Critical Chain Project Management

Improves Project Performance

Larry P. Leach, Quality Systems, 1577 Del Mar Circle, Idaho Falls, Idaho 83404 USA

I Abstract
This paper describes the theory and practice of critical chain project management (CCPM). CCPM provides a substantial
step in the ongoing improvement to the Project Management Body of Knowledge. The critical chain differs from the critical
path by including resource dependencies and never changing. CCPM improves the project plan by ensuring that it is feasible
and immune from reasonable common cause variation (uncertainty or statistical fluctuations). It does this by aggregating
uncertainty into buffers at the end of activity paths (or chains). The project buffer protects the overall project completion on
the critical chain path, and feeding buffers protect the critical chain from path merging. Buffer management enhances mea-
surement and decision-making for project control. CCPM implements required changes in resource behavior, including elim-
ination of date-driven activity performance and multitasking. Most of all, CCPM improves the focus of the project manager
and performers. Projects that use CCPM have a greatly improved record of schedule, cost, and scope performance. Experi-
ence with CCPM projects demonstrates completion in less than one half the time of projects using previous planning and con-
trol methods.

Keywords: critical chain project management; project schedule; project buffers

©1999 by the Project Management Institute — 1999, Vol. 30, No. 2, 39–51 — 8756–9728/99/$5.00 per article + $0.50 per page

C
ritical chain project management (CCPM) adds to scheduled dates and milestones within a project plan. It
the present Project Management Body of Knowl- focuses on developing and managing project perfor-
edge (PMBOK™), as described in A Guide to the mance to meet or exceed reduced activity times, thereby
Project Management Body of Knowledge (PMBOK™ Guide) reducing overall project duration.
(Project Management Institute Standards Committee, CCPM departs from the PMBOK™ Guide and sup-
1996) and supporting literature, in the areas of project porting literature as follows. CCPM:
planning, activity performance, and control. It reengineers I Specifies the critical chain, rather than the critical path,
project planning and management to eliminate common as the project constraint. This path includes resource
problems that lead to poor project performance. dependencies, and does not change during project
CCPM emphasizes focusing on the project execution.
schedule. Further, it also reduces project changes and
I Uses 50% probable activity times, and aggregates
the major source of project cost overruns by improving
allowances for uncertainty of estimates and activity per-
schedule performance. It accomplishes these results by
formance into “buffers” at the end of activity chains.
changing the project plan, the project measurement and
I Uses the buffers as an immediate and direct mea-
control system, and certain behaviors by the project
team and supporting personnel. The critical chain plan surement tool to control project schedule.
I Defines the constraint for multiple projects as the
effectively eliminates most resource contention before
the project starts, and uses buffers for project control. constraining company resource. It links projects through
System thinking and the Theory of Constraints this resource, using buffers to account for activity dura-
(TOC) were used to develop CCPM. The CCPM project tion variability.
planning and control process directly addresses uncer- I Seeks to change project team behavior; encouraging
tainty and variation in project activity duration. It helps reporting early completion of activities and elimination
eliminate undesirable behaviors fostered by using of multitasking.

