What does this article cover?
- What does the novel format actually achieve?
- Where does constraint logic genuinely transform?
- Where does the arithmetic stop working?
- What does that research not say?
- What happens across multiple plants?
- What should you do when the mix decision matters?
- Who should this article send away?
- Where does this break down?
- People also ask
- What I sell, and what I don’t
- About the Stagnation Assassin
My verdict on this book has been four out of five for a while and it has not moved. I have said elsewhere what costs it the fifth: the manufacturing framing that makes service leaders tune out, a personal subplot that interrupts the pacing, and a complete absence of guidance on the politics of implementing any of it. This article is about a fourth charge I have not made before, and it is the one that matters most on a factory floor. The rule that turns Goldratt’s philosophy into a daily decision does not always work, and the demonstration of that is formal rather than anecdotal.
What does the novel format actually achieve?
Retention, and the fact that people finish it. A method delivered through a plant manager solving a problem gets read by operators who do not read business books, and the Socratic device of Jonah asking rather than telling forces the reader to derive the answer.
The pedagogy is the underrated achievement and it is worth being precise about why it works. A framework presented as a framework invites evaluation; the reader assesses whether they agree. A framework presented as a problem being solved invites participation; the reader works ahead of the character and arrives at the conclusion themselves. Conclusions you reach yourself are held differently from conclusions you were told.
The Boy Scout hike is the clearest instance. Goldratt demonstrates that statistical fluctuations in a chain of dependent events do not average out, they accumulate, and he demonstrates it with children walking in a line rather than with a queueing model. Any operator who has watched a schedule slip recognises it immediately, and the recognition does the teaching.
The cost of the format is the one I have named before. A reader in professional services or software has to translate every example, and a large number of them do not bother.
Where does constraint logic genuinely transform?
Where output is measurable and effort is not reaching it. In one turnaround, years of competent Lean work had not closed a nine-figure deficit, and two years of constraint work took it to zero. The method was not harder. It was pointed somewhere else.
That case is the strongest evidence I have for the book and I have written about it elsewhere in more detail. The part worth repeating here is what it shows about the five focusing steps. Nobody in that business was working badly. Flow improved, waste fell, and the metrics owned by the people doing the improvement all moved in the right direction for years. What was missing was a mechanism for establishing that the improvement was happening somewhere that could not affect the outcome.
Identify, exploit, subordinate, elevate, repeat. The step operators skip is subordination, because it requires telling a productive department to slow down, and that instruction is politically impossible to give without the authority the book assumes you have.
The diagnostic signature remains the most useful thing in the book. Local metrics improving while the final number stays flat is the constraint telling you it is somewhere nobody is looking.
Nobody in that business was working badly. What was missing was a mechanism for establishing that the improvement was happening somewhere that could not affect the outcome.
Where does the arithmetic stop working?
At the product mix decision, which is the one most operators reach for. The rule of ranking products by throughput per constraint minute is a heuristic that looks like a calculation, and research has demonstrated it does not always find the optimum.
This is the charge I have not made before and it is the one that belongs in a review written from the floor, because the product mix decision is where constraint logic meets a real order book.
The rule is the operational core of throughput accounting. Take each product, calculate the throughput it generates per minute of constraint time it consumes, rank by that ratio, and fill the constraint from the top. It is elegant, it is teachable in ten minutes, and it is the thing most people take away from the book.
It fails when several resources bind at once. A line of work running from Plenert in 1993 through Fredendall and Lea, Hsu and Chung, and Balakrishnan and Cheng has established that the original heuristic does not reliably reach the optimum in multi-constraint environments, and each of those authors proposed revised versions precisely because of it. In a multi-product plant, several resources binding simultaneously is the normal condition rather than an edge case.
And it can fail with a single bottleneck. This is the finding that surprised me. Linhares, writing in the International Journal of Production Economics in 2009, demonstrated that the product-mix decision is NP-complete, and constructed cases where the optimal mix includes products with both the lowest margin and the lowest throughput per constraint minute, violating the conventional accounting rule and Goldratt’s rule at the same time. The cause is that real production comes in whole units. You cannot make three-sevenths of a machine, and the integer requirement is enough to break the ranking.
NP-complete is not jargon here, it is the practical point: the number of feasible combinations grows exponentially with the product count, which is why no simple ranking rule can be guaranteed correct across them.
What does that research not say?
That the theory is wrong. The same authors who demonstrate the heuristic’s failures describe Theory of Constraints as remarkably successful, and one review states plainly that these shortcomings are premises for development rather than a case against the philosophy itself.
I want this section in the article because leaving it out would be the dishonest version, and because the distinction is the whole of the fourth charge.
Separate three things. The philosophy, that systems have limits and improvement away from the limit produces little, which nothing in this literature challenges. The five focusing steps, which are a sound procedure and are not what the papers examine. And the product mix heuristic, which is one specific decision rule inside throughput accounting, and which is where the failures live.
Goldratt’s own later work acknowledged that a product mix problem under his framework can be expressed as a linear program, which is the formal statement of the same point. The heuristic exists because it is usable by a plant manager on a Tuesday, not because it is provably optimal, and reading it as arithmetic is the reader’s error more than the author’s.
It is also worth noting that some of this literature runs the other way. Luebbe and Finch compared TOC and linear programming and several authors have found the heuristic performing comparably to formal optimisation in realistic cases. The honest summary is that it usually works and provably does not always, which is a different claim from either side’s headline.
What happens across multiple plants?
The constraint stops being a place and becomes a routing decision. In a single plant you find the bottleneck and work there. Across a network, where the bottleneck sits depends on how you allocate volume, which means the answer and the question determine each other.
The book does not address this and it is the situation most mid-sized manufacturers with more than one site are actually in.
