doing.
The one action that survives the ignorance: pick the shortcut you are proudest of this quarter, name the party absorbing the difference in one sentence, and count how many of the last ten instances produced a visible consequence. If you cannot get to ten instances, you do not yet have the evidence to keep the shortcut or to kill it, and the honest next step is to start counting rather than to decide.
Who has to move
This belongs to whoever owns the process, not to the executive team, because the transfer happens at the point where the shortcut is taken and nowhere else. The natural instinct when an exported cost finally surfaces is a company-wide policy, which is the overshoot the second model predicts. The cheapest first test is one process, one named absorber, one count of ten. It takes an afternoon, and it usually produces a number the founder was not expecting in either direction.
Sources and notes
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. Externalities, the divergence between private and social cost, and the conditions under which bargaining between affected parties resolves them are treated in the chapters on public goods and externalities. Used here for the two correction mechanisms in section 1, and for the specific point that bargaining is effective when the number of parties is small and the cost of negotiating is low.
- John D. Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill. The behaviour of systems containing delays between action and effect, including overshoot and oscillation arising from lagged negative feedback, is developed in the chapters on system dynamics and on delays. Used in section 2 for the claim that the swing is a structural property of the loop rather than a failure of the people in it.
- Matthew A. Cronin, Cleotilde Gonzalez and John D. Sterman, Why don’t well-educated adults understand accumulation? A challenge to researchers, educators, and citizens, Organizational Behavior and Human Decision Processes 108(1), 2009, pages 116 to 130. Author copy: https://www.mit.edu/~jsterman/CroninGonzalezSterman061210.pdf. The abstract states that highly educated people are often unable to infer the behaviour of simple stock-flow systems, and that persistent poor performance is not attributable to an inability to interpret graphs, contextual knowledge, motivation, or cognitive capacity.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. Base rate neglect and the wider treatment of judgement heuristics sit in the part on bounded rationality. Cited for the direction of the effect only. The magnitude of base rate neglect is sensitive to how the problem is presented, and this article makes no claim about size, only that the frequency should be counted rather than recalled.
A note on a number this article does not give. There is no threshold frequency above which a shortcut becomes wrong. The threshold depends on what one return costs you and what the speed is worth, both of which are specific to your business. What transfers is the shape of the calculation, not a cutoff.
Joshua Agonya Pi’Rwot, Founder.