Eliyahu Goldratt's book, Critical Chain [The North River Press, 1997], is widely regarded as the most profound recent publication in project management. Written as a novel, this book applies the Theory of Constraints (TOC) to project management. The obvious constraint in projects, of course, is the critical path. Critical chain project management extends critical path analysis to the constraining resource. Many managers have seized upon the critical chain approach as an improved way of planning and managing projects.
Critical chain addresses several key problems in project scheduling. However, the method ignores much of what we've learned in decision analysis about modeling and understanding risk. Critical chain is fundamentally a deterministic methodology. This article compares and contrasts critical chain with a traditional decision analysis approach, then describes ways to blend the best of the two styles.
This article and new series in 2000 continues the “Decision Analysis in Projects” series that appeared in PM Network during 1993–1995, and which was embodied in Decision Analysis in Projects [Project Management Institute, 1997].
Critical Chain. Eliyahu Goldratt is considered by many to be the father of the theory of constraints, which is widely recognized in manufacturing. His first best-selling book, The Goal [The North River Press, 1992] is about de-bottlenecking manufacturing operations. Simply put, greater throughput can be achieved by attacking the bottleneck constraints. Goldratt's process, whether talking about inventory or time, is essentially the same:
■ Identify the constraint.
■ Exploit the constraint.
■ Subordinate other activities to the constraint. Don't start work prematurely.
■ Elevate the constraint. Seek ways to raise the productivity or availability of the constraining resource.
John Schuyler, PMP, of Decision Precision in Aurora, Colo., provides training and assistance in economic decision analysis and in project risk management. Questions about this article should be directed to [email protected]. Comments on this series should be directed to [email protected].
Exhibit 1. This chart highlights features of critical chain and Monte Carlo simulation and compares the two approaches.
Managers want to ensure that the constraining resource is fully used. They avoid starting work prematurely as this builds excess work-in-process inventory.
In Critical Chain, Goldratt applies his philosophy to project management. The analogy between production management and project management is effective. The obvious bottleneck is the sequence of activities along the critical path. The analog of excess work-in-progress inventory is off-critical-path activities that are started prematurely. Manufacturing's throughput analog is project completion time.
Critical Chain has been widely reviewed and discussed, and is an enjoyable way to challenge one's beliefs about project management and to gain new insights. Other references that I've found most helpful include articles by Larry Leach [“Critical Chain Project Management Improves Project Performance,” Project Management Journal, June 1999], Jeffrey Elton and Justin Roe [“Bringing Discipline to Project Management,” Harvard Business Review, Mar.–Apr. 1998], and Robert Newbold's Project Management in the Fast Lane: Applying the Theory of Constraints [St. Lucie Press, 1998].
Critical chain project management's schedule model is deterministic, with buffers to allow for uncertainty. Deterministic means that the inputs to the model are all singly determined values. This model specifies that the user:
Exhibit 2. Distribution forecast for when all of activity AE (assemble equipment) predecessors will be complete. The distribution in this case was obtained from a submodel. The project model calculates a criticality index of 0.35 for this activity.
Remove Padding From Activity Estimates. Otherwise, usually, the slack will be wasted. Goldratt suggests estimating activity completion times at 50 percent confidence points (medians). Put a main buffer at the end of the project to protect the customer's completion schedule. Also place buffers at the end of activity chains that feed into the critical path. Schedule high-risk activities early so that problems can be detected and addressed early.
Determine the Critical Path and Resource Constraints. Adjust the project plan(s) as necessary to exploit the constraining resource(s). The driving objective is minimizing project completion time.
Avoid Wasting Slack Times. Part of this is accomplished by using median rather than conservative time estimates. Encourage early activity completions. Instill a culture in which an activity team is ready to begin working on its critical activity and will devote 100 percent of its effort to this activity for the duration.
Work to Plan, and Avoid Tampering. Monitoring and communicating buffer statuses are key in managing the ongoing project.
“It's always a people problem” is an old adage in management consulting. Critical chain deals with project management's challenge to handle uncertainty effectively. The critical chain approach reduces the effects of two behavioral problems: biasing estimations and wasting slack times. However, we can do much more to manage uncertainty.
Decision Analysis With Monte Carlo Simulation. Decision analysis is the discipline for making decisions and evaluations under uncertainty. It is characterized by analyses having three key features:
■ Judgments about uncertainty, and often analysis results, are expressed as probability distributions.
■ The quality of an outcome is determined by a numeric value function that measures goodness or progress toward the organization's objective (most popularly net persent value).
■ The expected value (EV) calculation compresses the distribution of possible outcome values into a single number. EV is the probability-weighted average value. It is identical to the mean statistic.
The optimal decision policy is to choose the alternative having the best EV.
