When an organisation raises its effort and its output does not move, the problem was never effort. Somewhere in the chain, one unaddressed constraint is absorbing everything that gets injected into the system. People, hours, pressure and tools added anywhere other than that constraint produce noise, cost and fatigue, and no additional throughput. The expensive part is the months spent working around that constraint without ever naming it.
Effort is a flow, and every flow has a slowest point
An organisation is not a container you fill. It is a sequence of handoffs. A request arrives. Someone qualifies it. A system processes it. Someone validates it. A decision is taken. Something leaves the building.
The rate of that sequence is not the average of the rates of its steps. It is the rate of the slowest step. This is not a management opinion, it is arithmetic, and it is the arithmetic most organisations refuse to look at because it is humiliating in its simplicity.
Eliyahu Goldratt built an entire body of work on that single observation, first in a novel published in 1984, then in the discipline that became known as the Theory of Constraints. The core claim is blunt: only by increasing flow through the constraint can total throughput increase. Everything else is local improvement, and local improvement at a non-constraint does not show up in the result.
He put it in a form managers remember: an hour lost at the bottleneck is an hour lost for the entire system, while an hour saved anywhere else is a mirage. I have watched boards spend a year chasing mirages with perfect discipline.
What makes the claim uncomfortable is that it contradicts something every executive has been rewarded for believing. Effort produces results. Want more results, add more effort. That equation is true, and it is true most of the time, which is exactly why nobody checks whether it is true today.
Capacity added away from the constraint turns into inventory
Here is the mechanism that nobody describes in a steering committee.
You add capacity upstream of a bottleneck. The people you hired are real, they are competent, and they work. What they produce arrives at the bottleneck faster than the bottleneck can consume it. It does not vanish. It queues.
In a factory, queued work is visible. It sits on a pallet and takes up floor space, and at some point someone trips over it. In an office, queued work is invisible. It lives in a ticket backlog, in an inbox, in a shared drive, in a column of a spreadsheet nobody opens on Fridays. It has no physical footprint, so it generates no alarm.
Worse, it generates activity. Queued work produces status meetings, escalations, follow-up emails, dashboards built to explain why the queue is long, and occasionally a new tool bought to manage the queue. All of that is work. All of it is paid for. None of it moves anything through the constraint.
This is the part that bankrupts the argument for “just push harder”. Pushing harder at a non-constraint does not produce zero. It produces negative. It converts salary into queue, it converts queue into coordination overhead, and it converts coordination overhead into the exhaustion of the people you most needed to keep.
And the queue protects the constraint from scrutiny. As long as there is always work waiting in front of it, the constraint looks productive. It is running at one hundred percent. In most organisations, a unit running at one hundred percent is praised, not examined.
The arithmetic was settled in 1967, and it had nothing to do with organisations
In 1967, Gene Amdahl presented a short paper at the AFIPS Spring Joint Computer Conference with an unglamorous title: Validity of the Single Processor Approach to Achieving Large Scale Computing Capabilities. It runs to three pages. It is one of the few pieces of computing literature a chief executive should read, and it is readable in ten minutes.
Amdahl was arguing against the enthusiasm of his time for throwing more processors at a problem. His point was that the speedup you can obtain by improving one part of a system is bounded by the share of total time that part accounts for. If a job spends ninety percent of its time in a portion you can parallelise and ten percent in a portion you cannot, then even infinite capacity on the first portion caps your total gain at a factor of ten. You can buy a thousand processors. The ten percent does not care.
Transpose that into an organisation and the implication is brutal. The share of your lead time that sits inside the constraint sets a hard ceiling on what any amount of effort elsewhere can achieve. If eighty percent of the elapsed time of a client request is spent waiting for one validation step, then doubling the speed of everything else buys you an improvement of a few percent, at full cost.
Executives do not usually resist this once it is stated. They resist the next sentence, which is that they do not know what their percentages are. Almost nobody does. Elapsed time per step is one of the least measured quantities in professional organisations, because measuring it requires watching something boring for several days rather than asking a team how long they take.
A case: twelve months of maximum effort for one centimetre of progress
What follows is one organisation, anonymised, with the identifying details removed and nothing added. The shape is common enough that readers tend to recognise their own company in it, which is the point.
A European manufacturer of specialist industrial equipment, a few hundred people, three production sites, a direct sales force and a long technical sales cycle. Profitable. Well regarded by its customers. Growing slower than its market.
