This article was first published on LinkedIn by Dr Kelvyn Youngman. It is re-published here with permission.


Very few people understand subordination. No, it’s true! In contrast, almost everyone understands exploitation. So, how can this be? Well, exploitation is instinctful and subordination is not. Subordination is thoughtful. We most easily understand exploitation, instinctfully, as a matter of productivity; productivity of the whole that is, productivity of the system rather than of the so-called individual parts. But what if we were then to try and describe this thing that we call subordination? What if we were to describe subordination thoughtfully? Well, we can most easily understand subordination, thoughtfully, as a matter of utility; the utility of the whole that is, utility of the system rather than of the so-called individual parts. I think that developing this understanding would be helpful. So, let’s start with a simple model.

A Simple Model

Imagine a tube if you will. It’s hypothetical, or metaphorical, or all of those things. But let’s seat it in reality for a moment. Have you ever seen a sausage extruder? A hopper with a screw, or an auger feed, that forces sausage meat into a forming tube. Alright, you are vegetarian, pasta machines work the same way. Let’s just imagine for a moment that the forming tube is two foot long. Let’s also assume for a moment that it takes two minutes for the sausage meat to pass through the tube from the hopper to the sausage casing. That two minute period is our utility.

Now, imagine, if we double the auger speed. Now the time it takes for the sausage meat to traverse the two-foot tube is just one minute. That is our new utility. And of course you will recognise that we have also just doubled our productivity as well. Our utility and our productivity have both increased. Set that specific example aside for a moment.

What if we now double the length of the tube to four feet and keep the improved auger speed. It now, once again, takes two minutes to traverse the tube even at our doubled productivity, and this takes our utility back to where we started from. You could say our utility has now declined even though our new productivity remains constant.

Lastly, what now, if we put a bigger auger in the hopper and a larger diameter tube on the extruder? Its still four feet long, and let’s say it still takes two minutes to traverse, but more stuff, more sausage meat, is produced each minute. Our utility, now, is still the same as before, but now our productivity has increased once again.

The tube length is really a proxy for time, for duration. The utility that it imposes is the wait-time for something to come out of the process. The tube length and diameter is a proxy for capacity, and the speed of extrusion is a proxy for productivity. The productivity that it imposes is the amount of material that is extruded over a given period, the rate in fact, of material that comes out of the process.

Forget about bottlenecks, forget about hour-glass crimps in pipes, and all that stuff, and just hold the thought that rate is productivty, and duration is utility. Think of yourself as a customer at the end of the process. Think of yourself as the first customer in the process after it starts up and you have to wait until all the air in the pipe is pushed out and the new mix reaches the end of the forming tube.

OK, as I said, hypothetical and metaphorical, but I hope it will help. Let’s take a couple of steps back now, and return to the basics of exploitation and subordination.

The Five Steps of Focusing as an Erroneous Two-Step Process

Eli Goldratt formulated the five steps of focusing and gave it formal expression in his 1990 book What is this thing called Theory of Constraints and how is it implemented.The five-step focusing process as presented is:

  1. Identify the system’s constraints.
  2. Decide how to exploit the system’s constraints.
  3. Subordinate everything else to the above decision.
  4. Elevate the system’s constraints.
  5. If in the previous steps a constraint had been broken, go back to Step 1, but do not allow inertia to cause a system constraint.

You are most welcome to go back to this original source for early descriptions of each step, and to later verbalizations in The Goal and other places since. You may also be aware of my view that many people, regardless of what they might say they do, in their actual doing, treat this five-step process as a simple and erroneous two-step process. The essential elements of such a two-step approach are:

  1. Identify the system’s constraints.
  2. Exploit the system’s constraints.

or worse still:

  1. Identify the system’s constraints.
  2. Elevate the system’s constraints.

The salient point being that although exploitation is sometimes missing, subordination is always missing. And, frankly, subordination can’t be missing. If we are to have a system in any engineering or biological sense of the word (a box or a body) then subordination must exist. Seeing that subordination must exist, it then behoves us to understand it too. I think this is where, many times, we fall short. In fact, my argument is that subordination is at the very centre of the power of Theory of Constraints.

At TOCICO in Chicago 2019 I did what amounted to a “compare and contrast” of exploitation and subordination to show just how much more important subordination really is. That it is, in fact, a higher logical level. While reviewing that material recently a new distinction occurred to me as I have mentioned: productivity and utility. It also occurred to me that this new distinction, in conjunction with several others that we already know, might help us all to better understand the sometimes elusive nature of subordination.

The whole competitive advantage of Theory of Constraints, the whole value proposition in MBA-speak, lies in our ability to think in a way that our competitors do not. It’s that simple. The power isn’t in the first step: identify, and it isn’t in the second step: exploit, the power is in the third step: subordinate. Anything less than good subordination will eventually lead to failure. And often just these three steps alone, applied with skill and understanding, are more than enough to drive the constraint “into the market” and to keep it there. And that is where we want it – right?

I would be lying to you if I didn’t add that I was also concerned recently over some discussion on professional social media (by that I mean LinkedIn) about “what is subordination?” and it seemed to me from the answers that were forthcoming, that for many, subordination is “whatever you want it to be!” Yet, subordination is the step that we most often fail to mention, or when we do mention it, it then looks a lot like Exploitation II. I think that the differences between exploitation and subordination are both profound and (at first) quite subtle.

Again, its that “liminality” thing. You know: people talk of systems thinking and then proceed to pull the system apart and try to analyse each of the individual parts mechanistically. Probably not you, but most other people do. Of course, if the system is mechanical and discrete it is easy to do that, the trap is when we try to port it to systems that are organic or social and continuous. In the same vein, if exploitation is the only thing that we can think in (and who can’t!) then any subordination is sure to look like exploitation. In contrast if subordination is the thing that we have learnt to think in, then exploitation begins to look like subordination too. But, in fact, subordination is a different kind of thing. As I said, exploitation and subordination are at different logical levels, a point that we will come back to.