The project model is used to forecast the behavior of the project as a system. Monte Carlo simulation, a straightforward process that allows distributions to represent activity time estimates and other uncertain inputs, is perhaps the easiest and most useful way to add details about uncertainties. Simulation solves the project schedule, providing distributions and EVs for items such as time to complete and cost. There are several commercially available programs that provide Monte Carlo capabilities for models in Microsoft Excel, Microsoft Project, and various other project planning tools.
The key analytic benefit of simulation is a more accurate estimate. Project managers have recognized that uncertainty in merging activity paths produces a longer schedule. This is because it is the latest predecessor completion time that drives the start of the next activity. With stochastic (probabilistic) project models, this is often called merge bias (more generally, I apply the term stochastic variance to this correction).
Contrary to popular belief, summing up activity completion times does not produce a normal distribution. Usually, in real projects there are many correlations. Often, when things start to go wrong, the problems compound, affecting many activities.
Reader Service Number 059
With the simulation model the project manager can forecast such elements as:
■ The probability that an activity lies on the critical path (looking back at project completion); this is called the criticality index
■ The distribution for time to complete the project or any milestone sequence of activities
■ The distribution of project cost or, better, project value to the customer; cost/benefit analyses can be made for candidate risk mitigation actions and for possible activity-crashing efforts.
Traditionally, project scheduling programs start activities as soon as possible. The intent is to minimize possible delays from time overruns for most noncritical activities. Critical Chain explains why we should avoid default early starts.
Blending Critical Chain and Decision Analysis. Exhibit 1 highlights the features of critical chain and Monte Carlo simulation, and compares the two approaches. I believe the deterministic calculations are a deficiency in critical chain management. On the other hand, the critical chain approach remedies some problems with a traditional stochastic planning model.
Here is an approach to realizing benefits from blending critical chain and decision analysis styles to project management. Begin by developing the work breakdown structure and activity network diagram in the usual way. Then:
Obtain Quality Judgments About Activity Completion Times and Resource Requirements. Usually this done by interviewing the best available experts. These judgments should be expressed as probability distributions. Breaking apart important contingencies (probability and impact distribution if the contingency occurs) is a good way to obtain a better representation and a more realistic estimate. A baseline completion time (or cost) for managing an activity would be the EV time to complete, assuming no major contingencies occur.
Learning is essential to improving future judgments. A graph of actuals vs. estimates is an excellent feedback tool. Post-evaluations should be used in a nonpunitive way. A culture of trust and objectivity is essential for obtaining quality estimates and for other human facets of effective project management.
The activity manager maintains the model inputs (see below) for time to complete the activity. This function, and the cost of the activity, can incorporate dependencies for delayed start and other dynamic conditions.
Exhibit 3. Forecast of time to complete activity AE, provided by the team leader. In this activity, either or both of two significant contingencies could occur, lengthening time to complete AE. If neither of the two significant risk events occur, then the more narrow distribution is the forecast used.
Develop a Stochastic Project Model. This should be maintained perpetually throughout the project life cycle. As available resources permit, shorten the project plan to conduct more activities in parallel.
Reader Service Number 080
Exhibit 4. Total project cost as a function of time to complete activity AE. This function and the criticality index are useful in evaluating alternatives to delaying, slowing, or accelerating work on this activity.
Start times for low-criticality index activities should be scheduled to optimize project value. Simulation models will show that many noncritical activity starts should be delayed to reduce multitasking, premature investment, and rework due to changes.
The stochastic project model is the core of effective project risk management. Cost/benefit/risk analysis can be done in most cases with the activity information described below and in extreme cases by exploring changes to the top-level project model. Optimization techniques can plan best start times for each activity (topic of the next article in this series). In complex situations of shared resources, we need to expand the scope of the stochastic model to incorporate multiple projects.
Provide Activity Leaders With Convenient Access to the Project Model. The project manager should be especially involved in monitoring ongoing activities (dedicated efforts to the extent possible; reporting early completion) and in alerting workgroups to be prepared as work on critical (or near-critical) activities approaches.
Activity managers need to be informed about such elements as:
■ The forecast, as a probability distribution, for when all predecessors to the activity will be complete (see Exhibit 2)
■ Probability that the activity will be on the critical path, the criticality index
■ Time to complete the activity (see Exhibit 3), as judged by the activity leader or best available expert; in some cases it is useful to build a submodel of the activity that details the major contingencies
■ A graph of project value vs. the particular activity completion time (see Exhibit 4). Stay Flexible. The project is a dynamic system, and the best project and activity managers will be alert and responsive to changes as they occur. Regularly examine where resources might be better allocated.
CAPABLE LEADERSHIP OF a motivated, welltrained team is of paramount importance to project success. However, even skilled people are poor at complex planning processes without the assistance of quantitative tools. The methods described here can be very helpful in bolstering human judgment.
This is an exciting time in project management. The project management body of knowledge has changed dramatically during the past 20 years. Low-cost software and computing power now enable modeling projects as never before. We have the means at hand to dramatically improve productivity in project execution. ■
Reader Service Number 027