June 1999 Project Management Journal 39

Applications Demonstrate Effectiveness would increase without bound, or go to zero.” Most
Real-world applications of CPPM demonstrate effective- people readily accept this statement as self-evident fact.
ness over a wide range of project types. Large and small The primary message of The Goal is focus. Focus on
companies demonstrate similar improvements for project the goal of the company. Focus on the constraint that
types, including construction, software development, and blocks achieving the goal of the company. The Goal
high-technology research and development. Companies ends with five focusing steps, which apply to any phys-
such as Texas Instruments, Lucent Technologies, Honey- ical system. These steps are:
well, and Harris Semiconductor complete projects in one 1. Identify the system constraint.
half or less the time of previous or concurrent similar 2. Exploit the system constraint.
projects, or as compared to industry benchmarks. 3. Subordinate everything else to the system constraint.
Harris Semiconductor manufactures semicon- 4. Elevate the system constraint, and
ductor wafers. Harris used CCPM to build a new 8” 5. If, in the previous step, a new constraint has
wafer plant. The total investment for a plant of this size been uncovered, repeat the process. Do not let inertia
is in the range of $250 million. The revenue for such a become the system constraint.
plant is in the range of $2 million per day! (Raw mate- Dr. Goldratt used the focusing steps to develop crit-
rial cost is very small.) The industry standard to build a ical chain project management (CCPM), and CCPM
6” wafer plant is 30 months. The industry standard to managers use them to guide projects. Dr. Goldratt’s TOC
get the plant up and running to 90% of capacity is analysis identifies the core problem leading to most
about 46 months. Harris completed the plant, and project failure as, “failure to effectively manage uncer-
brought it up to 90% production in 14 months. Harris tainty.” The core problem leads to six undesired effects.
presented their results at a recent conference hosted by The CCPM process uses TOC and the following theory
the Avraham Y. Goldratt Institute. A video is available to eliminate the causes of these effects.
through their Internet page. Theory 2: Common Cause Variation. Dr. W.
The Israeli Aircraft Industry employs about 15,000 Edwards Deming (1989) included “an understanding of
people. One of their major functions is to maintain variation” as one of his four points of profound knowl-
jumbo jets used in passenger service. A particular type of edge. He identified two types of variation: (1) Common
maintenance called “type D” normally takes 46 days in Cause Variation: A cause that is inherent in the system. The
the industry. The penalty for nonperformance to responsibility of management. (2) Special Cause Variation:
schedule is very steep … $60,000 per day, because the A cause that is specific to some group of workers, or to a
airlines need the planes back into scheduled service. The particular production worker, or to a specific machine, or
company had been paying up to $25 million per year in to a specific local condition.
penalties. A letter from the manager to Dr. Goldratt Projects have common cause variation in the per-
(included on www.Goldratt.com) notes, “… we suc- formance time of activities. This variation represents
ceeded to drop our average turnaround time per aircraft uncertainty in the activity performance time. Although
visit from three months to two weeks and to increase the time to perform individual project activities may be
our backlog from two months to one year.”
Lucent Technologies applies CCPM to increasing
numbers of R&D projects. Many of the projects are still in
operation; but general feedback demonstrates plan I About the Author
reductions in excess of 25%, and performance to date
using very little of the project buffers. Early application to Larry P. Leach is the principal of Quality Systems, a
manufacturing consulting firm that
a software development project that was thought to be specializes in leading the implementation
impossible completed in June 1997, with buffer to spare. of the new critical chain method of
project management. He has an M.S. in
business management from the
Critical Chain Theory University of Idaho and mechanical engi-
CCPM uses three theory tools to improve project per- neering from the University of
Connecticut. He is a member of PMI and the American
formance. It applies the theory to eliminate six specific
Society for Quality Control. He has published many
project effects that lead to project schedule overruns. papers on related topics, and authored a self-published
(TOC identifies the six items as effects, rather than book, The Critical Chain Fieldbook. Prior to founding
causes, because they have underlying causes.) Quality Systems, he worked at VP-level in several Fortune
Theory 1: Theory of Constraints. CCPM applies 500 companies, and was a systems analysis division
director for the U.S. Department of Energy. His experi-
the TOC to project management. Goldratt first ence as a project manager includes research and devel-
described TOC in The Goal (1984) when applied it to opment projects for construction projects.
production systems. TOC can be summarized by: “Any
system must have a constraint. Otherwise, its output

40 Project Management Journal June 1999

 0.02 1.00
0.90
0.016 0.80
0.70
P (x)
0.012 Cum Prob 0.60
0.50
0.008 0.40
0.30
0.004 0.20
0.10
0 0.00
0.00

20.00

40.00

60.00

80.00
Figure 1. Typical Project Activity Performance Time Probability Distributions

independent of each other, project activity networks Theory 3: Statistical Laws Governing Common
define activity dependence. The project logic demands Cause Variation. The PMBOK™ Guide (Table 11-2)
that successor activities cannot start until the predecessor describes the well-known statistical law of aggregation:
activities complete (for the most frequent finish-to-start “The project variance is the sum of the individual
activity connection). Thus, projects have statistical fluc- activity variances.” Note that in statistical terminology,
tuations and dependent events, the same key issues that variance is the square of the standard deviation, usually
Dr. Goldratt (1984) addresses for production. represented by “s2” or the Greek sigma squared. For a
Dr. Goldratt’s improvements for production take given statistical distribution, it requires a given number
advantage of (exploit) the reality of common cause vari- of standard deviations to provide a cumulative proba-
ation. Figure 1 illustrates a typical project activity perfor- bility to that point. For example, with a normal distri-
mance time distribution. The solid curve (left ordinate) bution, one standard deviation plus or minus repre-
shows the probability of a given time on the abscissa. sents 67% of the data, or a cumulative probability that
The dotted line shows the cumulative probability of 67% of the time a result will fall within one standard
completing the activity in a time less than or equal to the deviation of the mean, plus or minus.
time on the abscissa. Note the left skew of the distribu- The statistical method to combine variances means
tion and the long tail to the right; this is typical of the that we can protect a chain of activities to the same level
common cause variation for many project activities. of probability with much less total contingency time
This common cause variation in activity perfor- than we can protect each individual activity. Aggregation
mance is not an exceptional event, such as discrete of the contingency times dramatically reduces the
project risk events. PERT attempted to estimate the overall estimated time for a chain of activities. Consider
impact of this common cause variation using three a chain of four activities, each of which has a 50% prob-
activity duration estimates, but for a variety of reasons ability estimated duration of one time unit, and a 90%
did not succeed. The PMBOK™ Guide and other litera- probable estimated activity duration of two time units.
ture still mention PERT in this fashion, although from If we include the contingency in each activity, the chain
my experience it is little used today. The PMBOK™ Guide of activities is eight units long. If we use the law of aggre-
acknowledges this. PERT diagrams, as referred to in gation, we can protect the whole chain to 90% proba-
much of the project literature, and in many project soft- bility by scheduling the individual activities at their 50%
ware packages, are simply a way to show the project net- estimates (a total of four units), and adding a two-unit
work logic independent of the time scale; not an appli- buffer at the end of the chain, for a total of six units.
cation of the three time estimates. Some projects use A second factor that comes into play in aggregating
methods such as simulation and Monte Carlo analysis activities is the central limit theorem. The central limit the-
to assess the impact of activity duration and cost uncer- orem states “as sample size increases, the distribution of the
tainty. While these methods propose a way to estimate sample mean becomes closer to the normal distribution”
uncertainty, they do not pose in my view an effective (Moore & McCabe, 1993). Many project activities have a
systematic method to manage it. skewed probability distribution. Figure 1 illustrates this

