It’s About Time

The Competitive Advantage of

Quick Response Manufacturing

Rajan Suri
Appendix D
Simple System Dynamics
Calculations for Quick Insights

In this Appendix, I show you how to perform some calculations that will
give you insights into the main drivers of Flow Time in your organization.
This will include insights into capacity planning, batch sizing, and other
decisions related to operating your QRM Cells.
First, I show you some simpler calculations that you can perform easily
to get initial insights without needing to obtain a lot of detailed data. Next,
if you are willing to engage in more data gathering and analysis, I show you
how to get more accurate estimates of the impact of various decisions. After
this I also show you a formula that can work in certain situations to help
you evaluate your batch sizes. Finally, for a more complete analysis of sys-
tem dynamics in your factory, you will need to use some type of dynamic
modeling software and I give you a few suggestions to explore further.
This Appendix continues in more depth the explanations I gave you
in Chapter 3. I will assume you have reviewed that chapter before you
continue with the material here. The basis for all the discussions in this
Appendix will be the formula in Chapter 3 for the average Flow Time of a
job visiting a resource:

Flow Time = (AV × M × TJ) + TJ

As a reminder, in this formula AV is the average variability, which

includes variability in both the arrivals and the job times, and I go into
more detail on this in the “How to Calculate Variability Values” section.
M is the magnifying effect of utilization and equals u ÷ (1 – u) where u is
the utilization of the resource. Finally, TJ is the average time taken by the
resource to work on a job. All these quantities were defined and explained

D3
D4  •  Supplement to the book, It’s About Time, by R. Suri

in Chapter 3. In the rest of this Appendix I show you how to calculate

these quantities for your situation and the types of insights you can get by
using these formulas. Before proceeding with the calculations, you need to
make an important decision.

First Determine What Time Period You Will Use

Since all the formulas here are based on statistical data from your opera-
tion, you need to be sure to use data for a time period that is reasonably
representative of the operation you wish to analyze. On the one hand, the
time period should not be too short, or your data might not be represen-
tative. On the other hand, if the time period is too long, then you might
average out peak-demand months with low-demand months and you don’t
want to do that either because then you lose insights into how to operate in
these two different situations.
Common sense should help you pick a time period for your particular
situation, but in general I have found that anything less than a week is
too short, while a month or a quarter are reasonable periods. If you have
high- and low-demand periods, you should perform separate analyses for
each one—for example, you could do an analysis for a month with peak
demand, a second one for a month with normal demand, and a third anal-
ysis for a month with low demand.
You can use historical data if you feel it is representative of expected
operations in the near future, or you can use planned (estimated
future) data if you feel that operations will be significantly different in
the near future.
Finally, note that the purpose of this Appendix is to give you insights
into the main causes of delays and the key decisions that will help you
reduce them. The approach here is not aimed at accurate day-to-day run-
ning of your factory, but rather, at giving you the major direction you need
to take. For this purpose, rough data is good enough, so don’t spend too
much time in data gathering. Get some initial numbers and get going on
the analyses here! In most cases I have seen that there are some significant
insights obtained right away.
I will start with the calculation of utilization (u) for a resource as that is
always the first step.
 Simple System Dynamics Calculations for Quick Insights  •  D5

Remember, in QRM “Utilization” Includes All

the Time the Resource Is Occupied for Any Task
Chapter 3 included a detailed discussion of this issue. To calculate the
value of u for a given resource, select the period you will analyze. For this
period, add up all the times that this resource was occupied for all tasks
plus any times it was not available for other reasons such as unexpected
downtimes due to machine failures or absenteeism. Let’s call this the Total
Utilized Time. If you are doing this for a past period and have reasonable
historical data in your MRP system, you can add up the times spent by jobs
at this resource. If you feel that the actual data is not accurately recorded,
you can also work with the standards in the MRP system if you feel they
are reasonably representative. In either case, be sure that the data includes
setup times or else you will need to add in estimated setup times. Finally,
be sure to add in any downtimes and other times as just mentioned.
Next, add up all the times that the resource was scheduled to work; call
this Total Scheduled Time. This can usually be done just by checking the
calendar and knowing how many shifts the resource operates. However,
be sure to check past records and add in all instances of overtime that was
used beyond the scheduled times that you included. With these calcula-
tions in place, the utilization of this resource is simply:

