A Guide to Implementing the Theory of Constraints (TOC)





Next Step



Bottom Line


Supply Chain

Tool Box



& More ...



Drum Buffer Rope

Implementation Details

Batch Issues

Quality/TQM II





How Do We Implement Drum Buffer Rope?

In the previous section we examined the mechanics of drum-buffer-rope in a rather basic form – an idealized framework.  In this section let’s try to flesh out this framework with a little more implementation detail.  Of course the best way to learn this detail is to just to go out and do it.  Drum-buffer-rope and the Theory of Constraints is a robust methodology so long as we keep the rules of engagement and plan of action in mind.

We will use the 5 focusing steps – our plan of attack – as our template.  We probably can’t repeat these steps too many times so let’s repeat them again right here;

(1)  Identify the system’s constraints.

(2)  Decide how to Exploit the system’s constraints.

(3)  Subordinate everything else to the above decisions.

(4)  Elevate the system’s constraints.

(5)  If in the previous steps a constraint has been broken Go back to step 1, but do not allow inertia to cause a system constraint.  In other words; Don’t Stop.

OK then, let’s start at the start.

How Do We Identify The Constraint?

In any normal production process, either manufacturing or service based, there will be work-in-process everywhere, especially if the plant is run as a balanced line.  Even more so if you use some form of MRP or ERP.  How do we find the weakest link?  In this situation every step looks like the weakest link with large amounts of working waiting to be done at each stage.

There are 3 ways;

(1)  Find the step with the longest waiting time for work to be completed.

(2)  Find the step that most often causes disruption downstream.

(3)  Nominate something.

Doing something, anything, might seem counterintuitive, but if you haven't chosen the real constraint, then it will make its presence felt very quickly. This is a process of on-going improvement; don't be scared to do something – anything – because each step will take you closer to your company's goal.

Let’s Collect Data And Write A Report!

No, no.  Please don’t do this.  Collecting data and writing a report should not be considered as doing something, at least not something that effective people do.  Most likely it should be considered as doing nothing.  Data collection is usually a stalling tactic.  The data is never good enough to make a decision on – which means you will need to collect even more data.  Remember this is a dynamic complexity problem, so don’t go and use detail complexity solutions such as searching for more accurate data on it.  Normally with drum-buffer-rope you should have demonstrable results in the same amount of time as others would have spent collecting data.  So it becomes a choice, take an action, or collect data.

Moreover, report writing usually doesn’t generate ownership or understanding of the real problem, it might generate an understanding of the symptoms, but not usually of the underlying core problem or core conflict.  Reports generally do not involve all the needed people and are usually perceived to apportion blame.  Note, perceived to apportion blame.  So even the most neutral and factual report will be perceived by some people – usually the people in the problem area, to be apportioning blame.  There is usually only one response to this by the people involved, defensiveness and deflection of the perceived blame onto another area.  Naturally enough this will not help to improve things at all.

The Thinking Process graphical trees are the tool of choice for generating buy-in and understanding of the problem and then seeking consensus on the solution.

In addition don’t be afraid to go and stand in the middle of your process and just watch what is happening.  People will get used to you being there fast enough.  And you are in good company, Taiichi Ohno the inventor of Toyota’s just-in-time once said; “We can talk about work improvement, but unless we know production thoroughly we can accomplish nothing.  Stand on the production floor all day and watch – you will eventually discover what has to be done.  I cannot emphasize this too much (1).”

It is likely that people with an affinity for understanding flow will naturally spend considerable time on the floor; don’t be afraid to do this just because no one in the company has done it before.  This is the reality check between what people tell you is happening in production and what really is happening.  The Japanese have a saying; go to the actual place, see the actual problem, and speak to the actual people.  Just common sense really.

Wandering Constraints

Almost all implementations in the pre-drum-buffer-rope stage claim to have wandering bottlenecks.  Don’t worry, there will almost certainly be work-in-process everywhere as well, and this is the cause of the impression.  However, later – as things begin to flow – there may still be the impression that there are wandering bottlenecks from time to time.  In this instance it is most likely a case of wandering large batches.  Large batches will hold themselves up at slow process stages.  Reduce the maximum batch size and the problem will go away.

Usually the people in the process have the best idea of where the constraint might be.  If all else fails then just nominate something, and the real constraint will make itself known within a short space of time.

How Then Do We Exploit The Constraint?

Initially the constraint is protected by a large pile of work-in-process; there is always something to do.  But is the constraint always doing something productive?  Is the constraint sometimes doing things that are unproductive?  The best way to find this out is to write a plan, a schedule for the constraint.  If you already have schedules for the plant in place, this might be the very first time that you write a schedule in detail for one resource.  This can be as simple as a whiteboard on the production floor.  The important thing is to write a plan of how to exploit the constraint over the next short period.  Then we have a basis to evaluate our actions and our results.  The schedule will let you peel away the obstacles to proper exploitation one by one.

Exploitation tactics on the constraint in manufacturing include; making sure automated machines run during breaks, making sure set-ups are staggered, fixing break-downs quickly (or preventing them completely), replacing absent workers quickly, avoiding downtime during shift changes, avoiding multitasking of set-up personnel, keeping faulty work or unnecessary work off the constraint.