In continuing to address these issues of exploitation and subordination I also tried to further codify the subordination portion, the thoughtful portion, of each of the logistical solutions: drum-buffer-rope, critical chain project management, retail/distribution/supply chain, and organizational structure (you didn’t know that was a solution but it is, think cooperation, specialization, and innovation). Each is determined by a unique set of just a few subordination sub-routines or sub-codes. I’ve started to call these road codes. These are simple steps that address dynamic complexity, and make it, well, simple – inherently simple.

As I did this the distinction between productivity and utility became clear to me:

  • Exploitation of the constraint is about productivity of the system.

and that:

  • Subordination of the non-constraints is about utility of the system.

Productivity and utility: the the more I think about it, the more that I believe that these two words offer a better clarity on the distinctions between exploitation and subordination. I think that they pave a way to move the point of liminality forward, or away from, exploitation and towards, or to, subordination. I will also say here that, dialectically, exploitation becomes subsumed into subordination. That sounds round-about, so let’s do it this way: productivity becomes subsumed into utility. I hope this will become clearer as we proceed.

Let’s leave that as it is for the moment and go back to our five-step focusing process. Then we will pick up exploitation and subordination once again and begin to deal with them in earnest.

The Five Steps of Focusing as a Legitimate Five-Step Process

I want to stress several aspects of the original and proper five-step focusing process, and then to reduce it to a short-hand form. That short-hand form will become the structure around which we can hang the subsequent discussion of exploitation and subordination from.

So let’s start by the stressing aspects of the classical five-step focusing process – at least in the way that I know how I do it.

  1. Identify the system’s constraints
  2. Decide how to Exploit the system’s constraints
  3. Subordinate everything else to the above decision.
  4. Elevate the system’s constraints.
  5. If in the previous steps a constraint had been broken, Go back to Step 1, but do not allow inertia to cause a system constraint.

There is subtlety here too. Look at step two. It will help you to understand where we are heading. There is an alternative reading of this, let’s have a look.

  1. Identify the system’s constraints
  2. Decide how to exploit the system’s constraints
  3. Subordinate everything else to the above decision.
  4. Elevate the system’s constraints.
  5. If in the previous steps a constraint had been broken, Go back to Step 1, but do not allow inertia to cause a system constraint.

Goldratt’s first language was Hebrew, and I think he was especially careful when choosing English words for Hebrew concepts. So we too should be very careful to notice what he wrote. It is not by accident that step two is a decision step, and step three is a doing step. Most often we treat step two as a doing step – the do everything step, in the two-step process. So let’s keep that in mind as we progress through this discussion.

I now want to truncate the five-step focusing process a little. The second step reverts to its traditional form. This is how it looks:

  1. Identify
  2. Exploit
  3. Subordinate
  4. Elevate
  5. Iterate

So, I like to play with words. I can’t help that. Words are sometimes the only thing we have to convey our concepts. But now we do have a pared-down list; five simple steps, and we can use these for the remainder of the discussion. Let’s begin to tease things out a bit more for exploitation and subordination.

Instinctful and Thoughtful

We mentioned that everyone can think in terms of exploitation. But, how is that? Well, here’s the thing. It’s instinctful. By that I mean it has its heart in our instincts and in our emotions, capped with a thin veneer of thoughtful narrative over the top. It’s logical for sure, and it’s rational too; that’s the narrative part, but at its heart it’s instinctful.

When we see someone (someone else – because we would never do this to ourselves) sitting apparently idly by, it’s often a deep and emotional response that is aroused within our being. We want them to be busy (so that we are not – although we won’t admit to that). The alternative view, the fact that the person is indeed doing something useful by not doing very much at all at that moment (think firemen), is much, much, harder to do. In fact, in that situation, they are most likely subordinating to the greater whole. It’s harder to do because it is thoughtful. Sure, dig deep enough and we instinctively know that we ourselves can’t all be busy all of the time (and that in itself is hint towards the direction of the solution). But, by and large, not doing something, subordination in fact, is full of thought. It is thoughtful.

So, let’s start our distinctions of exploitation and subordination with the assertion that one is instinctful and the other is thoughtful.

  1. Identify
  2. Exploit – Instinctful
  3. Subordinate – Thoughtful
  4. Elevate
  5. Iterate

How do we know which is which? How do we know something is instinctful and something else is thoughtful. To put it bluntly that latter hurts and the former doesn’t. Let’s then further explore this powerful notion of the instinctful and the thoughtful a little more.

Thinking Fast and Thinking Slow

As I have said elsewhere the late Prof. Antoine van Gelder was at pains to get me to read Daniel Kahneman’s Thinking, Fast and Slowand I eventually did so and appreciated why he had been so insistent. Kahneman and his colleague the late Amos Tversky defined two modes of thinking: system one, and system two.

  • System one is parallel, automatic, effortless and associative – it is fast
  • System two is serial, controlled, effortful, and rule-governed – it is slow

System one, the one that is parallel, automatic, effortless and associative, the one that I meant doesn’t hurt, is instinctful. Instinctful thought is a consequence of our genes. System two, the one that is serial, controlled, effortful, and rule-governed, the one that I meant does hurt, is thoughtful. Being thoughtful is both a consequence of our genes and it is a consequence of our memes.

So let’s add these two to our structure in the appropriate places.

  1. Identify
  2. Exploit – Thinking Fast
  3. Subordinate – Thinking Slow
  4. Elevate
  5. Iterate

You have probably never done this, but imagine if you wrote an e-mail in anger (an emotion) and sent it right away (instinct). Wouldn’t it be better to have sat on it for a while to see if you come to the realization that quietly deleting it (thoughtful) was the better way forward. That’s the test for instinctfulness: its fast, its sometimes brutal, and in modern times it is often just plain wrong. The alternative, thoughtfulness, takes some time. I think that is part of the reason why Eli Goldratt developed his thinking process. It causes us to map out in a graphical way the serial nature of our thoughts and it slows us down enough to ensure that they are thoughtful thoughts at that. Yep, its painful and that’s why some people disparage it.