Total Utilized Time
u=
Total Scheduled time

As a check on whether you have included all the potential instances in

the Total Utilized Time, I’d like to remind you of the alternative approach
in Chapter 3, which helps to “back into” the value of utilization. If you
are not sure whether a given category of time should be included in Total
Utilized Time, then ask this question: Was this resource actually avail-
able to start an unexpected job if it had arrived at the resource during
this time? If the answer is no, then this time should be included in Total
Utilized Time. I’ll give you two examples:

• A machine breaks down during in the middle of a shift and is down

for the rest of the shift (4 hours) and still down for the first 2 hours of
the next shift. Then you definitely need to add 6 hours into the Total
Utilized Time value.
D6  •  Supplement to the book, It’s About Time, by R. Suri

• A machine is available but the operator unexpectedly has to go home

and is gone for 5 hours during the shift and no one else is available to
run the machine. Again, you need to add these 5 hours to the Total
Utilized Time.

If you are concerned that your data isn’t good enough to record such
events, don’t worry, just talk to managers or supervisors and get rough
numbers for such instances.
Now I’ll give a different example. A machine needs preventive main-
tenance on its bearings and this is scheduled to be done on Thursday
and Friday. This is known ahead of time, and the Planning Department
does not include these 2 days when it is planning capacity and schedul-
ing the machine. In this case, the 2 days need to be taken out of the Total
Scheduled Time (or not included in the total in the first place).
In general, the rule is simple: If the time was scheduled and planned for,
and then it becomes unavailable, it needs to go into the Total Utilized Time
category. If the time was not scheduled to begin with (in the planning system),
then just be sure not to include it in the Total Scheduled Time. You might
think, if a machine fails and is unavailable for 10 hours, why not just take
these 10 hours out of the Total Scheduled Time? Why are they included in the
Total Utilized Time? The answer has to do with the system dynamics effect of
planned versus unplanned capacity utilization, as this example shows.
Suppose the Total Scheduled Time for a machine is 40 hours a week, but
during one week the machine unexpectedly breaks down and is not avail-
able for 10 hours of the working time. Also, say that 20 hours of work are
scheduled on this machine for that week. If you take the breakdown time
out of the scheduled time, then you get a utilization of u = 20/30 or 67% for
that week. But if you add the breakdown time to the Total Utilized Time
and leave the Total Scheduled Time at 40 hours, then you get u = 30 ÷ 40
or 75% for that week! This difference is even more pronounced when you
calculate its impact on Queue Time through the value of M as explained
in Chapter 3. In the first calculation you get M = 2, while in the second
case you get M = 3, a 50% increase in Queue Time! So this calculation pro-
cedure is important. But which one is right? The answer is the second one
(with the higher values of u and M). The reason is the system dynamics
effects occur based on what you plan, and then how the real world evolves.
If you originally expected the time to be available and planned for that,
and now the time is not available, this creates dysfunctional effects that
are estimated by the system dynamics formula.
 Simple System Dynamics Calculations for Quick Insights  •  D7