The board’s reading of the situation was straightforward and, on the face of it, reasonable. The sales cycle was too long. Quotes took weeks to leave the building. Customers waited, and some of them stopped waiting. The conclusion was that the commercial function was under-resourced.
So they resourced it. Over twelve months they hired three additional pre-sales engineers, bought a configure-price-quote tool that the sales directors had been asking for across two budget cycles, restructured the commercial organisation into two regional units with clearer ownership, and introduced a weekly pipeline review that the chief executive attended personally. Each of these decisions was defensible. Several of them were overdue.
Twelve months later the average time to issue a quote had improved by a figure the finance director described, without irony, as within the margin of error. Two of the three new engineers had left. The CPQ tool was in use for the simplest third of the catalogue and bypassed for the rest. The weekly pipeline review had become the most disliked meeting in the company.
The money was gone, the people were tired, and the position had not moved.
What actually happened was visible in a single afternoon, and it took an afternoon because nobody had spent one on it. Every quote above a modest technical threshold required a feasibility sign-off from the engineering office. That sign-off required one specific senior engineer, because he was the only person who had the full picture of how the product family interacted with the custom options the sales force kept selling. There was no document that said this. The organisation chart showed a team of six.
His calendar was the company’s throughput.
Everything the board had funded sat upstream of him. More pre-sales engineers meant more technically ambitious quotes reaching his desk. The CPQ tool automated the parts of the quote that were never the delay, and it could not automate his judgement, so for anything non-standard the tool produced a document that then waited for him exactly as before. The regional restructure gave two organisations the right to escalate to the same person. The weekly review ensured he was asked, every Thursday, by the chief executive, to explain the delay he was personally the cause of and personally unable to fix.
Each intervention was rational on its own. Together, they increased the load on the constraint while adding zero capacity to it. The organisation spent a year building a bigger funnel into the same hole.
The engineer, for the record, was neither incompetent nor obstructive. He was the best technical mind in the building and he worked longer hours than anyone. That is the normal profile of a human constraint: the constraint is usually the person everyone trusts most, which is exactly why work keeps being routed through them and exactly why nobody is willing to say the word “bottleneck” out loud in a room where they are sitting.
What they eventually did is less interesting than what it cost to find out, but it is worth stating because it was unspectacular. They separated the two things his sign-off was conflating: a genuine engineering judgement, which only he could make, and a compliance check against a set of rules which, once written down, three other people could apply. Writing those rules down took him eleven days of protected time, which the company had refused to grant him for four years on the grounds that he was too busy. After that, the share of quotes needing him personally fell to the minority of cases that actually warranted a senior engineer’s judgement.
The measurable gain arrived within a quarter. The three hires, the tool, the restructure and the weekly meeting had produced nothing in twelve months. Eleven days of the right person’s time, aimed at the right place, produced the result.
That ratio is the whole subject of this article.
The constraint never complains, which is why nobody looks at it
A constraint does not announce itself. It has three properties that keep it invisible, and they compound.
It is always busy, so it reads as productive. Utilisation and throughput are different quantities, but organisations measure the first and manage as if they had measured the second. A unit at one hundred percent utilisation is a unit that cannot absorb variability, and the variability has to go somewhere. It goes into the elapsed time of everything passing through.
It is usually trusted, so routing work through it feels like quality control rather than risk concentration. Nobody designed a single point of failure. It accreted, one reasonable exception at a time, over several years, and each individual decision to route something through that person was the right call in isolation.
There is a measurement habit that keeps all three properties intact. Organisations measure how busy people are, because that number is easy to obtain and it feels like management. They do not measure how long a unit of work waits, because that number requires patience and because it tends to implicate the way the organisation is arranged rather than the way its people behave. A company that only measures busyness will always conclude that it needs more hands, whatever the actual shape of its problem, because busyness is the only variable it can see moving.
And it is almost never where the complaints come from. Complaints come from the teams waiting on the constraint, and they phrase them in their own vocabulary: sales say sales is under-resourced, delivery say delivery is under-resourced, and both are describing the same queue from different sides. A chief executive hearing both complaints concludes, reasonably, that the company is under-resourced in general. That conclusion is the most expensive sentence in this article.