Constraint and Non-constraints

Let me ask you then, in everyday circumstances, where is it that we apply our instinctful reaction to, where is it that we apply our fast system-one thinking to? Well, the answer should be clear enough, it is everywhere! It’s instinctful to do so – right? And in many walks of life this is good enough. Its good enough because in many everyday places we at least have a semblance of independence (more imagined than real, but it will do nonetheless). However, within our modern invention, the modern process organization, or project organization, or distribution organization, dependencies rule the game. There is no longer a semblance of independence; things are strongly interdependent in a lock-step sort of way – both horizontally and vertically. However, in a nice twist, our instinctful response still has a place, one place, and that place is what we have come to know of as the constraint. Now instead of applying our instinctfulness everywhere we apply it to a few things, maybe just one thing, the constraint.

Let’s map this.

  1. Identify
  2. Exploit – Constraint
  3. Subordinate – Non-constraints
  4. Elevate
  5. Iterate

What then of subordination? Well, you can already see where I have put it. We have to apply thoughtfulness to everything else! We apply thoughtfulness to all of the non-constraints!

Can you see why subordination is so important, why thoughtfulness is so important? They apply to everything else. Now, it kind of feels like we have been trapped or tricked or tricked and trapped. Theory of Constraints is meant to be simple – right? And what could be simpler than applying our instinctfulness to just one or a few places. By a few places I mean that sometimes we do have similar things in similar places across multiple lines. And sometimes we have a number of similar things in one place in one line. Regardless they are still very few in number and we call this the constraint. So it is tantamount to treason that we should suddenly find everything else important too. However, once the gnashing of teeth and clenching of fists has subsided, once the anguish of the situation has abated, once we beginto apply some thoughtfulness to the situation, we find that paradoxically we still only have to apply a very few things to everything else (Phew!!!) It really is that simple. So what are they? Let’s begin to have a look.

(1) Don’t-Stop and Start-Stop – General

If I ask what exactly is it that we must instinctfully apply to the constraint? Then the answer at first blush is likely to be something akin to keep it busy! We do that without even thinking – right? That, after all, is pretty much what we try to do now everywhere all of the time. When there is only a somewhere, almost a nowhere, where we have to do this, then it suddenly becomes easy. If the constraint is a single machine or a type of machine or a group of similar machines this is relatively easy. We might stagger operator breaks so there is little down-time. We do everything possible to reduce set-up time because that is “non-productive,” We nip in and do preventative maintenance rather than wait for a break-down (I know the world has moved on, but I just needed to say that just-in-case). We ensure that there is a pile of work in front of the constraint so that it doesn’t run out of work (the fact that we most often have piles of work everywhere else too is beside the point at this moment). I know this is all motherhood and apple pie as the Americans would say, and it suddenly becomes wonderfully easy when there is only one or few places where you need to do it.

And by-the-way, you can also speed up a constraint by slowing it down. Strange but true. I’ve seen it. Machine manufacturers have recommended rates. People, being ever helpful, will “up” these rates. If there is an automated gluing stage involved you might not want to do this. You would think the cumulative down-time would be obvious – and you would be wrong. It won’t be obvious unless you go and stand in the process and watch.

Now you can say these things until you are blue in the face and come day one, the laptop for set-ups will be somewhere else! Well that’s a subordination problem, and as per usual I am getting ahead of myself.

So let’s ask, what if the constraint is a person or a group of people? Perhaps this is more common in services or in software. We can’t make people work across breaks and force-feed them pizza. But we can excuse them from those incessant planning meetings and we can take their Dilbert Comics away from them. Sad but true, and as I said, not too difficult.

The difficult part is subordination. Subordination is the difficult because instead of “don’t-stop” we have “start-stop.” We have to start-stop, start-stop, start-stop. Let’s add these to our structure.

  1. Identify
  2. Exploit – Don’t-Stop
  3. Subordinate – Start-Stop
  4. Elevate
  5. Iterate

I hoped that you might cool down as you read through that. Yep, I know, there isn’t a snow flakes chance in hell of start-stop working at your place of work. But it does, and here’s the thing, it is there right now, and for the most part it is hidden. It is thoughtful, but our instinctful selves win over every time and we cover it up.

Let’s back up a bit. If the non-constraints start and stop, then while they are stopped they must, in the English language sense, have “excess” capacity, or “spare” capacity. In biological systems (remember the body at the start) we would call this necessary redundancy. Bilateral symmetry aside, as far as I know we don’t begrudge ourselves two kidneys – quite obviously each working at only 50% each. So why is it any different for our organization (the box at the start). The answer is it isn’t. We bang on about the organization as an organic form and then run it like its clock-work. Go figure.

Let’s look at start-stop in another way. Do you have long duration for your project or production? Or perhaps more pertinently do you have a similar long duration as everyone else in your industry? Do you have months of work-in-process on hand? In case you don’t know the answer to this, just take the average sales value per month over say the last year and divide it into the final sales value of all of the work that is currently released to the process. You will be surprised. Include all work you have released “but haven’t really started on yet” that all counts! Do you have 40 foot containers out the back with spare stuff in them? Same thing. It all counts.

Part of this is a consequence of close-coupled dependency and inherent variability, but the more important point is you can’t do this (have work everwhere) without the capacity to make it. Instead of start-stop you have don’t-stop. You have what I will continue to call excess capacity at the moment and it is wasted (on stuff that you store – everywhere). What if you didn’t waste it, what if you stopped instead?

OK, at this point I have to say Eli Goldratt muddied the waters. His English expression of his Hebrew intent ran smack bang into the misunderstanding of professional instinctfulness. Goldratt used the term “sprint capacity” and he illustrated it with the cartoon character of Road Runner. And many people took it at its literal and face value: that the worker sprinted. This is wrong! In actuality the work sprints! But the work station or worker cannot. Let me explain. Sprint capacity is the “off” period or the “stopped” period of start-stop. The more often you have stopped (and frequency is incredibly important and underrated aspect) then the more often you can start again – and work at your normal rate. Now I’ve also slipped into that something about frequency but that kind of goes with the territory and we will come back to that. Imagine now that some work that is “falling behind,” we would call this a buffer penetration, in fact red zone buffer penetration. If the next, say, three stations are waiting in succession for that work, the work will sprint through, but the stations will work at their normal rate (how could they do otherwise?) and then stop. (Or if there is other work, do other work – but please, please, stop the BS of it’s management’s job to find “them” something productive to do – you’ve just missed the point entirely.)