Get Initial Insights with a Quick First Analysis

Now that you understand how to calculate utilization, get rough estimates
of its value for key resources. You can do this for most of your resources
if your data gathering can be automated (e.g., using your MRP system),
or else pick some resources that you know are bottlenecks and causing
long MCT. For example, there may be some cells that still have large MCT
values and you can start your analysis with the resources in these cells—
or if you suspect that there are a few specific resources within the cells
that are problematic, you can restrict your first analysis to those resources
and then expand to other resources later. Note that “resource” can be a
machine or a person such as an assembly operator. It is important to try
this approach on a few resources first as you will also discover how to
gather and use your data and whether it is good enough or if more effort is
needed in getting accurate data.
I’ll use an example to show you how to conduct this first analysis. Let’s
say you suspect that in a particular QRM Cell, a large milling machine
is a bottleneck and has a long Flow Time, thus contributing to the large
MCT value for the cell. You need to calculate only two data values for this
machine (I’ll call it the Large Mill) to get initial insights. First, calculate
the utilization (u) as previously described. Next, for the time period you
are analyzing, also calculate TJ, the average time that it takes for a job on
this machine (this includes setup time plus the run time for all the pieces
in the job).
Now to make this first analysis simple and to get quick insights, you
will use the value 1.0 for both the arrival variability and the job variabil-
ity. I will explain both of these later in this Appendix, but here is a short
explanation for now. For arrival variability, the value of 1.0 is commonly
used in system dynamics models to represent random arrivals such as cus-
tomer orders when there isn’t better data available, and is accepted as a
good starting point for analysis. For job variability, the value of 1.0 also is
commonly used to denote a reasonable amount of variability (e.g., some
very short jobs, some long ones)—it represents a statistical compromise
between very high variability and very little variability. Next, as explained
in Chapter 3, the average variability is obtained by squaring these two
variability values, adding the squared values and dividing by two. Since
both these values are 1.0, when you do this calculation the average vari-
ability AV just comes out at 1.0 also.
D8  •  Supplement to the book, It’s About Time, by R. Suri

When you put these values into the formula for Flow Time it simplifies to

TJ
Flow Time =
1−u

Now let’s see what this formula tells us. Suppose for the Large Mill you
find that TJ = 6 hours and u = 0.88. Then the formula predicts a Flow
Time of 50 hours. So you see that while on average there are only 6 hours
of work to be done on a job, it takes 50 hours to get through the Large
Mill. You can do similar analyses for other resources that appear to be
problematic. The results of these analyses will enable you to target a few
resources that need improvement. You can start working on two aspects
of these resources right away—namely, reducing the values of TJ and u
by applying some of the improvement ideas discussed in Chapter 3. For
the Large Mill as an example, suppose you think you can reduce the aver-
age job time by 5% to 5.7 hours and this reduction, along with improved
maintenance and uptime, would bring the utilization down to 81%, what
would that do to the Flow Time? The formula predicts a new Flow Time of
30 hours—a 40% reduction—which may help you to get the investments
needed to make these improvements. After you have targeted a few such
opportunities for improvement, you can then proceed to conduct more
detailed analyses as explained next, in order to pinpoint additional oppor-
tunities for improvement.

How to Calculate Variability Values

In order to get to the next level of accuracy with your system dynamics
analysis, you need to estimate the arrival time variability and the job time
variability for the resource being analyzed. I’ll start with arrival time vari-
ability. To calculate this you need data on the times between job arrivals.
If a job arrives at the resource at 9:00 a.m. and then the next job arrives at
2:00 p.m. then the time between the first and second arrival is 5 hours. It is
important that this data is calculated by using only working hours. Let’s say
the resource works one shift, from 8:00 a.m. to 4:00 p.m. Continuing with
the same example where the last job arrived at 2:00 p.m., let’s say the next
job arrives at 9:00 a.m. the next day. Then the time between this new arrival
 Simple System Dynamics Calculations for Quick Insights  •  D9

and the previous one should be recorded as only 3 hours (2 hours of work
time on the previous day and 1 hour of work time on this day).
If your MRP system has such data on job arrival times, you can auto-
mate this calculation, as long as you can remove nonwork times including
nights and weekends. It’s possible that you don’t have this detailed data for
the resource in question, or it is not accurate enough, or it is hard to extract
the values I described. In that case you have a few alternative options:

• You can look at data for initial job releases in the MRP system, that
is, the times when jobs are released to the shop floor. Be sure to look
only at jobs that are destined for this resource, and then use the times
between these job releases. Essentially, you are assuming that as vari-
ous jobs work their way through the shop floor, the arrivals to this
resource have a similar pattern. If a bunch of jobs destined for this
resource are all released at the same time, you give the first job in this
bunch the interarrival time value of the actual time from the previous
job release, and all the other jobs get an interarrival time of zero.
• If jobs on the shop floor are closely related to customer orders, then
you can look at arrival patterns for customer orders for the relevant
jobs and use those in the same way as previously described.
• Use manual observations for a period of time, such as a week. Ask a cell
team member or a planner or scheduler for that area to record data.
• As a last resort, if you are really having difficulty getting good enough
data, you can again resort to using the value 1.0 for the arrival vari-
ability. As I explained earlier, this is a commonly accepted start-
ing point for system dynamics analysis. This will allow you to get
going on some of the other calculations in the following paragraphs.
However, this is a stopgap measure so that you can do some initial
number crunching. Do not accept this as an end solution. Start the
process of gathering data so you can get a more accurate number in
the near future.

Once you have all the data you need, list these values of interarrival
times in a spreadsheet. (In the following section, I suggest additional cal-
culations that need a spreadsheet, so you can keep in mind that you could
integrate those into this spreadsheet.) Use the spreadsheet functions to
calculate the average of all the interarrival times:; say this value is TA (for
average Time between Arrivals). Next also use the spreadsheet functions
to calculate the standard deviation of all these values; say this is SA (for
D10  •  Supplement to the book, It’s About Time, by R. Suri

Standard deviation of time between Arrivals). Then the final value that
you need, namely the variability of arrivals (I’ll call it VA) for the resource
is given simply by

SA
VA =
TA

Next I’ll show you how to calculate the variability in job times. Here
you need data on how long it takes the resource to work on each job.
Remember that this must include the setup time plus the time to complete
all the pieces in the job. In order to do some of the “what-ifs” that I will
discuss in the following section, it would be best if you would record sepa-
rately, for each job, the setup time, the batch size used, and the run time.
The run time can be either the total for the whole batch, or the average
time per piece—either is fine because if you know the batch size you can
calculate either value from the other. For example, if you use actual data,
you might have only the total run time recorded. On the other hand, if you
use standards from the MRP system, you will probably have the run time
per piece. As in the previous case, you should organize all these values in
a spreadsheet as shown in Table D.1.
The last column in the table is the total time to complete each job.
From this column you can use the spreadsheet functions to calculate
the average time to work on a job (I’ve already called this TJ) and the
standard deviation of the time to complete a job (I will call this SJ).
Then the variability of job times (I’ll call it VJ) for the resource is given
simply by

SJ
VJ =
TJ

Table D.1
Example of Spreadsheet Needed to Gather Job Data for System Dynamics Analysis
List of Jobs for NC Mill in Housings Cell during December
Run Time per Total Job Time
Job Number Quantity Setup Time (hrs) piece (hrs) (hrs)
13011928 16 1.25 0.33 6.53
18121952 13 0.67 0.20 3.27
52133910 4 0.83 1.12 5.31
26522007 18 1.45 0.27 6.31
 Simple System Dynamics Calculations for Quick Insights  •  D11

Finally, you now have the data to calculate the average variability (AV),
which is needed in the Flow Time formula. This is given by

AV = (VA2 + VJ 2) ÷ 2

Note that if you set up a spreadsheet to calculate TJ and VJ, you can use
the same data to get the utilization u as well. Just remember to add any
downtimes or other times as I explained earlier.