June 1999 Project Management Journal 41

general shape. The distributions have an absolute min- little positive variation. If estimates are 50% probable,
imum time, and a long tail to the right, meaning that they people should complete and report 50% of the activi-
can take much longer than the average time. The central ties early. If estimates are 99% probable, 99% of the
limit theorem means that a project chain of activities has activities should report as completed early. Usually,
a more symmetrical distribution. This is true whether we people report most of the activities as done on the
know the real distributions or not. scheduled date, and they report a significant fraction
Undesired Effect 1: Excessive Activity Duration (~10%) of the activities as late.
Estimates. Most project managers include contingency Goldratt (1997) described several effects that led to
time within each activity estimate to account for indi- performance systematically overrunning estimates,
vidual activity common cause variation. Contingency is although the estimates initially had extensive contin-
defined as the difference between the 95% probable esti- gency time. Many people have a tendency to wait until
mate and the 50% probable estimate. The existence or activities get really urgent before they work on them.
the amount of this contingency time is not usually spec- This is especially true for busy people in high demand:
ified. People estimating activity times for a project usu- that is, all of the most important people the project
ally believe that the project manager wants low-risk manager is counting on to get the critical path work
activity times; perhaps a probability of 80% to 95% com- done on time. If people believe they have some extra
pletion on, or less than, the activity duration estimate. time in their estimates, they are often willing to accept
Most project plans that I have examined fail to other “higher priority” work at the beginning of the
specify the probability and confidence expected for scheduled activity duration. This tends to waste their
activity duration estimates. Most fail to provide a quanti- contingency time before they start the activity, forcing
tative basis for the activity duration estimate. The them to perform most of the work in the later portion
PMBOK™ Guide states, “Activity duration estimates should of the scheduled activity time. Then, if problems occur,
always include some indications of the range of possible there is no time to recover.
results,” and refers to risk management as the control Figure 2 shows the typical work pattern of many
technique. However, risk management in the project people. Goldratt calls this the student syndrome. They
management literature generally fails to differentiate do less than a third of the work on an activity during
between common cause and special cause variation. the first two-thirds of the activity duration, and the final
Construction projects are somewhat exceptional, two-thirds during the last third of the activity duration.
having access to extensive quantitative data to estimate They are more likely to find they have a problem to
activity duration. For example, the 1997 National Con- complete the activity in the remaining time during the
struction Estimator (Kiley, 1996) uses an extensive data- last third of the scheduled activity time. If they are
base, which lists many potential contributors to working above 100% capacity already to complete two-
common cause uncertainty in the estimates. It states that thirds of the work in one-third of the time, it is unlikely
many of these uncertainty items have ranges of several they can keep to the activity duration. They have little
tens of percents of the cost estimate. Therefore, in many chance to recover from an unanticipated problem, such
cases, they have the same potential impact on schedule. as computer failures. This makes it feel like the activity
Figure 1 illustrates that the “low-risk” duration esti- was underestimated to begin with.
mate can be two or more times the 50% probable esti- Meredith and Mantel (1995) state, “… operation of
mate. In most project environments, people feel good if Parkinson’s Law … clear and present danger. The work
they complete an activity by the due date, and feel bad done on project elements is almost certain to ‘expand to
if they overrun the due date. This reinforces their fill the additional time.’” In Goldratt’s words, “the safety
attempts to estimate high probability completion times. time is wasted.”
Walter A. Shewhart (1986), mentor to W. Edwards Failure to provide resources when needed impacts
activities by extending the duration from the time the
Deming, wrote:
input is available until the activity can complete. Pro-
“It should be noted that the statistician does not viding fewer resources than planned also extends the
attempt to make any verifiable prediction about one activity duration.
single estimate; instead, he states his prediction in Undesired Effect 3: Failure to Pass on Positive
terms of what is going to happen in a whole sequence
Variation. Projects do not get the benefit of many
of estimates made under conditions specified in the
actual early activity completions. Even if completed
operational meaning of the estimate that he chose.”
“early,” performing resources often fail to pass on posi-
Thus, attempts to deal with uncertainty by including tive variations. In most cultures, there is little or no
contingency in individual activity estimates are fruitless, reward for completing individual activities early, and
and significantly extend project plan duration. punishment for being late or having quality problems.
Undesired Effect 2: Little Actual Activity Posi- In many project environments, there is a significant dis-
tive Variation. Activity performance behavior leads to incentive to reporting an activity complete early. Work