So starting and stopping on non-constraints is thoughtful. It is not our instinctful reaction. Our instinctful reaction is to work regardless – either we produce too much, or we slow down in order to ensure that we are perceived as being “busy.” This destroys utility. There’s that word again.

In our general discussion of don’t stop/start-stop; one is instinctful the other is thoughtful, one addresses system productivity, the other addresses system utility. Let’s continue with more specific examples.

(2) Don’t-Stop and Start-Stop – Protection

There are several more thoughtful things that we must do in order to subordinate the parts (the non-constraints) to the whole (represented by the rate-limiting step of the constraint). Bear with me, let’s stay with production examples even though variations on these themes apply to all of the logistical solutions.

If we were to exploit everything everywhere then our instinctful reaction is that we need stock, a buffer, local protection, in front of every station, ever step in the process, so that none ever starves because they must all be kept busy. In fact, if you talk to a manufacturing jock, safety didn’t mean human health and safety (ever) it meant having a shit load of stuff in front of each and every machine. How much safety – well no one ever quite knew – but surely a heck of a lot.

Let’s add this to our structure.

  1. Identify
  2. Exploit – Don’t-Stop – protect each point with an individual stock/time buffer
  3. Subordinate – Start-Stop – protect the whole with a smaller aggregate buffer
  4. Elevate
  5. Iterate

Maybe I should have said “don’t protect each point” for subordination. But we know that protecting everything all of the time isn’t necessary in a linear serial dependency because only one place or one similar type or group of things will be rate limiting. Only one place or one similar type or group of things constrains the whole system. And because the non-constraints can stop – we’ve already decided that – then we don’t need individual stock in front of all of them at all times. In thoughtful subordination we aggregate the sum of the individual buffers and size that appropriately as a global buffer from the gate to the constraint and from the constraint to the end of the process. Actually this is more of an art than a science, and buffer management will soon tells us if we have it wrong. Actually I suspect the science would say that the aggregate buffer is somewhere near the square root of the sum of the individual buffers. If I took the proverbial 12 weeks of work-in-process that seems to automatically inhabit most batch manufactures, I’d be quite happy with √12 = 3.5 weeks instead. You can round it down to three weeks or round it up to four, but now I know why I’d drop work-in-process by a third to a quarter and not worry about it.

This is a very significant, all else being equal, we maintain our productivity but reduce our work-in-process considerably. When we reduce our work-in-process like that, we decrease our process lead time by the same amount. This is, wait for it, an increase in utility by any customer’s measure.

Alright there is some science there after all. In our discussion of don’t-stop/start-stop and protection, one is instinctful the other is thoughtful, one addresses system productivity, the other addresses system utility.

(3) Don’t-Stop and Start-Stop – Sequence and Schedule

Sequence, schedule, and indeed batch size, are all related to each other via the notion of set-up on the constraint and on the non-constraints. All three fall out of our don’t-stop/start-stop argument. Because batch size has other important implications I have decided to split that out and address it separately in the next section. In short, we don’t want to stop the constraint and therefore we sequence and schedule accordingly. Let’s make this explicit.

  1. Identify
  2. Exploit – Don’t-Stop  sequence and schedule
  3. Subordinate – Start-Stop – don’t sequence and don’t schedule
  4. Elevate
  5. Iterate

In my loose English I would say we exploit the constraint by writing a schedule and that the schedule in-turn subordinates to the constraint. Well, it does in fact, but let’s be quite clear that for the non-constraints we don’t sequence and we don’t schedule. Hold onto that thought, because I am about to walk into a shit fight. A shit fight between the instinctful and the thoughtful. Let’s back up and look at exploiting the constraint first, then subordinating the non-constraints second (spoiler – that’s were the fight is).

In a perfect world, a perfect world of make to order, we would schedule new orders as they arrived. But have you ever heard the rebuke “we will do that next month when we do the next run.” Did they want your work or not?? Well they do, but they also have a sequence and they almost certainly have a batch-sizing issue as well. If there is a constraint and we want to exploit the productivity of the constraint, then we want to reduce “unproductive” setups as much as possible. Many times, in many process, we might make things that are incrementally bigger in the next batch, or incrementally smaller. Such things usually only require small adjustments to the setup. Setup is faster, production is more productive. Ditto things that are lighter and lighter in colour or darker and darker. Then the issue might not be setup by rather clean up, or clean down as you prefer. The waiting issue is overcome by increases in frequency. There’s that frequency thing again. It sounds wrong – right? We want to decrease setup and increase the number of them at the same time.

Nonetheless, you will arrive at a sequence and then a schedule for the constraint. In a way this is ground-zero for your operation. Remember that shit happens so don’t be afraid to reschedule your reality, especially early-on in your new adventure. Which brings me to the next point. What to do with the non-constraints, which are not sequenced and are not scheduled. Let’s have a look.

Firstly when manufacturers say not scheduled or sequenced, they know that station B comes after station A – just like projects. Can I say it; that’s just common sense, stations don’t move around, although work can indeed go “back” through an earlier one. The issue is that on a non-constraint customer X can actually come after customer Y rather than before, it doesn’t matter. Or so you would think. My advice is that unless you have a red zone buffer penetration (and you will have to go and look in a text if you don’t understand that) you should do strict FIFO – meaning first in (FI) the line at that point in the process is next to be worked on and first out (FO). That is a thoughtful instruction. This is where the fight normally starts. It is thoughtful because it ensures against soldiering. The instinctful will reply that customer X must be pulled forward of customer Y as determined by buffer data/due date data sequence. But the moment you allow that to happen you ruin Road Runner. Remember Road Runner? The moment you do that, in walks soldiering (because there isn’t a “pressure” to do Y expeditiously so that you can get to X). In walks cherry picking because now anything goes. Next thing, carefully spaced batches (coming to that) will be suddenly get glued together again (just by accident!) The thoughtful rule is “do FIFO,” the instinctful response (at first) is every excuse under the sun not to. I mean every excuse.