How to Use the Detailed Flow Time

Formula to Get Insights
Let’s say you want insight into a QRM Cell that was put in with a target
MCT of 3 days but currently has an MCT value of over 6 days. The cell
team feels that a particular vertical turret lathe (VTL) in the cell is causing
long Flow Times and needs your help in planning the load on this VTL.
You’ve done some initial simplified analysis as shown previously but now
you want to do some more detailed analysis to see what other improve-
ments might be possible. Your analysis of the data for this VTL produces
these values: u = 0.87, TJ = 4 hours, and VJ = 0.73; you also realize that
you don’t have an easy way to get interarrival time data, so as explained
previously you assume VA = 1 for this analysis. Now you can calculate AV
using the preceding formula and you get AV = (12 + 0.732)/2 = 0.766. You
can also calculate M, which equals u/(1 – u) = 6.69. For convenience, the
Flow Time formula is:

Flow Time = (AV × M × TJ) + TJ

Putting all these values into the formula, you find that the average Flow
Time through this VTL is 24.5 hours. If the cell works one shift a day (8
hours), this is a Flow Time of over 3 working days, so this one machine
alone accounts for a Flow Time that exceeds the target for the whole cell.
Note that the average job time on the VTL is only 4 hours in this example,
so it is taking 3 days to do a 4-hour job—the cell team is right in its gut
feeling that the main problem is the VTL. Even before I discuss what action
you could take, note that just doing this calculation helps you rationally
ascertain whether there is a problem at the resource rather than having to
go on people’s opinions.
D12  •  Supplement to the book, It’s About Time, by R. Suri

Now that you know there is an issue with the VTL, you can look at poten-
tial solutions. In Chapter 3, I gave you a three-pronged strategy to reduce
Flow Time along with several practical suggestions. Let’s use a couple of
these ideas here. You find that some parts are being made to stock in large
quantities and then some of those pieces don’t get sold—with your shorter
lead times you need less safety margin and you can run roughly the same
number of batches but with smaller quantities; there will be the same num-
ber of setups but less run time so this will reduce some load on the VTL. You
also get management to agree to spend money on 4 hours of overtime per
week for the cell (essentially a 10% increase over the 40-hour week). Finally,
you find that the value of VJ is high because of occasional long breakdowns
of the VTL. You discuss this with the maintenance crew and come up with
a preventive maintenance strategy. You also get estimates for what the new
downtime data might be in the future. You put this revised data into your
spreadsheet and get a new value for VJ. With all these improvements, you
estimate that the new values will be: u = 0.7, TJ = 3.9 and VJ = 0.5, and VA
remains the same for now. Putting these estimates into the formula gives
you the new Flow Time of 9.6 hours, or just over one shift. So the next
insight you get from the formula is that these improvements can help you
cut almost 2 days out of the Flow Time through this cell.
There is one more important decision that you would normally explore
and that is to do with the batch sizes being run on the VTL. Here I will
show you one way of analyzing this issue, and then I will give you an alter-
native approach later. Let’s say that in looking at the historical data that
you used to get the TJ and VJ values (the data in your spreadsheet similar
to that in Table D.1), you note that there are occasionally very large total
job times; you find that these are usually associated with jobs being run in
large batch sizes. So you can investigate the impact of cutting these down
as the following examples show. If a part is being made to stock in batches
of 20, you could consider running it in batches of 10. This would mean that
during the time period you analyzed you would need to run two orders of
10. So in your spreadsheet you replace one order of 20 with two orders of
10. Remembering that each of these will take a setup time, you can’t just
cut the job times in half, so make sure to calculate the new total of setup
and run times and enter the right total times for each of these two jobs. If
cutting a batch size in half seems too drastic, you can be more conserva-
tive. Say that during the period you had two orders of the same part with
30 pieces in each order for a total of 60 pieces. You could consider cutting
the batch size to 20 and going with three orders during the period.
 Simple System Dynamics Calculations for Quick Insights  •  D13