42 Project Management Journal June 1999

 Milestone
Date

Effort

Student Syndrome

Activity Time

Figure 2. Typical Work Pattern

performed on “time and material” contracts results in Usually, the path merges tend to concentrate near the
less revenue if the work is completed early. Many com- end of the project. One reason for this is that “assembly”
panies budget work performed by internal functional or “test” operations tend to occur near the end of the
organizations as if it were time and material contract project, requiring many elements to come together. The
work. If the functional organization completes the work following discussion demonstrates how this becomes a
in less time than estimated, they cannot continue to primary cause of the well-known project “truth” that,
charge to the project. They must find alternative work “many projects complete 90% in the first year, and com-
for the resources. If individuals complete activities early, plete the final 10% in the second year.”
they get more to do. These cultures drive local optima, Activity path merging creates a filter that eliminates
which means delivery on the scheduled date, but not positive fluctuations, and passes on the longest delay.
before. There are many ways to justify keeping the The reason is that merging activity paths means that all
potentially early result. We can put its review or com- of the feeding paths are required to start the successor
pletion at “low priority,” because it is not due yet, or we activity. Therefore, the successor activity cannot start
can “polish the apple.” The result is the same: we waste until the latest of the merging activities completes.
contingency time originally included in individual Consider an activity on the project critical path that
activity time estimates. requires three separate inputs in order to start (see Figure
Examination of milestone performance in a very 3). This is frequent in assembly operations, and in many
large project (over 15,000 activities) conforms very project results, such as a major show or meeting event
closely to Goldratt’s prediction that about 80% of the where everything has to be ready on opening day. Usu-
activity milestones are achieved exactly on the working ally, there are many more than three. However, even with
scheduled date (i.e., exactly on the original planned three, if each has a 50% chance of being done in the esti-
duration). Only one or two activities have reported mated time, the probability that at least one is late is over
early completion (out of a sample of 3,000), and the 88%! Even if each individual activity had a 90% proba-
rest later, including a few significantly later. This project bility of completion, the probability that at least one is
consists of about 30 large subprojects, some of which late is still 30%, or nearly one out of three times.
contain yet smaller subprojects. The PMBOK™ Guide states, “Schedule simulation
Undesired Effect 4: Project Delay Caused by should be used on any large or complex project since
Activity Path Merging. Most projects have multiple traditional mathematical analysis techniques such as
activity paths. All activity paths must merge into the crit- the Critical Path Method … do not account for path
ical path by the end of the project; if for no other reason convergence … and thus tend to underestimate project
than into a milestone that identifies project completion. durations.”

June 1999 Project Management Journal 43

 15 days late

Path A

On schedule 15 days late

Path B Merged Path

15 days ahead of schedule

Path C

Figure 3. Impact of Activity Path Merging

Undesired Effect 5: Multitasking. Multitasking is duration for this activity. Performance data supports
the performance of multiple project activities “at the this inflated activity duration. If this is a critical chain
same time.” Some people refer to it as the “fractional activity, the practice directly extends the duration of the
head count.” Humans are not too good at rubbing their project. Most companies admit to encouraging exten-
tummy and patting their head at the same time. People sive multitasking.
actually perform in a multiactivity mode by dividing Undesired Effect 6: Loss of Focus. Several aspects
time between the multiple activities. They might do this of current project planning make it difficult for the proj-
during the course of the day by working on one project ect manager to know where to focus to ensure project
in the morning and one in the afternoon. delivery. These include:
Most people think of multitasking as a good way to I Early start schedules, which allow all activity paths to
improve efficiency. It ensures everyone is busy all of the start at the same time. The instant jump to a high-
time. Often, I have to wait for inputs or for someone to activity level causes the project manager’s attention to
call back before I can get on with an activity. Multi- become diffused.
tasking makes good use of this time. I Changing the critical path during project performance.
Goldratt (1984) demonstrated how focus on local I Attempts to exclusively use earned value for project
efficiency could damage the overall performance of a control. Earned value does not discriminate between
system, an example of the phenomenon of subopt- activities that affect the critical path and those that do
mization. He used the example of robots, which oper- not. It weights activities by dollars, not schedule impor-
ated all of the time in order to show high efficiency. In tance. It gives no indication of the potential impact of
the case of production, this leads to producing excess paths parallel to the critical path.
inventory, and may “plug” the constraint with work not I Frequent use of earned value action thresholds that
necessary for current orders, increasing operating are too tight, causing too many control actions. Deming
expense and delivery times with no positive benefit to (1989) called this damaging behavior “tampering.”
the company as a whole. Tampering is attempting to fix variation that is within
Multitasking on project activities has a much worse the statistical bounds of common cause variation. Tam-
impact. Consider a person who has to do three one- pering always damages process performance.
week activities for three different projects (see Figure 4).
If the person were permitted to work exclusively on
each one, the first project would have its result in one Critical Chain Plan Process
week. The second project would have its output at the
end of the second week, and the third project at the end Critical Chain (Identify the Constraint). Goldratt
of the third week. However, if the activity performer (1997) extended the production application of TOC to
multitasks, spending for example one-third time each projects. He identified the constraint of a project as the
day on each project, none of the projects get their critical chain or “the sequence of dependent events that
output until the end of the third week. All three activi- prevents the project from completing in a shorter
ties have a three-week duration, potentially extending interval. Resource dependencies determine the critical
the overall duration of each project. chain as much as do activity dependencies.”
If multitasking is a normal way of business in a Defining the constraint of a project in terms of the
company, three weeks becomes the normal activity schedule derives from the impact that schedule has on