Consider what changes. With one plant, capacity is fixed and the constraint is discovered. With three plants sharing a product range, you choose where each product runs, and that choice creates the constraint pattern rather than revealing it. Move a high-volume line from one site to another and you have relieved one bottleneck and created a different one somewhere else, which the five focusing steps handle by telling you to go back to step one, having given you no help deciding the allocation that produced the situation.
Two practical consequences worth stating. The first is that constraint analysis in a network should follow the allocation decision rather than precede it, because running it first optimises a configuration you are about to change. The second is that the drum-buffer-rope scheduling logic assumes a single drum, and a network with routing flexibility does not have one; what it has is a different drum per allocation scenario.
None of this makes the philosophy wrong. It means that in a multi-site business the constraint question sits downstream of a capacity and footprint question that Goldratt’s plant manager never faced, because he had one plant and it was the one he was trying to save.
I should be plain that this section is reasoning rather than reporting. The multi-plant argument follows from the structure of the problem and I am not citing a study for it, which puts it on a different evidential footing from the mix-decision research above.
What should you do when the mix decision matters?
Solve it properly rather than rank it. An integer program will find the optimum where the heuristic may not, and the threshold for bothering is simple: whether the value at stake in the decision exceeds the cost of an afternoon with somebody who can formulate one.
Three practical positions, in ascending order of what is at stake.
Below the threshold, use the rule. For a weekly loading decision on a stable product set, ranking by throughput per constraint minute is fast, close enough, and a great deal better than ranking by gross margin, which is what it replaced. Optimising here is waste.
At the threshold, sanity-check the edges. When a mix decision is worth real money, take the products the ranking rule would drop and test a few of them back in by hand. The failures documented in the literature show up as products near the bottom of the ranking that belong in the optimal mix, so the bottom of your list is where to look.
Above the threshold, formulate it. Integer linear programming solves this properly and the expertise to set it up is purchasable. An industrial engineer or an operations research specialist will do it in a day, which is trivial against a mix decision worth seven figures a year.
The honest caveat on the third option: formulating an integer program takes expertise and can be slow to solve, which is exactly why Goldratt’s heuristic exists and why it deserves more respect than a purely technical critique gives it.
Who should this article send away?
Anyone treating the ranking rule as arithmetic. It is a heuristic, it is usually good, and the difference matters only when the decision is large enough that being close is not close enough. Below that threshold, optimising is waste and the rule is correct.
The failure I would guard against is a reader taking this article as permission to distrust the book. That would be an expensive misreading. The philosophy is intact, the five steps are intact, and the case I described above is what happens when an organisation applies them properly.
A second group: anyone in a service or knowledge business who reads the plant examples literally. The constraint concept transfers and the mechanics do not, and the translation work is yours because Goldratt did not do it.
Where to go instead, and some of it is better than the book for specific jobs. For the mix decision at scale, the operations research literature on integer programming solves properly what the heuristic approximates, and Linhares’s paper is the clearest statement of why that matters. For the theory beyond the novel, Goldratt’s own later work sets out the framework without the fiction and is more precise about its own boundaries. And for the political half the book omits entirely, nothing in the constraint literature helps, which is a gap I have written about rather than solved.
Where does this break down?
I reread this book every year, which is a disclosed bias rather than a neutral position. I am also publishing a technical critique of a book I have called formative, which gives me an interest in the critique being survivable rather than fatal.
Three further limits worth naming. Most of the papers cited here reach me through other papers’ citations rather than in the original, and I have described them as they appear in that literature rather than claiming a reading I have not done. The mix-decision critique is unillustrated by anything first-hand: I can tell you the research says the rule fails and I cannot show you a plant of mine where it did, which is a real gap in a review subtitled from the factory floor. And my own strongest case for the method is a single business, which is an anecdote regardless of how large the number attached to it was.
One more thing about that case. The turnaround it describes followed years of Lean work, and I have no way of establishing what those years contributed to making the constraint findable once somebody looked. It is entirely possible the constraint work succeeded partly because the system it inherited had already been simplified, and that reading would credit the approach the story is usually told against.
The verdict stands at four out of five, unchanged by any of this, because none of the technical failures touch what the book is actually for.
The heuristic exists because it is usable by a plant manager on a Tuesday, not because it is provably optimal. Reading it as arithmetic is the reader’s error more than the author’s.
People also ask
Is The Goal still worth reading? Yes, at four out of five. The five focusing steps and the throughput-first framing remain the most useful operational lenses available, and the novel format means operators actually finish it.
Does the throughput per constraint minute rule always work? No. Research has established it does not reliably reach the optimum when several resources bind, and Linhares demonstrated in 2009 that it can fail even with a single bottleneck once production is limited to whole units.
Does that mean Theory of Constraints is wrong? No, and the researchers making these demonstrations say so explicitly. The failures are in one decision rule inside throughput accounting, not in the philosophy or the five focusing steps.
When should you use an optimiser instead? When the value at stake in the mix decision exceeds the cost of having somebody formulate an integer program, which for most mid-sized manufacturers is a low bar for an annual decision and a high one for a weekly loading call.
Where does the book fall short for non-manufacturers? The constraint concept transfers and the plant mechanics do not. Goldratt left the translation to the reader, and a large share of service and knowledge-work leaders conclude the lesson does not apply to them.
What I sell, and what I don’t
I don’t sell consulting. My employment contract doesn’t allow it, so there is no engagement waiting at the end of this article and no proposal coming. Two things here are commercial: I speak for a fee, and I sell books. Everything else, including the audit, is free, and it stays free because it has to be. If you want a second read on whether your mix decision is above or below the threshold, ask at my contact page.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as Global VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books, The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
The talk version of this asks a room how they decide what to run on the constrained machine, and then asks what happens to the products at the bottom of that list. Formats and fee are on my speaking page.