Of course people will have always tried to do this; it is just that they didn't know where to focus on in the past.  Now they do.  The focusing itself will greatly improve the constraint.  For example when management sees the set-up man on the constraint waiting for a laptop to program a machine and that the laptop is currently at some non-constraint machine on the other side of the plant, they will take action.  Whereas one week prior the only action would have been to ask the constraint operator to be patient.

Schedule The Constraint

This doesn’t have to be sophisticated – especially if your operation isn’t particularly complex.  Even if your operation appears complex prior to drum-buffer-rope, you might be surprised how it looks with in a few short weeks.  A constraint schedule might be written on a white board on the work shop floor, or a sheet of graph paper, or travelers/paperwork hung in sequence on a rope near the constraint.  The synchronization might not be anything more than one job finished at the constraint means one job admitted by the gating operation.  The important thing is that is understood and that it works.

More complex operations in all likelihood already have an infinite capacity MRP-based scheduling system in place and then it is a matter of adjusting this to work in a finite capacity drum-buffer-rope environment.  There are drum-buffer-rope schedulers available, for instance see Manusync in the links and resources, however it is infinitely better to get the organizational understanding and  behavioral change accomplished first and then look at more automated options and to leverage off the automation.  Automation before understanding is like putting the cart before the horse and it is not known to be effective.

Shadow Constraints

What if, for some reason, the actual constraint is difficult to schedule?  No matter, you could consider nominating a shadow constraint somewhere else – just make sure that it is “hobbled” by the schedule so that it can’t exceed the throughput of the real constraint (2).

Schedule Discipline

When the schedule and reality deviate for whatever reason, then schedule reality.  Generally this means re-synchronizing the current constraint sequence to a new time.  Of course proper buffer management should avoid such resynchronization – but stuff happens.

How Then Do We Subordinate?

We must ensure that we subordinate our local actions to that of the goal (global outcome) of the system.  For actions read; decisions, behaviors, measurements, and policies.  Normally we do that via the plan we write for exploiting the constraints – the schedule.

When we have a written exploitation plan in place we can quickly compare actual results obtained with the planned results.  We will soon see deviations.  As we saw in the section on process of change, these deviations from the plan results from (12);

(1)  Not doing what was supposed to be done.

(2)   Doing what was not supposed to be done.

The deviations from plan result from our failure to properly subordinate.  Let’s have look at some the more common subordination issues in more detail.  Then we will look at two local performance measurements that we can use to align and monitor our subordination efforts with.  We will also consider a pivotal role in subordination; the role of the buffer manager.

Buffer The Constraint

It is not enough to write a schedule, we must also protect the constraint to make sure the correct material is always ready and waiting for processing in good time before the start date at the constraint.  We can achieve this by ensuring that we release material a rope-length in time prior to the scheduled consumption on the constraint.

Previously we applied a rule of thumb which was to make the rope length half the current lead time from gating release to the constraint start time.  We then divided this rope length into 3 equal buffer zones.  Essentially we expect most jobs to be ready and waiting at the buffer origin (a place in front of the constraint) within 2/3rds of the buffer time available.  But just how much is most?  It seems to vary; original estimates seemed to be of the order of 90% (4).  More recent estimates suggest 95% (5).  It seems much more likely that the actual percentage is best determined by the volume of jobs in each situation.  Even though zone 1 was called the expediting zone initially, more recent terminology seems to favor dropping this term and simply calling it zone 1.  Expediting has such negative connotations in manufacturing.

There is some confusion however in the literature over the definition of the buffer.  In fact this confusion arises more as a consequence of the difference in interpretation between the buffer and zone 1 of the buffer.  The total buffer time is divided into 3 parts.  The zone closest to the buffer origin should normally “be full.”  Some authors then call this zone 1 component of the entire buffer as “the buffer.”

Schragenheim and Dettmer, Stein, Caspari and Caspari follow Goldratt’s lead and take a total lead time approach to the buffer definition (6-8).  Woeppel, Cox and Spencer, and Umble and Srikanth take a zone 1 definition of a buffer (9-11), that is they treat it as a discrete addition of time in front of the constraint.  It seems that those who relate more closely to MRP environments prefer the zone 1 definition of a buffer.

Sometimes a zone 1 definition of the buffer helps people to understand the local versus global safety argument.  Let’s have a look at that next.

Local Safety Argument

Umble and Srikanth were perhaps the earliest to publish on the detail of drum-buffer-rope independent of Goldratt.  They chose a Zone 1 definition of the buffer (11).  Basically this develops out of an argument for local safety everywhere versus safety as strategic buffers in a few key places.

Probably few managers would deny the importance of a special buffer in front of an identified constraint – many more, however, would be hugely reluctant to give up their local safety everywhere else.  They would prefer to keep all their local safety and create some additional safety in front of the constraint.

Because the concept of local safety everywhere is second nature to manufacturers it is sometimes difficult for managers to see the important characteristics of global buffers.  As previously mentioned, if an MRP or ERP system has been implemented, the local safety values might be quite explicit.

If you are interested the argument for moving from local safety everywhere to global safety in a few key places is here.