In this fashion, you can pick out some of the batches with the longest
job times and revise the data for them. With modern spreadsheets you
can automate this calculation to make your job easier. After you revise the
data, three values will change: the utilization, u, will become larger due to
more setups, TJ will become smaller as some of the job times will be much
smaller than before, and SJ will become a bit smaller too since the very
long jobs will be cut down so there will be less variability overall.
Let’s say after you revise the VTL data with your “what-ifs” for the
batch sizes, these new numbers are: u = 0.72, TJ = 3.1, and VJ = 0.45.
Putting these estimates into the formula gives you the new Flow Time of
7.9 hours, so now the VTL Flow Time is reduced to under a shift. While
this is not a big reduction from the previous value of 9.6 hours, it does
set the stage for the next improvement. If long runs are an issue on this
machine, clearly one item to explore at this point is whether setup times
can also be reduced, enabling even further reduction in batch sizes. So
you look at the impact of a 50% reduction in setup time for some of the
parts with big batch sizes, and also reduce the batch sizes some more.
You play around with some numbers and find batch sizes that will keep
the utilization from going over its last value. With this new data in your
spreadsheet, you get u = 0.72, TJ = 2.7, and VJ = 0.4, and the formula
predicts a Flow Time of 6.7 hours. Again, this may seem like a small
improvement over the previous value, but remember where you started—
at 24.5 hours! The system dynamics analysis here has helped you come
up with a series of improvements that will reduce the Flow Time through
the VTL by 75%.
If this data manipulation seems too time consuming, note that using
spreadsheets and with the availability of query methods on modern ERP
systems, one of your systems analysts could help you automate much of
this “what-if” analysis. Alternatively, you could look into using one of the
modeling packages mentioned at the end of this Appendix.

Determining a Good Batch Size

While the previous analysis will help you look at any resource, performing
the “what-ifs” can be time consuming. If you feel that poor choice of batch
sizes may be affecting the performance at a particular resource, there is
a simpler rule that you can use for quick insights in certain cases. The
D14  •  Supplement to the book, It’s About Time, by R. Suri

resource must satisfy the condition that for all products going through
this resource, the setup times are about the same across products, and
similarly, the run times per piece are reasonably similar across products.
In that case you can use the averages of these setup times and run times
to come up with a good batch size. Specifically, the rule I give you will tell
you the batch size that should result in the shortest Flow Time through
that resource.
To apply the rule you will need to calculate some more quantities using
the data you already have in the spreadsheet discussed previously. I will
use a time unit of hours for illustration; you can pick any other time unit
as well. The quantities you need to calculate are:
• S = Average setup time for a job, in hours
• H = Total Scheduled Time at the resource for the selected time
period, in hours
• Q = Total number of pieces (of all products) made by the resource
during the same time period
• R = Total of all run times at the resource during the period
• Z = Total of all “other times” at the resource during the period (see
following explanation)
I have used the phrase “other times.” These are the times that the resource
is not in setup nor is it running a product, but it is still not available to run
another job for some reason (such as when a machine is broken down). In
Chapter 3 and at the beginning of this Appendix, I discuss the importance
of including the total of such times in your utilization calculation.
From these values, you calculate uR , which is the “run utilization” of the
resource (proportion of time it was actually working on a product), and
uZ , which is its “other utilization” (proportion of time it spent in these
“other times”):

R
uR =
H


Z
uZ =
H

With all these quantities in hand, the recommended batch size for this
resource can be calculated by the formula
 Simple System Dynamics Calculations for Quick Insights  •  D15

S × Q × (UR +√ UR × (1 − UZ) )
B* =
H × UR × (1 − UZ − UR)