44 Project Management Journal June 1999

 One Unit of T One Unit of T One Unit of T

Task A Task B Task C

Task A
Task B
Task C

Figure 4. Multitasking Extends Activity Duration

project cost and project scope. The three conditions are The PMBOK™ Guide’s definition of critical path
dependent. As schedule increases with fixed deliverable states that the critical path may change during the per-
scope, cost usually increases. As scope increases with fixed formance of the project. This occurs when other paths
cost (or resources), schedule tends to increase. As scope experience delay, and redefine the longest “zero float”
increases with fixed schedule, cost tends to increase. path to complete the project. The critical chain does not
Critical path project planning contains an often change during project performance. This is partly a
hidden assumption: an acceptable way to account for matter of definition, but mostly a result of the overall
potential resource constraints on the project is to first critical chain plan construction procedure.
identify the critical path, and then perform resource lev- Project Activity Estimates (Exploit the Constraint).
eling. Network specialists know that there is no CCPM seeks to use the best estimate, or 50% probable,
optimum method to resource level. Network configura- individual activity time estimates. First assemble the plan
tions and some resource-leveling algorithms give very using the “low-risk” activity estimate duration provided
poor results. For most networks, applying resource-lev- by the project resources. Teach the estimators to under-
eling algorithms lengthens the overall schedule. For this stand variation and the critical chain method, including
reason, few projects use the resource-leveling tools. assurance that they will not be criticized or otherwise
Figure 5 illustrates a typical deterministic project impacted by either underrunning or overrunning esti-
schedule. The different shades represent unique mated duration. Next (the order is extremely important),
resources. The plan identifies the last activity as a critical solicit their estimated “average” times, assuming every-
path activity. Resource leveling has eliminated the rest thing went as they would hope it would, they have all
of the critical path, an often-mystifying result for the inputs when they start, and they are able to devote 100%
inexperienced project manager. Still, this is a frequent effort to the activity once started. Finally, build the critical
result of resource-leveling methods. chain plan using these reduced activity duration esti-
Since the resource constraint is often a significant mates, and collect the differences (D) between the low-
project constraint, the Theory of Constraints method of risk and average estimates to develop buffers.
project planning always considers it. Thus, the critical Subordinate Noncritical Chain Paths (Late Finish
chain includes the resource dependencies that define Plan). Project managers understand that early start sched-
the overall longest path (constraint) of the project. The ules can reduce project risk by getting things done early,
method resolves all resource constraints while deter- while late finish schedules:
mining the project critical chain. The project critical I Reduce the impact of changes on work already performed
chain may have gaps between activities. Figure 6 illus- I Delay the project cash outlay
trates the comparable critical chain plan. I Give the project a chance to focus by starting with
The improvements that result from CCPM do not fewer simultaneous activity chains, allowing the project
depend on having significant resource constraints or team and processes to come up to speed.
conflicts in the project. For a project without resource Much project management guidance recommends
constraints, the critical chain is the same initial activity that project managers use an early start schedule. Many
path as the critical path. The project plan differs signifi- scheduling computer programs use the early start
cantly, as described below. schedule as the “default.” Early start means permitting

June 1999 Project Management Journal 45

 Jun 1, 1997 Jun 15, 1997 Jun 29, 1997 Jul 13, 1997 Jul 27, 1997 Aug 10, 1997 Aug 2
ID Task Name 1 5 9 13 17 21 25 29 3 7 11 15 19 23 27 31 4 8 12 16 20 24 28

2 Plan
3 Permit
4 Site Prep
5 Hole
6 Landscape
7 Drive and Walks
10 Foundation
Carpenter
11 Frame
Carpenter
12 Roof
Carpenter
13 Sheath
14 Trim Carpenter