Measurements & Communication

Proper subordination not only means making sure that we use the right measures, but also means making sure that the right measures are continuously communicated to the line.  The most important global measure is physical output and/or financial throughput.  If the constraint is internal then this should go up, if the constraint is external then it may initially stay constant, but it too should go up. These results should be visible to everyone and updated as frequently as possible.  Updating once a day is good and is not too frequent.  But having visible results is not enough.  The line management must reinforce positive results with a few well chosen words to the right people at the right time.  That reinforcement, like water, will filter down to all levels and in all directions.

Reluctance to praise will be viewed as failure to support the initiatives.  You would be amazed how some managers are reluctant to praise significant results.  You can be almost sure that the manager concerned does not fully support the implementation.  It is a very passive way of communication, do everything requested but say nothing.  Such managers can block praise from the leadership meant for the line by failing to pass it on.  It happens.

By the same token the leadership must reinforce positive results with a few well chosen words to the right managers at the right time.  Without that feed-back the managers may decided that leadership support is lacking.

Speed Wobbles

All implementations develop speed wobbles.  The analogy is a child who has just learnt to ride a bicycle, maybe straight-off, or maybe for the first time without dummy wheels or stabilizers or what ever you wish to call them.  As the child goes a little faster, at some stage they will decide that if they go any faster they will fall off – and they do.  Our people may never have produced as much output before without some sort of assistance (overtime maybe), they certainly won’t have done it with less inventory as well – they will convince themselves that they can’t continue to do it and things will come unstuck for a few days.  And then we just pick ourselves up again and go straight past the point where we thought we couldn’t go past before.  What is really happening is that people are learning to subordinate properly for the first time.

Work-In-Process – A Previous Indicator Of Security And Success

We have mentioned many times of the need to reframe the environment from one of a reductionist/local optima approach to one of a systemic/global optimum approach.  One of the most important aspects of this reframing in a processing setting is the perceived value of work-in-process held by the guys on the floor.  It is the epitome of local optima.  A large pile of work-in-process in front of each work center in the morning is a measure of importance and security to the members of that work center.  Seeing that work-in-process moved to the next work center by the end of the day is a measure of success and satisfaction to the members of the work center who processed it.  If we reduce work-in-process in this setting then we must “unhook” these measures of security, success, and satisfaction and “rehook” them to the global measures of the system.  People will get used to the new reduced inventory way quite quickly, they will develop pride in how “the work I started this morning is now way over there,” but it needs careful nurturing and the positive reinforcement mentioned above during the transition.

If managers find the importance of work-in-process hard to believe, then try operating with a clean desk policy (a really clean desk) for a couple of days and count the number of people who will question whether you have any work to do.  These pressures are pervasive and very powerful and can undo an implementation.

Gating Issues

To properly subordinate the process to the constraint we must pace new material release to the pace of the constraint.  To do more only builds work-in-process.  Many times early on there are likely to be pressures to admit extra work.  Often there are not-so-subtle pressures from sales people to favor certain clients – always their clients and not anyone else’s clients.  It requires good leadership in both sales and manufacturing to avoid this.  Another reason for using the holistic approach prior to the implementation.

An important part of subordination is to make sure that the “drumbeat” is relayed to the start or gating operation and is observed.  This means that new scheduled work should not be admitted into the system any sooner than scheduled.  Again in the early stages there is a great temptation to admit work early to keep the next several stages “busy” and to “get ahead” at little.  Doing this will result in non-constraints at some stage working on work that is not needed to be worked on and not working on work that is needed to be worked on.  Local work-in-process will go up, local lead times may increase, and if this is a consistent policy, holes in the buffer will surely arise, and throughput will go down.

The subordination challenge for gating operations and all other non-constraints is to keep work-in-process and lead time low and constant and the buffer full (downstream).  To do this we must protect their sprint capacity.

Protective Capacity/Sprint Capacity

The capacity that all non-constraints have above and beyond the capacity of the constraints is often termed protective capacity and sometimes called sprint capacity (13).  Sprint capacity seems to be a good description of this factor.  It is a description of the ability of non-constraints to speed up (actually working longer at their normal rate) as the situation requires in order to get work to the constraint on time after some variability in the system has intervened such as; worker absence, machine downtime, rework issues, or material supply.  Sprint capacity is a fundamental and absolute component of drum-buffer-rope.  Let’s examine this in more detail.

All systems have 3 elements in common (14);

(1)  Finite Capacity.

(2)  Variability.

(3)  Dependency.

This much we know already.

However, for a given set of dependencies, there exists a dynamic relationship amongst;

(1)  Sprint capacity.

(2)  Variability.

(3)  Inventory (work-in-process).

So, the better the sprint capacity, the more likely we can handle variability and maintain our work-in-process.  If we reduce our sprint capacity without reducing variability, the work-in-process must go up (and lead times will go out).  If we can maintain or increase our sprint capacity and/or reduce our variability through total quality management, kaizen, or whatever, then work-in-process can go down.