Alright, that still looks like a sequence on the non-constraints, but it isn’t, different things go to different places at different speeds, none of which matter to anyone unless there is a red zone penetration, so just do the next job in the line.

Now, somewhere in there batching kept turning up. Let’s turn our attention to this and tease out the salient points.

(4) Don’t-Stop and Start-Stop – Batch Size

I am not sure that anyone other than a professional planner has any real feel for the safety in their process. We’ve discussed a seat-of-the-pants aggregation rule that approximates the square root of the old unaggregated safety. That one works for me. But confounded within that are batch sizing rules that it seems need to be addressed concurrently. And because they always are addressed concurrently it is hard to know which is cause and which is effect, or whether they are both intimately interwoven and interrelated.

This is what we need to do.

  1. Identify
  2. Exploit – Don’t-Stop  fewer setups, larger batches
  3. Subordinate – Start-Stop – more setups, smaller batches
  4. Elevate
  5. Iterate

Just as there are three types of carburettor (no there aren’t really, its just a joke, but never say it at a party with mechanical engineers or other tech-heads because (1) they have no sense of humour and (2) they will then spend the rest of the evening discussing the merits of the statement), there really are three environments where setup is important.

  1. Mass production: this is the bailiwick of Ohno and Shingo; they had metal press stamping changeovers down to 15 minutes in Japan in the early 1960’s and forging changeovers down to 15 minutes in Toyota Brazil a decade later in the early 1970’s. Now that’s a story that is not told too often – Brazil not Japan. We like to pick and choose our narratives. If it is not clear, as you reduce the non-productive time in set-up or change-over, or you can increase the don’t-stop time on potential constraints and improve the productivity, and all sorts of other good things as well. And while I am at it, Ohno is frequently represented as reducing waste. Do you know what he was really doing? Creating capacity. Both in the sense of productivity and utility.
  2. Make-to-order: Goldratt’s great breakthrough (sorry, like all these guys – just one of his great breakthroughs) was the recognition that a process batch through a constraint can be as large as necessary in order to avoid set-ups (regardless of any concurrent set-up reduction activity). In the remaining non-constraining part of the system, subdivision of the single large process batch into any number of smaller transfer batches allows things to flow (assuming you have drained your excess work-in-process).
  3. Make-to-stock: I don’t know whether those of us who do this just know it, and therefore don’t say it, or whether most people simply don’t know it, however, halving or quartering the 10-20% of the batches that account for 80-90% of the volume will produce a small increase in set-ups on the constraint (the batch frequency is doubled or quadrupled) and everything else as well. But 80-90% of the batches are unaffected. The process lead-time is halved or quartered.

I always find this intriguing. In the first two cases we have at the very least reduced or maintained “don’t-stop” on the constraint and have not impacted upon the non-constraints at all. In the third case we do impact upon the constraint, we do additional set-ups. We also do additional set-ups on the non-constraints as well. So how come things are better? Well simply stated we have a much better flow. The flow comes from frequency. And you know you will find the additional capacity on the constraints the moment you focus on them. I’ll give you an example: the constraint sits idle because there is only one laptop for setups on that line and it is in use setting-up a non-constraint.

I don’t need to tell you that the non-constraint will wait, indeed can wait. The constraint can’t. This is also where visible constraint schedules (a schedule board) becomes imperative at first. At first because we are changing our nature and our habits. Latter it will become second nature.

In our discussion of don’t-stop/start-stop and batching one is instinctful the other is thoughtful, one addresses system productivity, the other addresses system utility.

(5) Don’t-Stop and Start-Stop – Staggering

This is perhaps the simplest action to do, it also the last major one. We have to gate the admission of new work in concert with the rate at which it is processed on the constraint. We stagger the work admission, and rather than instinctfully admit it as soon as possible, we thoughtfully hold it back as long or as late as possible. Hey, another subordination step!

  1. Identify
  2. Exploit – Don’t-Stop – in ASAP
  3. Subordinate – Start-Stop – in ALAP
  4. Elevate
  5. Iterate

In the original boy scout analogy the rope is physical, it has tension in it between the drum and the gate and everything else can move, everything else has slack in effect, within those limitations, we call this movement the buffer. In that analogy we often consider the rope length to be the buffer size. I’m not sure that there is anything wrong with this. But I want you to also entertain the idea that the buffer might just be that – the buffer. A notion derived from the interaction between the aggregation of overall safety and reduction in maximum batch size/increase in frequency. To help you in this understanding consider if the drum was way down the line somewhere (imagine you are now in a factory – one of those U-shaped Japanese ones that require a tricycle to get around), the drum can signal to the gate with a flag, or a whistle, or a light (some mirrors might be involved), or maybe now a text on a smartphone. That is what the rope does, it synchronises the gate with the drum. You know you could actually run a huge plant like that. But hopefully that helps to understand why we don’t need a rope “length” in the picture. The rope is a signal first and foremost. Of course in any MRP or DDMRP environment, these conceptual issues are incorporated into the schedule.

In our discussion of don’t-stop/start-stop and staggering; one is instinctful the other is thoughtful, one addresses system productivity, the other addresses system utility.

Each of the subordination steps in the preceding sections: 3, or 4, or 5 of them depending upon how you want to slice and dice matters are thoughtful, they are counter-instinctual, they apply to the non-constraints (but benefit the whole through the rate-limiting step of the constraint) and they are subordination of the parts to the whole. There are, however, still one or two more general things. Let’s have a look

(6) Don’t-Stop and Start-Stop – What Didn’t We Mention?

I put this article aside for several weeks in mid-March, in fact when I got to this section. I did that in part because of preparation for, and distraction of, an imminent stay-at-home for coronavirus, and in part because I couldn’t rightfully express throughput accounting in the way that I thought that I wanted to. If at the moment you want to jump right past this section you are most welcome. But then, also, come back to it sometime soon.