17 Plumbing
18 Electrical
19 Cabinets
20 Drywall Carpenter
21 Paint
22 Trim Carp

Figure 5. Example of Resource-Leveled Critical Path

all of the noncritical path activities to start earlier than is instead, they are use the information to compute the
necessary to meet the schedule date. People working on shared buffer.
those activities know that there is slack in their activity. Critical Chain Feeding Buffers (Subordinate
CCPM uses “late start” for all project activities. Merging Paths). CCPM protects the critical chain from
Feeding buffers provide an explicitly sized buffer to pro- potential delays by subordinating critical chain feeding
tect the overall project from late completion of the paths; placing an aggregated critical chain feeding buffer
feeding paths. This maximizes the advantage to the (CCFB) at the end of each path that feeds the critical
project, while ensuring project schedule protection. chain. This includes paths that merge with the critical
Project Buffer (Subordinate to the Constraint). chain at the end of the project. The feeding buffer pro-
CCPM protects the overall project delivery time with a vides a measurement and control mechanism to protect
project buffer at the end of the critical chain. This the critical chain, as described in Figure 7, which illus-
exploits the statistical law of aggregation by protecting trates how the buffers absorb the late paths. Figure 6
illustrates the location of CCFBs.
the project from common cause uncertainty of the indi-
This innovation immunizes the critical chain from
vidual activities in an activity path using buffers at the
potential delays in the feeding paths. It also provides a
end of the path. Buffers appear as activities in the
means to measure the feeding paths, while keeping
project plan, but have no work assigned to them.
focus on the critical chain.
Goldratt (1997) suggests a very simple method to
Resource Buffers Exploit the Critical Chain.
size buffers: use one half of the sum of the activity dura-
Resource buffers protect the critical chain from unavail-
tions in the chain of activities that precedes the buffer. ability of resources. They are a signal to the project
This is the critical chain for the project buffer. Others manager and resource managers to ensure resources are
use a method developed and applied by Lucent Tech- ready to work on critical chain activities as soon as the
nologies. Sum the “Ds” (from above, the difference activity input is ready. Resource buffers do not take time
between the low-risk and average estimates). Following in the critical chain and are sometimes called resource
the “law of aggregation,” use the square root of the sum “flags.” In risky situations, and in subcontracts, it may
of the squares (SSQ) to sum the Ds. be appropriate to include financial incentives in the
While the SSQ method has some theoretical attrac- resource buffers, such as paying for early delivery, penal-
tion, and generally leads to shorter buffers, the primary ties for late delivery, or paying for standby time. CCPM
reason to apply this method is to get the buy-in of the only applies resource buffers to critical chain activities,
project team. This way, management does not arbi- because the feeding buffers provide added protection to
trarily cut the “low-risk” estimates people provide; noncritical activity chains.

46 Project Management Journal June 1999

 June July
ID Task Name 6/1 6/8 6/15 6/22 6/29 7/6 7/13 7/20 7/27 8/3

2 Plan
3 Permit Resource
4 Site Preparation Flag
5 Hole
6 Landscape
7 CCFB-1
8 Drive and Walks
11 Foundation
12 Frame
13 Roof Critical Chain
Resource
14 Sheath Flag
15 Trim
18 Plumbing
19 Electrical
20 Cabinets
21 CCFB-2
22 Drywall
23 Paint
24 CCFB-3
25 Trim
27 Project Buffer

Figure 6. Example of Critical Chain Schedule

15 days late 20 day Buffer

Path A Buffer

On schedule

Path B Buffer

15 days early Start 5 days early

Path C Merged Path

Figure 7. Critical Chain Feeding Buffers Absorbing Delays

Critical Chain Human Performance long as the resources (a) start the activity as soon as
they had the input, (b) work 100% on the activity (no
Eliminate Date-Driven Behavior. Critical chain multitasking), and (c) pass on the activity output as
project plans only provide dates for the start of activity soon as it is completed. This is called “roadrunner”
chains and the end of the project buffer. This enables activity performance. They expect 50% of the activities
the project team to focus on completing the project as to overrun. (In practice, date-driven behavior persists;
soon as possible. For the rest of the project, the plan but with CCPM it causes delivery to the 50% probable
provides approximate start times and estimated activity time, thus automatically passing on positive variation
duration. This helps avoid date-driven behavior. relative to previous “low-risk” estimates.)
Critical chain project managers do not criticize per- Elevate Activity Performance by Eliminating
formers that overrun estimated activity durations, as Multitasking. CCPM seeks to eliminate this type of