Argued from a different perspective; increases in sprint capacity can allow decreases in buffer capacity.  Conversely decreases in sprint capacity will require increases in buffer capacity.  The latter is very important where demand grows as a consequence of implementing drum-buffer-rope.  Increased demand on the non-constraints decreases their sprint capacity and may cause holes in the buffer to emerge.  There are two options; increase the sprint capacity of the non-constraints or increase the buffer capacity of the constraint.  In fact sprint capacity, cross training of employees, duplication of capabilities, having more that one supplier, are all cases of buffering the system (15).  Whether the buffering is explicit or implicit the effect is the same.

Ideally, we must continually hunt out additional sprint capacity.  We saw in the section on accounting for change the strategic implications of near-capacity constraints and we will repeat the argument at the end of this page.  Fortunately buffer management gives us the control mechanism to stay a couple of steps ahead of eroding sprint capacity.  There is, however, one issue that will destroy sprint capacity faster than anything else – poor road runner ethic.

Road Runner & Activation/Utilization

The kanban in just-in-time gives workers permission to work only when there is work to be done, and permission not to work when there isn’t work to be done.  Of course in processes where there is total quality management or total preventative maintenance the chance that there is nothing to do at all is pretty slim indeed.  In fact these systems need slack built in so that such activities can take place (16).  This is in fact how additional sprint capacity is built although it is not called that in these methods.

Theory of Constraints drum-buffer-rope uses the road runner ethic instead.  The analogy is that the road runner (both fictional and in real life) has just two speeds, fully-on, and fully-off, and this is how non-constraints should work in drum-buffer-rope.  If there is work; work on it at normal pace, if there isn’t work; don’t work on it.  However, it seems that some people take the cartoon interpretation too literally and believe it to mean working at more than normal pace.  This is not the case.  The road runner ethic says just two things;

(1)  When there is work, complete it promptly.

(2)  When there is no work, find something else to do.

Implementations get into trouble when there are reluctant road runners.  Reluctant road runners are a management or leadership issue; it is not a worker issue.

Another way of looking at this situation is through the terminology of activation and utilization (17).  Activation at its most basic interpretation means working for the sake of appearing to be busy.  Utilization means working only to create or protect throughput.  We saw previously that one of the synchronous manufacturing principles was: Resources must be utilized, not simply activated (17).  This is a re-verbalization of the need for adequate subordination.

Reluctant Road Runners

If there are local efficiency measurements still in place then don’t expect to see roadrunner.  Don’t expect to get the same results either.  Implementing roadrunner while having another older and stronger counter measurement will cause it to fail.  Timesheets broken down by job or task for non-constraints is a powerful signal to be busy all the times.  Even when the system owner in such an environment includes a code for “Roadrunner” on the time sheets – which in effect means “I’m idle” – do not expect to see it used.  Believe me.

There is quite an old term used in manufacturing to describe the practice where people work just as slow as necessary and no faster – soldiering (18).  Taylor sought to a large extent to remove this through standardized work practice “not how long did a job take to complete but, how long should it take.”  Of course this occurred within the framework of the reductionist/local optima approach known as scientific management, or as Goldratt calls it “efficiency syndrome (19).”  We know this as; “If a worker does not have anything to do, find him something to do! (20).”  The corollary is “I’ll always make sure it appears that I have something to do.”

As we reframe the situation from a reductionist/local optima approach to a systemic/global optimum approach we must be aware of not only the presence of new measurements and behaviors, but also of the presence of old measurements which continue to underline old behaviors.  If you have reluctant roadrunners then look for “legacy” measurements that are causing that behavior.

I have seen the so-called “laziest” group in a factory embrace roadrunner, and use the spare time to maintain their gear – something that previously they had been hard pressed to do.  To understand this, just think back to the section on people.  People will make the best decision that they can based upon their map of reality.  If you change the map of reality you will change the behavior as well.

If you measure people based on local efficiency you will get soldiering – guaranteed.  If you measure people globally, you will get roadrunner – guaranteed.

You can “wing” a large amount of improvement in a process just from improved exploitation without necessarily utilizing full subordination such as roadrunner.  For instance you can obtain impressive and rapid improvements in output where time sheets are still in force, but the result will be soldiering in several departments and the illusion within middle management that it is impossible to improve any further – after all the guys are fully busy.  It really depends whether we want a process of on-going improvement or a one-off gain.

That Traffic Light Analogy Again

Let’s complement our understanding of the roadrunner ethic with the traffic light analogy that we developed in the section on process of change.  When a non-constraint is working it is creating throughput, and when a non-constraint is not working it is creating protection for throughput.

We can consider this as stop and go, or on and off, or creation and protection.  This is roadrunner.  It says nothing about going faster than normal; it just says go (normally) or stop.  However, when we go slower than normal, when we soldier, we are at a half way position, let’s look at that.

When we soldier we do two things.  We often fail to fully create throughput and we often fail to fully protect throughput.  It is a double edged sword.  Basically we destroy throughput.   We destroy throughput on two different fronts at the same time.  We need to gain a much better appreciation of this.

Batch Sizing

Batch sizing is an important part of subordination.  Sufficiently important that we will give it a page of its own.  Obtaining agreement to modify batch-size is a major part of most process implementations.