I’ve been using a pattern language as I have progressed down the page, it’s one of my habits. That’s one way to check that your fundamental thesis is correct, because if it is not at some point something will come unstuck. However when I didn’t feel confident with my expression of accounting, I was fairly sure it was something about my thinking rather than something was about to become unstuck with accounting. And like all good things, if you wait a while it will often sort itself out. It has, so let’s continue.

There are really two aspects to this section, both are measurements. Let’s have a look.

  • The physical yield expressed in units per period (or periods per unit for projects).
  • The financial yield expressed in currency per unit.

What we (in Theory of Constraints) tend to do, is take the sum of the product of each of the the different physical and financial yields and call that Throughput. It might be so automatic for us that we forget that for many others it is not. Throughput then is always a rate. Even if it is something as absolute as the total throughput for the month, the “for the month” makes it a rate expression. Clearly any other accounting system will be also be similar, certainly at the output end. At the input end things might not be so similar. But now we are getting to the gist of the matter. But before we get there, I want to chop the problem back into two halves. First physical measurements then financial, at the end of which we can pull these two threads together again.

How do we express physical measurements in terms of our five step focusing process? Lets have a look.

  1. Identify
  2. Exploit – Don’t-Stop – physical output measurements
  3. Subordinate – Start-Stop – no physical measurements
  4. Elevate
  5. Iterate

Our instinctful response to measurements, to put it bluntly, it is to measure everything everywhere all of the time (and then report something as pointless as the average once a month and a month late at that!) If it moves we will measure it. Where do we find that instinctful expression in our constrained system? At the constraint or course, in the second step – exploit. We measure what goes out compared to some idealised target (the schedule in fact). And by extension that becomes a proxy for the system as a whole. Remember the constraint is rate limiting for the whole system, there’s just a lag between leaving the constraint and leaving the system entirely. Rather than monthly reporting, frequency is of the essence, once a day, the very next day, is not too much to ask, and not at all difficult to achieve.

What then of the non-constraints? What is the paradoxical action that we undertake there? Well, of course, it is not to measure them! “Whoa,” I can hear you say, dangerous – right? We might lose control. Not at all. [In fact you are not in control now so your loss is an illusion in any case.] What do we measure in front of the constraint and in front of shipping? We measure buffer penetration, lateness in fact. We measure things that we expect to be ready and waiting in two thirds of the time since their release at the gate. If they are not there, we should go looking. If root cause analysis, or Pareto analysis, suggests a similar upstream problem or a similar upstream pattern is frequently occurring, then that is the measurement. It occurs at the constraint.

Now, ONE MORE TIME, THE BUFFER IS NOT THE PILE OF WORK SITTING IN FRONT OF THE CONSTRAINT – I don’t care how many consultants and experts say otherwise; that is wrong. Its that liminality thing again. More than that, it is an illustration of exactly what I am trying to say here, it is an instinctful interpretation of Eli Goldratt’s thoughtful response. “Yes but our clients understand this better!” Well I rest my case. And while I am at it, the constant “cherry picking” of what is “in” and what is “out,” especially the necessary condition of secure and satisfied employees is exactly the same issue. It is one thing to further construct and build upon what Eli Goldratt produced, it is another thing altogether to deconstruct and disregard the pre-existing before we even start.

Let’s now bring the financial aspect into the mix. We have:

  1. Identify
  2. Exploit – Don’t-Stop – financial output measurements
  3. Subordinate – Start-Stop – no financial measurements
  4. Elevate
  5. Iterate

Our instinctful response to financial aspects, to put it bluntly, is to measure the cost of everything everywhere all of the time. If it moves, say a light switch, we will cost the light produced it in watts per square meter and allocate it as appropriate (or inappropriately actually). Where do we find that instinctful expression in our constrained system? At the constraint or course, in the second step – exploit.

NOW HERE COMES AN IMPORTANT POINT. We are concerned with decision analysis, we are concerned with management accounting. However, we report externally in terms of financial accounting. These are two different things. We do not have to use financial accounting for internal management decisions but we do have to use financial accounting for external reporting. The Japanese have a proverb “it is impossible to win over a crying child or government officials” and on that basis our external finacial reporting to government officials (or the stock exchange) remains an accrual-based one developed around 400 years ago in Rennnaisance Italy and which fundamentally assumes independent events. But there is no law that says you have to make internal management decisions using it.

Teleport foward to America at the start of their industrial revolution. H. Thomas Johnson and Robert Kaplan their 1987 classic Relevance Lost: the rise and fall of management accounting note some pretty clever management accounting systems developed by engineers (sacré bleu! engineers doing accounting, perish the thought!) at about the time of Taylor and his contemporaries. But here’s the thing, they understood systems and the management decisions based upon these were very good. [As an aside many people find H. Thomas Johnson’s 2006 paper very useful: Lean Dilemma: Choose System Principles or Management Accounting Controls, Not Both. A Shingo Prize winner no less].

Coming back to the story. What do we do? Well we still do apply our instinct to the constraint but we do it in a way that was best described by Debra Smith in 1995 in a report sponsored by The Institute of Management Accountants and Price Waterhouse as “an extreme form of variable costing in which the only costs assigned to products are direct materials” (pg xxii). This gives us the thing sometimes called “octane,” or Throughput/unit time on the constraint, the sum of which is throughput in dollars or whatever currency you choose per constraint period. Remember it is a rate. If its not obvious, some things that might produce a moderate throughput might not use much time on the constraint. Other things that might produce a moderate throughput might use much more time on the constraint. Which one will you favour if you can?

In many cases what we chose to make and market today from our decision analysis using less extreme forms of costing means that we are making less money for the firm than we could. I will suggest that a significant part of any increase in throughput comes from making things that make money over things that do not make as much. Rocket science – right? It works beautifully for internally constrained systems.

Now you think that I have forgotten about the subordination and the non-constraints and in fact I haven’t. What is the paradoxical thing that we do here? Well, we do nothing. All of the rest of the system is accounted for in the operating expense. If you find that hard to imagine then here is a little guidance (remember you don’t have to make a profit and you don’t have to avail yourself to any of these). In chronological order.