June 1999 Project Management Journal 47

 0/3 1/3 2/3 3/3

No Action
Project
Buffer X

Plan
CCFB-1
X

Act
CCFB-2
X

Figure 8. Example of Using the Buffers

multitasking by eliciting 100% focus on the project appropriate time intervals for the project, usually
activity at hand by all resources supporting the project. weekly but at least monthly. For this tool to be fully
Thus, eliminating “fractional head counts” is a primary useful, the buffer monitoring time must be at least as
consideration in planning a critical chain project. frequent as one-third of the total buffer time. If the
Buffer Management Process. This process mea- buffers are negative (i.e., latest activity on the chain is
sures drive actions that move the project toward the early relative to schedule date) or less than one-third of
goal. Goldratt (1990) notes, “The first thing that must the total buffer late (e.g., less than 10 days if the total
buffer is 30 days), take no action. Buffer management
be clearly defined is the overall purpose of the organi-
provides a unique anticipatory project management
zation—or, as I prefer to call it, the organization’s goal.
tool with clear decision criteria.
The second thing is measurements. Not just any mea- Project managers update the buffers as often as
surements, but measurements that will enable us to they need by simply asking each of the performing
judge the impact of a local decision on the global goal.” activities how many days they estimate to the comple-
Goldratt (1990) also defines data as, “every string tion of their activity. They do this without pressure or
of characters that describes something, anything, about comment on the estimate. They expect these estimates
our reality.” He defines information as “The answer to to vary from day to day and some of the activities to
the question asked.” Goldratt suggests that the infor- exceed the original duration estimates. As long as the
mation system should incorporate the decision. resources are working on the activities with the CCPM
The improved measurement system for critical activity performance paradigm, managers evaluate them
chain project management follows the practice estab- positively, regardless of the actual duration.
lished by Goldratt for production operations. It uses Multiple Project CCPM Process. The impact of
buffers (that is, time) to measure activity chain perfor- multitasking on a single project is significant. In a mul-
mance. We set explicit action levels for decisions. The tiple project environment, it is a disaster. The impact
decision levels are in terms of the buffer size, measured gets worse and worse as managers push more and more
in days: projects into the project performance system. CCPM
1. Within the first third of the buffer: No action. project managers work to eliminate multitasking and
2. Penetrate the middle third of the buffer: Assess create a “pull” system for the multiproject environment.
the problem and plan for action. Figure 9 illustrates an example critical path multiproject
3. Penetrate the third third: Initiate action. scenario. The patterns on the bars represent resources.
These measures apply to both the project buffer Using conventional low-risk activity estimates and con-
(PB) and the CCFBs. Figure 8 shows an example of how sidering three project multitasking, we estimate each
buffer penetration provides the essential measurement activity duration as 90 days.
for CCPM project control. Identify the Multiproject Constraint. TOC
Project teams monitor the PB and each CCFB at the applies directly to plan and manage projects in the mul-

48 Project Management Journal June 1999

 Qtr 1, 1998 Qtr 2, 1998 Qtr 3, 1998 Qtr 4, 1998 Qtr 1, 1999 Qtr 2, 1999 Qtr 3, 1999 Qtr
ID Task Name Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct

1 Activity 1
2 Activity 2
3 Activity 3
4 Activity 4
5 Activity 5
7 Activity 1
8 Activity 2
9 Activity 3
10 Activity 4
11 Activity 5
13 Activity 1
14 Activity 2
15 Activity 3
16 Activity 4
17 Activity 5

Figure 9. Example of Critical Path Multiproject Schedule

tiproject environment. First, we have to identify the implementations of CCPM may not deploy the drum
company capacity constraint resource. This is most often buffer.
a certain type of person, but may be a physical or even a Do not attempt to schedule all resources across all
policy constraint. The company constraint resource projects. Companies demonstrate repeatedly that this is
becomes the “drum” for scheduling multiple projects. a losing proposition. It has never proven possible to get
This terminology comes from Goldratt’s production enough current information together and processed
methodology, where the drum sets the beat for the quick enough to exceed the ongoing variations in all
entire factory. Here, the drum sets the beat for all of the activities. CCPM allows for this variation with the
company projects. Think of the drummer on a galleon. resource flags and buffers within each project.
What happens if even one rower gets out of beat? Figure 10 shows the CCPM plan completing the
Figure 10 illustrates the CCPM method—activity three projects (including the project buffer) near the
times reduced to 15 days, eliminating the three times end of August 1998. It shows the first two projects com-
multitasking, and using 50% probable duration esti- pleting even earlier. Contrast this to the critical path
mates. The resource supplying activities 2 and 3 is the multiproject plans of Figure 9, all of which were sched-
“capacity constraint resource.” The plan exploits the uled to complete in May of 1999. Single-project experi-
resource by synchronizing the projects, using this ence suggests that the CCPM projects should be fin-
resource as the drum. The plan subordinates to this ished early. Experience with critical path projects
resource by adding capacity buffers between the proj- suggests that they will be late for even these extended
ects. The capacity buffers ensure that the capacity con- schedules.
straint resource (drum) is available for the subsequent Also note that synchronizing the projects this way
project. eliminates resource contention for all resources; not just
The project system becomes a “pull” system because the drum resource. This happens in the example
the drum schedule determines the sequencing of proj- because the projects are identical. While most multi-
ects. It pulls projects forward in time if the drum com- project environments do not have identical projects,
pletes project work early. It delays subsequent projects synchronizing projects to the drum usually eliminates
when the drum is late. For this reason, projects in a mul- some resource contention.
tiproject environment also require buffers to protect the Exploit Multiproject Resource Allocation.
drum … to ensure that they never starve the capacity Resource managers prioritize resource allocation across
constraint for work. The plan should schedule the proj- projects according to the importance of activities to the
ects to ensure that they are ready to use the drum projects. They give priority to (in order):
resource should it become available early. (The figure 1. Critical chain activities over noncritical chain
does not show the drum buffers for simplicity.) Early activities

June 1999 Project Management Journal 49

 Qtr 1, 1998 Qtr 2, 1998 Qtr 3, 1998 Qtr 4, 1998 Qtr 1, 1999 Qtr 2, 1999 Qtr 3, 1999 Qtr
ID Task Name Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct

19 Activity 1
20 Activity 2
21 Activity 3
22 Activity 4
23 CCFB
24 Activity 5
25 Project Buffer
27 Capacity Buffer
29 Activity 1
30 Activity 2
31 Activity 3
32 Activity 4
33 CCFB
34 Activity 5
35 Project Buffer
37 Capacity Buffer
39 Activity 1
40 Activity 2
41 Activity 3
42 Activity 4
43 CCFB
44 Activity 5
45 Project Buffer