Material Handling

As work-in-process decreases there are going to be a lot of devices left around that the work used to sit on and also maybe moved on.  Remove these at a healthy and proactive rate as they become free.  If there is somewhere for people to put work down on, then people will find work to put down on it.  When there is insufficient work available to cover the “empty” spaces people become uncomfortable.  Remove the cause of the discomfort – the empty and unnecessary pallets, trolleys, or what-have-you that is in excess.  Clean them up and make sure people know they have been put away for safe keeping.  We don’t expect to ever use them again; but most others will be expecting that they will be used again shortly.  This way everyone is happy until we all forget that it was ever an issue.

First-In First-Out (FIFO) Discipline

This one requires patience, persistent patience.  Let’s go back a step or two.  Many drum-buffer-rope implementations will have previously had some sort of MRP scheduling system in place, or in its absence some sort of “hands-on” method of expediting.  The whole rationale of the scheduling component of MRP in a job shop or repetitive batch environment is that it is able to reschedule any process stage by reprioritizing the sequence that work waits in queues prior to each stage of the process (21).

Drum-buffer-rope removes the need for queuing everywhere but not the bad habit of reprioritizing work in the queue.  Some jobs may simply float for a day or two until they are done – continuously passed by more recent jobs that have arrived.  In cultures where seniority is highly respected, and there are senior operators, it may be extremely difficult to get first-in-first-out operating initially.  Persist.

Because material is released to the system in the sequence that it will be processed on the constraint, then in most cases simple first-in-first-out will preserve the constraint schedule sequence.  Failure to observe this will quickly lead to holes in the buffer.  Work that is passed over at a non-constraint will be detected as late at the buffer.  Work that is favored will be detected as unusually early at the buffer.  Everyone knows which are the “sweet” jobs, if they start to turn up early then be paranoid.

At non-constraints; “The method of determining priority is first in first out (FIFO).  Since the schedule for the constraint and the amount of protection required determines the release date, the sequencing of orders has already been determined and should be by arrival data/time to the non-constraints.  Unless there is a problem causing an order to penetrate zone one in buffer management, no effort is made to change this rule (22).”

In large volume conditions however we need to note the following for non-constraints; “In the case of a shop that has many jobs arriving in a day, a method has to be devised to determine the sequence in which the jobs arrive.  This will dictate the sequence of processing.  The procedure addresses two things – how to know what arrived first, and how to make sure people work in that sequence (23).”

Local Performance Measures – Lateness

Even in the very best systems, occasionally “stuff happens” and we can’t always deliver on time or ex-stock as we initially promised.  Material shortages are probably the number one cause for this.  Over-promising by non-manufacturing personnel is probably the number two reason.  As a consequence usually the IT guys get given a project to do to build some meaningful measure of lateness.  Why the IT guys?  Well it has numbers in it and uses the MRP system so it must be an IT job.  Of course the IT guys come up against the age-old problem of whether to measure in units of time or units of sales value.  Do we measure the number of days late and under-report a big job that was just a day late or do we measure sales value and under-report the small job that was held up for several weeks while we waited for a part from the outsourcer?

The answer is that we measure both, using the value of throughput-dollar-days (12).  On any given day we multiply any late order’s throughput by the number of days that it is late and sum for all the late orders.  Of course it should be zero but sometimes it won’t be.  For orders whose lateness impinge upon our customers we should measure from the original required delivery date for make-to-order and the order date for make-to-stock (goods should be availably ex-stock at the time the order is placed).  However, Goldratt suggests that in order to avoid this most undesirable situation we should also “tie” the lateness to buffer holes (12).  "Sometimes an action by the organization is triggered due to a deviation in a local department.  This occurs whenever a task does not arrive at its buffer‑origin even though enough time had elapsed since its release, enough time to cause us, in quite high probability (let's say over 90 percent), to expect the task's arrival.  Thus, we might start to count the days from the point in time when the task penetrated into the expediting zone, rather than from the order due‑date.  This will give us time to correct the situation before damage to the entire company is fait accompli."

We might consider this to be a tactical application.  Thus a work order that is late to zone 1 of any buffer should have in the buffer report a value for throughput-dollar-days.  Let’s draw that.

The place where we expect to see the goods accumulate in front of the constraint is called the buffer origin (24).  The scheduled start time on the constraint also marks the end time for the buffer.  Let’s call this time exactly that – the buffer end time.  To ascertain if we have a problem we must check at one zone prior to the buffer end time to see whether the completed work is physically present at the buffer origin or not.  Let’s call this the buffer checking time.  If the work is not present at the buffer origin then we must go and find where it is and determine the cause of the delay.  If it seems probably that the work will not reach the buffer origin by the buffer end time without intervention then we must “facilitate” its progress.  When it eventually does turn up in completion at the buffer origin then we can calculate the lateness.  There are two distinct actions here.  The first must be done in real time; the second can be done from records.

Lateness = Rope Length – One Zone Length – Actual Duration

Using the example from the section on drum-buffer-rope.  Our rope length was 9 days and each buffer zone is 3 days.  Therefore any work not at the buffer origin in (9 - 3 =) 6 days should be located and any apparent cause for the delay recorded.  If the job actually takes 6.5 days then the hole in the buffer is 0.5 days long.

Note once again that the buffer end time, the zone length, and the buffer checking time are all measures of time which we have drawn in space on our diagram.  Viewing buffers like this is a major source of confusion caused by the limitations of our two dimensional representation.  