  • Noreen, E., Smith, D., and Mackey J. T., (1995) The theory of constraints and its implications for management accounting. The North River Press, 187 pp.
  • Corbett, T., (1998) Throughput accounting: TOC’s management accounting system. North River Press, 174 pp.
  • Smith, D., (2000) The measurement nightmare: how the theory of constraints can resolve conflicting strategies, policies, and measures. St Lucie Press/APICS series on constraint management, 184 pp.
  • Caspari, J. A., and Caspari, P., (2004) Management dynamics: merging constraints accounting to drive improvement. John Wiley & Sons Inc., 327 pp.
  • Du Plooy, E., (2016) Throughput accounting techniques. General Media Press, 244 pp.
  • Schragenheim, E., Camp, H., and Surace R., (2019) Throughput economics: making good management decisions. Routledge/Productivity Press, 276 pp.

There are others, Boaz Ronen’s work for instance, if I start to list these out I will omit someone, and they will never speak to me again, so let’s stop there. BTW if you thought you knew what an accounting book looks like, go check Etienne Du Plooy’s masterful book. What you will see is not just accounting built around an internal constraint, but the whole gamut of accounting decision and economic issues that arise in firms that recognise the dependency in their operations.

So lastly, what happens when the constraint is no longer tangible and internal, what happens when the constraint is somewhat more intangible and external. What happens when things move from our span of control into out sphere of influence? What happens when we move from the constraint, the drum, as a producer that is producing? Well the answer is that the constraint moves to the consumer who is consuming. That’s what our system does, it produces for consumers! Sometimes we measure at one end and sometimes we measure the other. If you should ever examine Viable Vision you will see this latter version in spades.

Either we know where the constraint is, or we know where we want the constraint to be. However, we always have two domains that we can explore, one is quantity: how to increase the physical amount, whether production or consumption, and the other is quality: how to increase the financial amount for any given production or consumption.

Remember one is like quantity, one is like quality, the outcome is throughput. Maybe that is a good place to wind up our discussion on measurements and get back to our central argument.

Let’s Pause for a Moment and Reflect

It might have occurred to you that in the above “analyses” we have something that resembles drum-buffer/batch-rope. Believe me that was not done on purpose, at least not at first. These elements were all there from the small sequence of matrices that this article fell out of, but I probably had the order all mixed up. Wasn’t Eli Goldratt consummate in encapsulating that “system” in just three words.

But also, please see something else. At every step that we have examined there are always two perspectives.

  • Exploitation – instinctful, constrained, and focused.
  • Subordination – thoughtful, non-constrained, and focused.

Exploitation and subordination really are two different things. One deals with system productivity and the other deals with system utility. Let’s pick this up.

System Productivity and System Utility

Let’s consider productivity a little more. The constraint is like the pump of the system. We want it to be busy, we want it to be productive in terms of quantity. But we also want it to be productive in terms of quality too. Do you want your heart to pump unoxygenated blood, or oxygenated blood? Same with a box rather than a body. Do you want to pump goods through that produce little profit, or significant profit? Well that’s exploitation of the constraint. Productivity is about quantity and it is about quality (even in the literal sense of defect-free). And BTW a few people like to write about TOC not addressing quality. But that’s one of the way in which you can immediately create capacity! Why on earth would you not do that?

Let’s put productivity into our structure and add utility too. Then we will discuss utility further.

  1. Identify
  2. Exploit – Productivity
  3. Subordinate – Utility
  4. Elevate
  5. Iterate

Let’s say you are going to the movies – there are two queues, one long and one short. The operator at the end of each queue is equally productive in each case. Which queue is for you?

Now you are a customer for a make-to-order part, or a customer for a die-in-the-ditch patch for your legacy computer system. You have two choices a long wait or a short wait. Which one is for you?

You see in each case the productivity is the same, but there is still a difference, the utility as defined in this case by the customer.

Now remember all that sprint capacity that was going to waste? Well we need that. Because when we offer better utility than our competitors, we will get some of their customers too, and our production will have to go up to accommodate that. It will eat into some of our protective capacity, our sprint capacity. So we actually have to keep building that, we have to keep building the capacity of the non-constraints – and believe me this is counter-instinctual, indeed it is thoughtful. In fact it is of a different logical level.

The Company and the Customer

The company cares most about productivity and the customer doesn’t care about that at all! Think about that. The Customer cares about utility and company doesn’t care about that at all! Think about that too. Want proof? Let’s have a look. Let me ask you, how is it that in the vast majority of cases the sum of all the improvements isn’t very much improvement at all? Of course there are a number of answers, we might scratch the surface with just four:

  1. We usually assume independent events when in fact we have dependency.
  2. We usually have no knowledge of process variation (we bury ourselves in product variation instead).
  3. We don’t usually know where the productivity rate-limiting step is.
  4. We don’t usually know how to address utility.

Now the last might strike you as odd, but its true. If it wasn’t true there would be a great deal more written about subordination. But, as you know, I’m biased (opinionated), and I am using a negative proof which hardly proves anything. So you can either take that or leave it. But what if we summarised all four points above as “we don’t know very much at all?” That would be getting closer to the issue.

But think about it for a moment. When did a tool-head or a tech-head ever shout out: “let’s halve the lead time for our customers!” No, only over their dead bodies – right? Instead, in truth, its “let’s write better and better code in the available time that we can wring out of everything else.” What does agile use as the buffer? Code for goodness sake. Well scope really, but code is the scope. As much code as possible written as correctly as possible within the existing time and resources (read money). Even doing that, we know we can delivery in half or much less of the current lead time, but we don’t. We don’t do so because we don’t really care about the utility to the customer (we might say we do, but action speaks louder than words). Remember that necessary condition: secure and satisfied customers now and in the future. Maybe it should be secure and more satisfied customers now and in the future. Imagine that. Imagine if existing customers gave us more work, or paid more for existing work. Its starting to sound like a win-win.