Finding the handbrake is an act of observation, not of analysis
The instinct, when a chief executive accepts that there may be a constraint, is to commission an analysis. Workshops, interviews, a maturity assessment, a diagnostic deck. This produces a map of what people believe about the organisation, which is useful for other purposes and almost useless for this one.
There is a reason the instinct is so strong. A diagnostic deck is a deliverable. It can be scheduled, scoped, priced and presented, and at the end of it everyone has something to hold. Watching a single order move through a building for nine days produces no deliverable until the last hour, and for most of that time it looks like nothing is happening, which is a difficult thing to defend in a steering committee. The method that works is the one that is hardest to buy.
Constraints are found by watching one real thing travel from end to end.
Take a single unit of work that matters: an order, a client request, a quote, a case. Take a real one, already in flight, not a reconstruction. Follow it without helping it. The discipline is in the second half of that sentence. The moment anyone knows the item is being watched, it moves faster, and the measurement is destroyed.
Record two numbers at every stage: the time somebody was actually working on it, and the time it was sitting. In professional organisations the ratio between those two is routinely one to twenty or worse, and almost every executive guesses it wrong by an order of magnitude when asked in advance.
The constraint is wherever the sitting time concentrates. It is usually one place. Not three, not a cultural condition, not a general need for transformation: one step, one queue, one calendar.
This takes days, not months, and it requires no methodology that a competent observer cannot carry in their head. What it requires is someone with no stake in the answer. That is the genuinely hard part, and it is organisational rather than technical.
The cost of misdirected effort is real money, and it appears in no ledger
Here is the accounting that nobody runs.
The direct cost is easy and is the smaller half: salaries of people hired to relieve a pressure they could not relieve, licences for tools bought to solve a problem they were not adjacent to, consulting fees for programmes aimed at the wrong layer.
The second cost is the elapsed time. A year during which the market moved and you did not. That cost is never recorded anywhere because it is an opportunity, and opportunities do not have invoices.
The third cost is the one that does the real damage: credibility. Each cycle of maximum effort followed by no result teaches an organisation something. It teaches that effort is decorative. The next time a chief executive asks for a push, the push is quieter, and the people who push hardest are the first to leave, because they are the ones who can.
None of this is in the profit and loss account. The salaries are there, the licences are there, the consulting is there, and they all look like ordinary, well-governed spending. There is no line item called effort dissipated against an unaddressed constraint. The number is large and it is invisible, and it stays invisible for as long as nobody looks for the handbrake.
There is no sophistication required to avoid this. There is only a willingness to ask one unglamorous question before funding the next push, and to accept that the answer may be embarrassing.
What is holding us back? Ask it before the next push is funded.
Sources
- Theory of Constraints, overview and the principle that total throughput rises only by increasing flow through the constraint: en.wikipedia.org/wiki/Theory_of_constraints
- Eliyahu M. Goldratt with Jeff Cox, The Goal, 1984, the origin of the bottleneck argument in management literature: en.wikipedia.org/wiki/The_Goal_(novel)
- Amdahl’s law, the bound on overall speedup set by the share of time spent in the part you did not improve: en.wikipedia.org/wiki/Amdahl%27s_law
- Gene M. Amdahl, Validity of the Single Processor Approach to Achieving Large Scale Computing Capabilities, AFIPS Conference Proceedings vol. 30, 1967, pp. 483 to 485: www3.cs.stonybrook.edu/~rezaul/Spring-2019/CSE613/reading/Amdahl-1967.pdf
FAQ
Why does adding more people or tools to a struggling team often fail to increase output?
Because an organisation’s throughput is set by its slowest step, not the average of its steps — the core claim of Eliyahu Goldratt’s Theory of Constraints. Capacity added upstream of the bottleneck doesn’t vanish; it queues invisibly in backlogs, inboxes and spreadsheets, generating status meetings, escalations, and new tools to manage the queue. That activity is paid for but moves nothing through the constraint, converting salary into coordination overhead and eventually into the exhaustion of the people you most needed to keep.
How do you actually find an organisation's real bottleneck?
By watching one real unit of work (an order, a quote, a case) travel end to end without helping it, recording at every stage the time someone actually worked on it versus the time it sat waiting. The ratio is routinely one to twenty or worse, and the constraint is wherever the sitting time concentrates — usually one step, one queue, or one calendar. It takes days, not months, requires no heavy methodology, and does require an observer with no stake in the answer, which is the genuinely hard part.