The only person who needs to know the buffer checking time for a job is the buffer manager.  Don’t add it to the plant schedule or else it will immediately become an intermediate schedule date and people will leave work until the last minute – and it will be late.  In effect having such a date on the schedule creates a “concrete” zone 1 buffer in most people’s minds.   This is not correct; the buffer extends from material release until the scheduled beginning on the constraint.

Using buffers to manage in this way we not only have information on the probable location of the problem and the frequency of incidence but we also have some measure of the severity from the system’s viewpoint.  The more days late, and the more throughput at stake, then the more severe the problem.  This information is currently missing from most buffer reports.  Stein advocates exactly this when preparing a buffer hole Pareto chart (25, 26).  Let’s try and show Stein’s argument graphically.

Currently a Pareto chart of buffer hole incidence might look something like this;

However, for each incidence we must do the following;

Severity = Throughput x Lateness

This can change the way that we view the information.  For instance the graph might now look like this;

Maybe this is a bit of an exaggeration, but it serves to illustrate the point that a buffer hole that is infrequent, compared to other locations or causes, might actually be quite substantial when it does occur and moreover endanger a considerable amount of throughput.  Alternatively, buffer holes that are frequent might represent repetitive “near-misses” of no great importance to the scheme of things.  We need to remember how important throughput-dollar-days are as a measurement.

This is valid information for the 5-10% of orders that we have to expedite.  But what if the order is no longer at the resource that caused the lateness.  How can we know what to improve in this instance?  The answer comes back that we must deepen the statistic (27).  A more satisfactory concept might be that of a “tracking zone” established expressly for this purpose (27).  Stein advocates extending the buffer checking point out to zone 2 (25, 26).  It is now far more likely that an order will be found at the location that is causing it to be “late” and it is also far more likely that emergent problems will be found before they become critical.  We might view this to be a strategic application of buffer management.  Let’s draw that.

To see what is happening for tracking purposes we must check at two zone lengths out from the buffer end time whether the completed work is physically present at the buffer origin or not.  If not we must go and ascertain at that time where the work is and any cause that might have produced a delay.  When, however, it does turn up in completion we can calculate the tracking hole;

Tracking Hole = Rope Length – Two Zone Lengths – Actual Duration

Again using the example from the section on drum-buffer-rope.  Our rope length was 9 days and each buffer zone is 3 days.  Therefore any work not at the buffer origin in (9 - 3 - 3 =) 3 days should be tracked and its location recorded.  If the job actually takes 4.5 days then the tracking hole is 1.5 days long.

I don’t think the intent is that tracking should be undertaken for every job, even though our manufacturing system might be able to show us every hole in the tracking zone.  It is unlikely however that we will know where the job is without physical inspection, and so this type of data analysis should be periodic at best – sampling the system performance from time to time in other words.

Local Performance Measures – Inventory

A process is like a sponge; we go about squeezing the excess work-in-process out of it, then let it rebound to its original shape and put it back down on the bench.  WARNING, if there is any water on the bench the sponge will simply soak it up again.  If you don’t have the vigilance of a hawk your system will start to build inventory as soon as possible.  The previous best way to keep a tag on this is either by a measure of work-in-process or a measure of lead time.

As with our measures of lateness we once again have two measures, either a monetary value of work-in-process or a time value.  Once again the best solution is to use the measure of inventory-dollar-days (28).  For a whole process we know the gating date, we know the current date so we can work out the duration that it has been in the process, and we know the raw material value so we can calculate and sum the inventory-dollar-days for the process.  If it goes up then work is “aging” or too much work is being admitted.

Moreover, as a drum-buffer-rope implementation begins to give increased sales the volume of work in the process will have to increase unless the rate of flow (the lead time) decreases.  Use inventory-dollar-days as a way of continually driving the system to better levels of performance.  Don’t let the value rise because it indicates that your process is becoming slower – customers don’t like that.

At a departmental level, inventory-dollar-days is a good way to keep a handle on aging work if sweet jobs are queue jumping other jobs.  Failing this, then at least have a daily graph of local work-in-process versus output.  From this you can derive an indication of local departmental lead time.  Why do this?  Well, as work-in-process goes down, people often slow down too.  This graph will focus on the positive of the decreasing lead time and warn whenever it is rising.  It also removes the focus from the total work-in-process.  Remember that maybe just a few weeks earlier a large work-in-process was a measure of how busy and important a department was.  We have to replace that measure of importance with another – short lead times.

Buffer Management & The Buffer Manager

The concept of a buffer manager is pivotal to the smooth execution of drum-buffer-rope.  The actual person who occupies the position will depend upon the size of the firm.  For a small firm the buffer manager may well be the owner or a foreman – indeed the foreman if the firm is not too large.  In larger firms it might be a senior planner or the production manager.  Whoever it is, they must have the vested authority to determine the priority of work in the non-constraint areas.

Buffer management has two functions which we have seen; strategic and tactical.  The tactical function is to enable the limited expediting of jobs that might not otherwise reach the constraint on time.  The strategic use is to monitor the size of the buffer and conversely the aggregate sprint capacity of the non-constraints.  This will identify potential emergent constraints.  We now have sufficient system-wide vision to decide whether to invest in additional capacity, or better subordination, or longer buffers.