Goldratt wrote something to the effect that: “Value is created by removing a significant limitation for the customer, in a way that was not possible before, and to the extent that no significant competitor can deliver.” It’s in quotes but I’m damned if I can find the original source. But we might add something more to this too. “Value is also created by adding a significant enhancement for the customer, in a way that was not possible before, and to the extent that no significant competitor can deliver.” That is utility.

Now we are getting somewhere. We pay lip-service to utility. It’s all company and no customer. We don’t understand the emergent effects of providing better utility. By that I mean we don’t understand, in fact can’t understand, the positive reinforcement that occurs out of better availability until, well, after we have experienced it. To cite a trivial but oh too common example, when you can provide ex-stock and the industry standard is, say, twelve weeks, you will get your stock on the sales floor when your competitors have nothing to supply. Retail customers don’t get to see competing stock after a while because it isn’t there on the shop floor anymore, and you sell more as a consequence. It is a virtuous cycle that has to be experienced to be understood. So you sell more, and now that you understand making money, you sell more of the things that make even more money. Nothing but nothing in accepted management accounting will ever tell you anything about that.

Now, I can see that you have jumped ahead of me and can see that productivity will have to rise as a consequence of improved utility. Well productivity will rise, because you are still an ace at instinctful exploitation. But now you are learning something different, something of thoughtful subordination. BTW long may our competitors continue in their instinctful ways everywhere else.

Its Like We are in a Time Warp

Truly, it is like we are in a time warp. We keep banging on about productivity, about exploitation, and about the company, whereas the customer wants subordination and utility. We can give that to them you know. And out of that will come increased productivity.

No alt text provided for this image

But at the moment we seem to have the cart before the horse.

Productivity, Utility, and Commonality

As I started this article I purposefully chose not to include the three necessary conditions. “Three?” I can hear you say. Well the owners of the system get to choose the goal, but regardless of that (and even if one of them is the same) we have:

  • Secure and satisfactory cashflow now and in the future.
  • Satisfied customers now and in the future.
  • Secure and satisfied employees now and in the future.

If you choose to look at what we have been discussing in this article then productivity and utility also map into two of these necessary conditions. I don’t know that they have to, but it just seems apparent that they do. Sometimes life is fractal, maybe this is one of those moments. Let’s have a look.

  • Secure and satisfactory cashflow now and in the future = productivity.
  • Satisfied customers now and in the future = utility.

That just leave the last one about employees. Its still a little early in the days for this, but I will take a stab at it.

  • Secure and satisfied employees now and in the future = commonality.

It doesn’t matter what you goal is, if you don’t have commonality you won’t get your goal either. That’s why it is a necessary condition. I guess that the thing that I am trying to express, and the thing that I have found that drives success in our endeavours more than anything else is roughly expressed as “common purpose.” To cite the mantra of the times: we are all in this together. Each of these three aspects is incredibly interrelated and reinforced through feedback. No feedback means no reinforcement. If there is a ‘them and us” dychotomy between the financial capital and the human capital, well then you are going to skate around on a very thin layer of ice thinking that you are very good, when you are not even within cooee of becoming great.

Lower and Higher Logical Levels

I hope you have caught on that every single subordination step that we have discusses is not only thoughtful, but more over that they are counter-instinctual. In fact they are paradoxical. I used to say counter-intuitive, its the same thing, but its clouded by the concept of expert-intuition. So, let’s stick to counter-instinctual.

Frankly, exploitation and subordination are different logical levels.

Exploitation is about the productivity of the constraint. It is driven by our instinctful reactions, it is driven by our fast thinking. It is inherent to us, it is hard wired into us via our genes. It is, frankly, a lower logical level and we should acknowledge that; least we never escape it. Let’s add this to our structure and for good measure add in subordination as a higher logical level and then we can examine that.

  1. Identify
  2. Exploit – Lower Logical Level
  3. Subordinate – Higher Logical Level
  4. Elevate
  5. Iterate

Subordination is about the utility of the non-constraints. Most of most processes are indeed such non-constraints. Subordination is driven by our thoughtful actions, it is driven by our slow, serial, sometimes quite painful thinking. It is also inherent to us, it is hard wired into us via our genes. But there is something more, it is also recorded in our social memes. This is a higher logical level. It is recorded in books and ideas and cultures that are passed explicitly and implicitly from generation to generation. We record it as a codification. That very simple code, those 3 or 4 or 5 things we listed for subordination is what makes Theory of Constraints so very powerful.


What is it that drove me in this digital age to put digits to the keyboard and pound this out? Part of that answer is a decade long belief that there is very little written about subordination or of its importance, and even less understanding. Don’t get me wrong, there are indeed people for whom subordination is implicitly obvious and this article is of absolutely no use to them in the least. People who understand organic systems (a body) and people who understand engineering and social systems (a box) will understand subordination because they see it all the time. For others I want to point in the direction that subordination is something more, something much more, than exploitation.

I will continue in this vein for a moment longer. I’ve always wanted to do a presentation built around the “S” in the middle of the road. It’s part of how I see subordination. The “S” comes from the “s” in subordination and it also alludes to Kuhn’s analogy of the curve, the bend actually, in the road that we experience in paradigm. We drive into this S-curve at the bottom-left full of ourselves and our instinctful knowledge of exploitation. And we leave this S-curve at the upper-right with a new knowledge and our thoughtful understanding of subordination which to a large extent makes our previous knowledge of exploitation a rather moot point.

As Kuhn said: “From the bend both sections of the road are visible, and its continuity is apparent. But viewed from a point before the bend, the road seems to run straight to the bend and then to disappear; the bend seems the last point in a straight road. And viewed from a point in the next section, after the bend, the road appears to begin at the bend from which it runs straight on. The bend belongs equally to both sections, or it belongs to neither. It marks a turning point in the direction of the road’s progress …” (1957, The Copernican revolution, pg 182.)

Who would have thought that historians of science could be so poetic!

I’ll leave you with one more thought. There is a razor here:

don’t mistake exploitation for subordination (because it isn’t)

Dr. Kelvyn Youngman