Figure 10. The CCPM Multiproject Plan

2. Activities in project chains with greatest project stands buffer management, measured in days of buffer
buffer penetration penetration. Projects can collect and process buffer data
3. Activities in project chains with the highest feeding daily, if desired. Buffer penetration provides the deci-
buffer penetration. sion on when to plan and when to act. Few people
Penetration of feeding and project buffers resolves really understand the meaning of earned value mea-
remaining resource contentions.
surements, and how to use them for project manage-
ment. For example, Powers (1997) notes, “… the par-
Simplicity ticipants from leading Fortune 500 companies
CCPM planning and project management is simple attending the last benchmarking forum were queried
compared to other alternative techniques such as simu- regarding their use of earned value calculations. None
lation, quantitative risk assessment, PERT three time indicated that they were using the format described in
estimates, Monte Carlo methods, and earned value. The the PMBOK™ Guide.”
primary concepts are simple; comprising 50/50 esti- The five focusing steps are clear and concise. CCPM
mates, the critical chain, buffers, and buffer manage- provides a recipe for the straightforward application to
ment. CCPM requires neither statistical sophistication
real projects. CCPM does not require new computer
nor possession of actual distributions of activity perfor-
software, although such software is already available to
mance data. Such data usually does not exist for most
simplify the activities of resource leveling, finding the
projects, and even where it does exist, such as in the
construction industry, it has not solved the problem of critical chain, sizing and placing buffers, and per-
time overruns. forming buffer management (Creative Technology
Busy project managers do not have the time or Labs). Project teams can create CCPM plans in a short
inclination to assimilate obtuse data. They need real time (within a week for projects with the activity net-
information, gathered in real time. Everyone under- work and resource estimates available).

50 Project Management Journal June 1999

Conclusion Deming, W. Edwards. (1989). Out of the crisis. Cambridge,
Critical chain project management provides a substan- MA: MIT Press.
Goldratt, Eliyahu M. (1984). The goal. New York: North
tial step in continuous improvement to the Project
River Press, Croton-on-Hudson.
Management Body of Knowledge. It derives from Goldratt, Eliyahu M. (1990). The haystack syndrome. New
applying TOC and statistical theory to the project York: North River Press, Croton-on-Hudson.
system. It provides a conceptually simple and practical Goldratt, Eliyahu M. (1997). Critical chain. New York:
process to plan and manage projects. North River Press, Croton-on-Hudson.
Kiley, Martin D. (1996). 1997 National construction esti-
Focusing on the project constraint (the critical
mator. Craftsman Book Company.
chain) causes the success of CCPM. The critical chain Meredith, Jack R., & Mantel, Samuel J. (1995). Project
provides the focus for the whole project. The buffers management: A managerial approach. New York: John Wiley &
provide focus and clear decision criteria for the project Sons Inc.
manager. Moore, David S., & McCabe, George P. (1993). Introduction
to the practice of statistics. New York: W.H. Freeman & Co.
The essential changes introduced by critical chain
Powers, Ray. (1997). Response from standards committee.
project management (relative to the current practices) are: Project Management Journal, 28 (2), 53.
I Develop the critical chain plan Project Management Institute Standards Committee. (1996).
1. Develop the critical chain, using both activity A guide to the project management body of knowledge (PMBOK™
logic and resource constraints. guide). Upper Darby, PA: Project Management Institute.
Shewhart, Walter A. (1986). Statistical method from the
2. Reduce activity-estimated times to 50% proba-
viewpoint of quality control. New York: Dover Publications
bility estimates to account for aggregation of the (Originally published in 1939).
activity contingency times.
3. Add a project buffer to protect completion of the
critical chain.
4. Add critical chain feeding buffers to immunize the
critical chain from delays in the feeding chains and
merging effects.
5. Add resource buffers to ensure critical chain
resources are available when needed.
I Multiple project plan
1. Plan each project plan as in above steps 1 and 2.
2. Identify the constraint (drum) resource.
3. Create the drum schedule.
4. Sequence projects to the drum.
5. Insert project, feeding, and resource buffers into
each project.
6. Insert drum and capacity buffers into each project.
I Measure and control to the plan
1. Use buffer management as the primary tool to
measure and control the project.
2. Use buffer reports to assign resources.
I Lead behavior to achieve earliest project completion
Use behaviors conducive to the global project optima,
such as roadrunner activity performance, passing on pos-
itive variation, and critical chain-based resource allocation
to satisfy the company project needs.
All projects that have diligently applied CCPM have
completed the project substantially under the original
time estimate, fulfilled the original scope, and came in
near or under the estimated budget. Project durations
normally reduce by at least 50% in the first pass, and
several companies have taken the early successes to
cause further substantial duration reductions.

References
Creative Technology Labs LLC, 37 Grieb Trail. Walling-
ford, CT.

June 1999 Project Management Journal 51