Buffer management reporting is the subordination control mechanism that replaces legacy efficiency reporting.  Buffer management reporting is a replacement, not an add-on mechanism (29).


“If anyone adjusts a stable process to try to compensate for a result that is undesirable, or for a result that is extra good, the output that follows will be worse than if he had left the process alone (3).”  The process of interfering in a stable system is often known as tampering.  It is the epitome of a combination of people wanting to do their best, and the effects of dynamic complexity.  A local action (in time) that is perceived to be positive may have negative ramifications for the system as a whole at some later time.  We mentioned the impact of this – the failure to “learn from experience” in the introduction.

Essentially the stages of identification, exploitation, and subordination seek to stabilize the system around the goal, usually maximal throughput in manufacturing or output in a not-for-profit organization.  Be careful however, just because you have removed expediting from the shop floor for instance, there will be other people who will “tamper” with the resulting stable drum-buffer-rope process with the best of intents.  However they will cause harm.  Again it arises from failure to learn from experience because the feedback is blocked by distance and time.

Consider for a moment the example where the sales function “forgets” to tell manufacturing about a promotion in good time – has that ever happened?  And then the sales function panics and interferes with the schedule.  Sales meet their short term objective of getting additional promotional stock, but harm other orders.  It is much better to allow the buffer manager to deal with the problem as best possible within the existing system.  It is absolutely essential that the harm caused by tampering is highlighted to everyone.  Better still make sure the holistic approach has been used so that all parts of the process are aware of the constraint and the effect of tampering before the implementation begins.

How Do We Elevate?

Elevation is so system specific that it isn’t possible to address it here.  It may however be some considerable time until we need to elevate if the exploitation and subordination stages are approached as a true process of on-going improvement.

If The Constraint Is Broken – Go Back

If a constraint is elevated sufficiently that it is broken, then we must go back to the first step and identify the next constraint.  Well, that is if we are not to allow inertia to take over as satisfaction with the current status.  Of course we would expect that buffer management would have identified the next physical constraint.  It may not however adequately identify the policy constraints that give rise to physical constraints.  So we shouldn’t forget to expose our investigation to the rigors of the Thinking Process current reality tree when we begin to identify a new constraint.

There is another aspect to this stage of the process.  We can make it quite proactive rather than passive by selecting a strategic constraint.  Indeed as knowledge of the way the business generates throughput or output improves it is very likely that a strategic constraint will be selected and managed.  Let’s look at this further.

Strategic Constraints

In the first stages of an implementation people are going to raise the issue of “if we exploit this constraint, we may in fact elevate it above something else and the constraint will move to there.”  “AND if we begin to exploit the new constraint, it too, might be elevated above something else and the constraint will move once again.”  Exactly right.

It could sound like a weak excuse to do nothing, in reality it is peoples’ intuition telling them that, yes, if we know where to focus we can break constraints easily.  And at the same time that same intuition is telling people that they really don’t know where the next constraint will be and with that there is a feeling of powerlessness or loss of control.

After we begin to dry a process out we can begin to make conscious decisions about where we, the management, want the constraint to be.  In effect we are choosing a strategic constraint; we are making a decision to always build sprint capacity ahead of building capacity on the constraint.  There are good reasons for this.  The constraint dictates the way in which the firm will make money – and capital investment, product development, marketing and sales.

Newbold, using the language of leverage points rather than constraints; reworded the 5 focusing steps to accommodate this strategic perspective (30).

(1)  Select the leverage points.

(2)  Exploit the leverage points.

(3)  Subordinate everything else to the above decision.

(4)  Elevate the leverage points.

(5)  Before making any significant changes, Evaluate whether the leverage points will and should stay the same.

The new key words are; select and evaluate.

This requires a more focused approach to non-constraints and near-capacity constraints.  Near-capacity resources must be removed (30).  Therefore;

(1)  Identify near-capacity constraints.

(2)  Evaluate their significance.

(3)  If necessary, remove their impact.

Essential this is a type of “subroutine” loop for near-capacity constraints running within the subordination step above.  We drew this on a sub-page off the page on evaluating change.  Let’s draw a similar diagram here.

In practice this means that some current non-constraints and all current near-capacity constraints must be evaluated and steps taken to ensure that they can continue to correctly subordinate to the strategic constraint.  If demand or product mix changes so that they have insufficient protective capacity then they must be elevated prior to the elevation of the strategic constraint and elevated sufficiently at that time relative to the future capacity of the strategic constraint that they do not inadvertently become the constraint themselves.

The identification and evaluation of near-capacity constraints is of course a strategic function of buffer management.  Using the tracking zone from time to time will help to answer these strategic questions.

A Postscript To Implementation Details

Don’t ever underestimate the power of being on the floor or wherever the real work is done.  Things will go wrong.  You will be able to solve problems with direct management involvement in 15 minutes on the floor that you will not have been able to solve with the same management in 55 minutes of “office” meetings.

Let’s repeat a useful trinity from the beginning of this page.

Go to the actual place

See the actual problem

Talk to the actual people

If you can’t do this with the management then you are being stonewalled.  Don’t worry, the management will be surprised at the new found speed of problem resolution and will grow to like it.