In case you are interested in the background to this formula, it is based
on system dynamics analysis of batch sizing. As I explain in Chapter 3
(and illustrate there in Figure 3.6), there is a trade-off between the large
batch sizes that cause long job times and the small batch sizes that create
a lot of setups. Looking at Figure 3.6, B* is the batch size for which you get
the shortest Flow Time through the resource. For more details on the for-
mula derivation, see “For Further Reading” at the end of this Appendix.
I’ll walk you through an example to show you the application of this
formula. Let’s say that in a QRM Cell a particular CNC lathe runs jobs
with typically 50 pieces at a time and you suspect that this batch size is
too large and is causing long Flow Time. An initial look at the data for
the lathe shows you that most jobs have similar run times for each piece,
and discussions with the cell team lead you to believe that setup times
are about the same across jobs. So you feel you can apply this formula
to get some insight. For this lathe, you get the following data for a one-
month period.

• S = Average setup time = 1 hour

• H = Total Scheduled Time = 200 hours
• Q = Total number of pieces made = 520
• R = Total of all run times = 130 hours
• Z = Total of all “other times” = 10 hours

These values result in uR = 130/200 = 0.65, uZ = 10/200 = 0.05. Putting all

these values into the formula (you can use a spreadsheet or a calculator for
this) you get B* = 19.1. Even after rounding off to the next whole number,
the recommended batch size is around 20 pieces, which is less than half
the current batch size. Although this is an approximate analysis based on
rough data, this is a significant enough difference from your current oper-
ating policy that it should convince you to start running smaller batches
on the lathe. Even if you don’t want to do anything as drastic as cutting
all the batches in half, you could perhaps reduce them by 25%, observe the
effects building up over the next month, and then reduce the batches some
more after the Flow Time goes down as predicted.
As a final general pointer on batch sizing, remember that even a few big
batches (big in the sense that their total setup plus run time is long relative
D16  •  Supplement to the book, It’s About Time, by R. Suri

to other products) can have a disproportionately large effect on overall

Flow Times, as I showed with the numerical example in Chapter 3. Hence
always keep your eye out for such situations and find ways to justify reduc-
ing these big batches using various tools discussed in this book.

Consider Using Dynamic Modeling Software

The calculations here will enable you to get started and use system dynam-
ics to get some initial insights. However, the analysis gets complicated
once you try to include more details of your factory such as cross-trained
people running multiple machines, products flowing through multistep
routings, different products with different setup and run times on various
machines, scrap and rework, and so on. If, after trying out some of these
formulas, you see the potential of system dynamics and want to work with
a more realistic model of your factory, you will need to use some type of
dynamic modeling software (see “For Further Reading”).

For Further Reading

If you would like more insight into the system dynamics formulas and
why utilization and variability impact Flow Time, but don’t want too tech-
nical an explanation, an easy-to-read discussion along with step-by-step
examples can be found in Chapter 7 of Quick Response Manufacturing: A
Companywide Approach to Reducing Lead Times, by R. Suri (Productivity
Press, 1998).
If you have a more theoretical background, the batch size formula is
based on an analysis in “Lot Sizes, Lead Times and In-process Inventories,”
by U. S. Karmarkar, Management Science, March 1987, pages 409–418.
However, note that I have modified the analysis to include the “other” uti-
lization. I did this to provide more accurate predictions in the presence
of significant downtimes, which often occur in real-world situations and
need to be taken into consideration.
An easy-to-use software package for system dynamics analysis of fac-
tories is the MPX Rapid Modeling System from Network Dynamics, Inc.
(www.networkdyn.com). It can model all the practical aspects of factories
 Simple System Dynamics Calculations for Quick Insights  •  D17

as mentioned in the text and at the same time takes just seconds to calcu-
late the output values, even for large systems.
There are more detailed modeling packages available for system dynamics
analysis, and they are known under the general category of “discrete-event
simulation” software. Several professional societies publish annual surveys
of these packages and also hold regular tutorials and conferences on them. A
few suggestions include the Institute of Industrial Engineers (www.iienet.org),
INFORMS (www.informs.org), the Society for Computer Simulation (www.
scs.org), and the Winter Simulation Conference (www.wintersim.org).