Where does this trinity come from?  Well, I learnt it from TQM but could never track the source down despite the smoking gun of Ohno’s admonishment to stand all day until you understand the flow of the process.  It does, however, indeed come from Toyota and is a verbalization of the Toyota principle of genchi genbutsu (31).  It is really an acknowledgement that the required knowledge can only be gained tacitly rather than explicitly.


We have seen in the previous section the basics of drum-buffer-rope.  In this section we have seen some additional implementation detail.  However, there are some issues that need to be addressed further.  Therefore, over the next 4 pages we will look at; batch issues, quality, alignment, and the time required for implementation.

Let’s have a look at batch issues next.


(1) Ohno, T., (1978) The Toyota production system: beyond large-scale production.  English Translation 1988, Productivity Press, pg 78.

(2) Woeppel. M. J., (2001) Manufacturer’s guide to implementing the Theory of Constraints.  St. Lucie Press, pg 87.

(3) Deming, W. E., (1982) Out of the crisis.  Massachusetts Institute of Technology, pg 327.

(4) Goldratt, E. M., (1990) The haystack syndrome: sifting information out of the data ocean.  North River Press, pg 131.

(5) Caspari, J. A., and Caspari, P., (2004) Management Dynamics: merging constraints accounting to drive improvement.  John Wiley & Sons Inc., pg 180.

(6) Schragenheim, E., and Dettmer, H. W., (2000) Manufacturing at warp speed: optimizing supply chain financial performance.  The St. Lucie Press, pg 124.

(7) Stein, R. E., (1996) Re-engineering the manufacturing system: applying the theory of constraints (TOC).  Marcel Dekker, pp 112-115.

(8) Caspari, J. A., and Caspari, P., (2004) Management Dynamics: merging constraints accounting to drive improvement.  John Wiley & Sons Inc., pg 175.

(9) Woeppel. M. J., (2001) Manufacturer’s guide to implementing the Theory of Constraints.  St. Lucie Press, pg 140.

(10) Cox, J. F., and Spencer, M. S., (1997) The constraints management handbook.  St Lucie Press, pg 86-87.

(11) Umble, M., and Srikanth, M. L, (1996) Synchronous manufacturing: principles for world-class excellence.  Spectrum Publishing, pp 186.

(12) Goldratt, E. M., (1990) The haystack syndrome: sifting information out of the data ocean.  North River Press, pp 146-155.

(13) 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, pg 67.

(14) Smith, D., (unpublished) Gain more speed and predictability through adequate protective capacity.  Chesapeake Consulting Inc., 4 pp.

(15) Schragenheim, E., and Dettmer, H. W., (2000) Manufacturing at warp speed: optimizing supply chain financial performance.  The St. Lucie Press, pg 175-176.

(16) Kawase, T., (2001) Human-centered problem-solving: the management of improvements.  Asian Productivity Organization, pg 111-119.

(17) Umble, M., and Srikanth, M. L, (1996) Synchronous manufacturing: principles for world-class excellence.  Spectrum Publishing, pg 74.

(18) Kanigel, R., (1997) The one best way: Frederick Winslow Taylor and the enigma of efficiency.  Viking, pp 141-142, 163-166, & 203-210.

(19) Goldratt, E. M. (1996) Production the TOC way, Tutor guide.  Avraham Y. Goldratt Institute, pp 46-48.

(20) Goldratt, E. M., (1990) The haystack syndrome: sifting information out of the data ocean.  North River Press, pp 88-89.

(21) Cox, J. F., and Spencer, M. S., (1997) The constraints management handbook.  St Lucie Press, pg 90-91.

(22) Stein, R. E., (1994) The next phase of total quality management: TQM II and the focus on profitability.  Marcel Dekker, pp 99-100.

(23) Woeppel. M. J., (2001) Manufacturer’s guide to implementing the Theory of Constraints.  St. Lucie Press, pp 128-129.

(24) Goldratt, E. M., (1990) The haystack syndrome: sifting information out of the data ocean.  North River Press, pp 130 & 129.

(25) Stein, R. E., (1994) The next phase of total quality management: TQM II and the focus on profitability.  Marcel Dekker, pp 92-99.

(26) Stein, R. E., (1996) Re-engineering the manufacturing system: applying the theory of constraints (TOC).  Marcel Dekker, pp 141-151.

(27) Goldratt, E. M., (1990) The haystack syndrome: sifting information out of the data ocean.  North River Press, pp 139-140.

(28) Goldratt, E. M., (1990) Essays on the Theory of Constraints, Chapter 3.  North River Press.

(29) Caspari, J. A., and Caspari, P., (2004) Management Dynamics: merging constraints accounting to drive improvement.  John Wiley & Sons Inc., pp 187-188.

(30) Newbold, R. C., (1998) Project management in the fast lane: applying the Theory of Constraints.  St. Lucie Press, pp 152-155.

(31) Liker, J. K., (2004) The Toyota Way: 14 management principles from the world’s greatest manufacturer.  McGraw-Hill, pp 223-236.

This Webpage Copyright © 2003-2009 by Dr K. J. Youngman